diff --git a/configs/config_1.14G_dp128_tp4_pp1_acc1_mbs1_seq32768_zero1_tpmodeRED_vocab32k.yaml b/configs/config_1.14G_dp128_tp4_pp1_acc1_mbs1_seq32768_zero1_tpmodeRED_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..a36f1834e5e6a7998922f49a50779f4c3d12b477 --- /dev/null +++ b/configs/config_1.14G_dp128_tp4_pp1_acc1_mbs1_seq32768_zero1_tpmodeRED_vocab32k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.14G_dp128_tp4_pp1_acc1_mbs1_seq32768_zero1_tpmodeRED_vocab32k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 32768 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 128 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 4 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 1 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 1 + sequence_length: 32768 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp1_tp8_pp1_acc1_mbs64_seq2048_zero0_tpmodeALL_vocab32k_prof.yaml b/configs/config_1.14G_dp1_tp8_pp1_acc1_mbs64_seq2048_zero0_tpmodeALL_vocab32k_prof.yaml new file mode 100644 index 0000000000000000000000000000000000000000..5b4faa7449a9ed812d944b7152d4e252a8fb0d4c --- /dev/null +++ b/configs/config_1.14G_dp1_tp8_pp1_acc1_mbs64_seq2048_zero0_tpmodeALL_vocab32k_prof.yaml @@ -0,0 +1,92 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.14G_dp1_tp8_pp1_acc1_mbs64_seq2048_zero0_tpmodeALL_vocab32k_prof + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 2048 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 1 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 8 + tp_linear_async_communication: false + tp_mode: ALL_REDUCE +profiler: + profiler_export_path: ./tb_logs +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 1 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 64 + sequence_length: 2048 + train_steps: 10 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp2_tp128_pp1_acc16_mbs16_seq8192_zero1_tpmodeALL_vocab32k.yaml b/configs/config_1.14G_dp2_tp128_pp1_acc16_mbs16_seq8192_zero1_tpmodeALL_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..4404c044f29c17024d6e0c6f7360feb0d854714d --- /dev/null +++ b/configs/config_1.14G_dp2_tp128_pp1_acc16_mbs16_seq8192_zero1_tpmodeALL_vocab32k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.14G_dp2_tp128_pp1_acc16_mbs16_seq8192_zero1_tpmodeALL_vocab32k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 8192 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 2 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 128 + tp_linear_async_communication: false + tp_mode: ALL_REDUCE +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 16 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 16 + sequence_length: 8192 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp2_tp128_pp1_acc32_mbs32_seq2048_zero1_tpmodeALL_vocab32k.yaml b/configs/config_1.14G_dp2_tp128_pp1_acc32_mbs32_seq2048_zero1_tpmodeALL_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e5784ad860e2cf37944a4f739b3662626f5298f2 --- /dev/null +++ b/configs/config_1.14G_dp2_tp128_pp1_acc32_mbs32_seq2048_zero1_tpmodeALL_vocab32k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.14G_dp2_tp128_pp1_acc32_mbs32_seq2048_zero1_tpmodeALL_vocab32k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 2048 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 2 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 128 + tp_linear_async_communication: false + tp_mode: ALL_REDUCE +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 32 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 32 + sequence_length: 2048 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp2_tp256_pp1_acc4_mbs64_seq2048_zero1_tpmodeRED_vocab32k.yaml b/configs/config_1.14G_dp2_tp256_pp1_acc4_mbs64_seq2048_zero1_tpmodeRED_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..310a6fdf24e2e6b8d87d959631bb808360ac6b1c --- /dev/null +++ b/configs/config_1.14G_dp2_tp256_pp1_acc4_mbs64_seq2048_zero1_tpmodeRED_vocab32k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.14G_dp2_tp256_pp1_acc4_mbs64_seq2048_zero1_tpmodeRED_vocab32k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 2048 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 2 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 256 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 4 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 64 + sequence_length: 2048 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp2_tp256_pp1_acc64_mbs16_seq2048_zero1_tpmodeALL_vocab32k.yaml b/configs/config_1.14G_dp2_tp256_pp1_acc64_mbs16_seq2048_zero1_tpmodeALL_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..3b21c96b048898a44a53b796f3e295fff9eeb6b1 --- /dev/null +++ b/configs/config_1.14G_dp2_tp256_pp1_acc64_mbs16_seq2048_zero1_tpmodeALL_vocab32k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.14G_dp2_tp256_pp1_acc64_mbs16_seq2048_zero1_tpmodeALL_vocab32k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 2048 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 2 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 256 + tp_linear_async_communication: false + tp_mode: ALL_REDUCE +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 64 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 16 + sequence_length: 2048 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp2_tp4_pp1_acc1_mbs64_seq32768_zero1_tpmodeRED_vocab32k.yaml b/configs/config_1.14G_dp2_tp4_pp1_acc1_mbs64_seq32768_zero1_tpmodeRED_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..ebfddcbaaec29434c3c46dbea2dd5aecfc6f4f5e --- /dev/null +++ b/configs/config_1.14G_dp2_tp4_pp1_acc1_mbs64_seq32768_zero1_tpmodeRED_vocab32k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.14G_dp2_tp4_pp1_acc1_mbs64_seq32768_zero1_tpmodeRED_vocab32k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 32768 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 2 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 4 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 1 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 64 + sequence_length: 32768 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp2_tp8_pp1_acc16_mbs4_seq8192_zero1_tpmodeRED_vocab32k.yaml b/configs/config_1.14G_dp2_tp8_pp1_acc16_mbs4_seq8192_zero1_tpmodeRED_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..83e24502f4a7d54483055e4c8c2901ced36fcf9b --- /dev/null +++ b/configs/config_1.14G_dp2_tp8_pp1_acc16_mbs4_seq8192_zero1_tpmodeRED_vocab32k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.14G_dp2_tp8_pp1_acc16_mbs4_seq8192_zero1_tpmodeRED_vocab32k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 8192 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 2 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 8 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 16 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 4 + sequence_length: 8192 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp32_tp1_pp1_acc16_mbs1_seq8192_zero1_tpmodeRED_vocab32k.yaml b/configs/config_1.14G_dp32_tp1_pp1_acc16_mbs1_seq8192_zero1_tpmodeRED_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..85656a5c7753e443b4a9783d19318c7ad97d0028 --- /dev/null +++ b/configs/config_1.14G_dp32_tp1_pp1_acc16_mbs1_seq8192_zero1_tpmodeRED_vocab32k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.14G_dp32_tp1_pp1_acc16_mbs1_seq8192_zero1_tpmodeRED_vocab32k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 8192 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 32 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 1 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 16 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 1 + sequence_length: 8192 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp4_tp2_pp1_acc1_mbs8_seq32768_zero1_tpmodeRED_vocab32k.yaml b/configs/config_1.14G_dp4_tp2_pp1_acc1_mbs8_seq32768_zero1_tpmodeRED_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..965c590b336022ec28d690649e95f66426dbb9cb --- /dev/null +++ b/configs/config_1.14G_dp4_tp2_pp1_acc1_mbs8_seq32768_zero1_tpmodeRED_vocab32k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.14G_dp4_tp2_pp1_acc1_mbs8_seq32768_zero1_tpmodeRED_vocab32k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 32768 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 4 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 2 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 1 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 8 + sequence_length: 32768 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp4_tp32_pp1_acc64_mbs8_seq2048_zero1_tpmodeALL_vocab32k.yaml b/configs/config_1.14G_dp4_tp32_pp1_acc64_mbs8_seq2048_zero1_tpmodeALL_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..687a8ea2bfe56e5ca3967ddea2efd114a8b252ca --- /dev/null +++ b/configs/config_1.14G_dp4_tp32_pp1_acc64_mbs8_seq2048_zero1_tpmodeALL_vocab32k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.14G_dp4_tp32_pp1_acc64_mbs8_seq2048_zero1_tpmodeALL_vocab32k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 2048 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 4 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 32 + tp_linear_async_communication: false + tp_mode: ALL_REDUCE +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 64 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 8 + sequence_length: 2048 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp4_tp4_pp1_acc64_mbs8_seq2048_zero1_tpmodeALL_vocab32k.yaml b/configs/config_1.14G_dp4_tp4_pp1_acc64_mbs8_seq2048_zero1_tpmodeALL_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..44c2b8625b294631ca81df845aba8ba3a22314d5 --- /dev/null +++ b/configs/config_1.14G_dp4_tp4_pp1_acc64_mbs8_seq2048_zero1_tpmodeALL_vocab32k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.14G_dp4_tp4_pp1_acc64_mbs8_seq2048_zero1_tpmodeALL_vocab32k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 2048 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 4 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 4 + tp_linear_async_communication: false + tp_mode: ALL_REDUCE +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 64 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 8 + sequence_length: 2048 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp4_tp8_pp1_acc1_mbs128_seq2048_zero1_tpmodeRED_vocab32k.yaml b/configs/config_1.14G_dp4_tp8_pp1_acc1_mbs128_seq2048_zero1_tpmodeRED_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..7d68cb3c230b392c98363f78864238995b552b58 --- /dev/null +++ b/configs/config_1.14G_dp4_tp8_pp1_acc1_mbs128_seq2048_zero1_tpmodeRED_vocab32k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.14G_dp4_tp8_pp1_acc1_mbs128_seq2048_zero1_tpmodeRED_vocab32k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 2048 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 4 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 8 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 1 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 128 + sequence_length: 2048 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp64_tp2_pp1_acc1_mbs8_seq2048_zero1_tpmodeRED_vocab32k.yaml b/configs/config_1.14G_dp64_tp2_pp1_acc1_mbs8_seq2048_zero1_tpmodeRED_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..7dd8eb3840f228ab0e5341d6859968f1b328149e --- /dev/null +++ b/configs/config_1.14G_dp64_tp2_pp1_acc1_mbs8_seq2048_zero1_tpmodeRED_vocab32k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.14G_dp64_tp2_pp1_acc1_mbs8_seq2048_zero1_tpmodeRED_vocab32k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 2048 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 64 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 2 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 1 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 8 + sequence_length: 2048 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp64_tp2_pp1_acc2_mbs1_seq8192_zero1_tpmodeALL_vocab32k.yaml b/configs/config_1.14G_dp64_tp2_pp1_acc2_mbs1_seq8192_zero1_tpmodeALL_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..80102a7dafe753497a9a8ba4c702ff934e095d7b --- /dev/null +++ b/configs/config_1.14G_dp64_tp2_pp1_acc2_mbs1_seq8192_zero1_tpmodeALL_vocab32k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.14G_dp64_tp2_pp1_acc2_mbs1_seq8192_zero1_tpmodeALL_vocab32k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 8192 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 64 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 2 + tp_linear_async_communication: false + tp_mode: ALL_REDUCE +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 2 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 1 + sequence_length: 8192 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp8_tp1_pp1_acc32_mbs2_seq2048_zero1_tpmodeRED_vocab32k.yaml b/configs/config_1.14G_dp8_tp1_pp1_acc32_mbs2_seq2048_zero1_tpmodeRED_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..4081b16d8ed6f85abedbeea924298d8277ede40d --- /dev/null +++ b/configs/config_1.14G_dp8_tp1_pp1_acc32_mbs2_seq2048_zero1_tpmodeRED_vocab32k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.14G_dp8_tp1_pp1_acc32_mbs2_seq2048_zero1_tpmodeRED_vocab32k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 2048 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 8 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 1 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 32 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 2 + sequence_length: 2048 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp8_tp64_pp1_acc4_mbs1_seq32768_zero1_tpmodeALL_vocab32k.yaml b/configs/config_1.14G_dp8_tp64_pp1_acc4_mbs1_seq32768_zero1_tpmodeALL_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..11b3d6210b7c7b34aa73747ae111310c525c53a9 --- /dev/null +++ b/configs/config_1.14G_dp8_tp64_pp1_acc4_mbs1_seq32768_zero1_tpmodeALL_vocab32k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.14G_dp8_tp64_pp1_acc4_mbs1_seq32768_zero1_tpmodeALL_vocab32k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 32768 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 8 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 64 + tp_linear_async_communication: false + tp_mode: ALL_REDUCE +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 4 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 1 + sequence_length: 32768 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.14G_dp8_tp8_pp1_acc128_mbs2_seq2048_zero1_tpmodeRED_vocab32k.yaml b/configs/config_1.14G_dp8_tp8_pp1_acc128_mbs2_seq2048_zero1_tpmodeRED_vocab32k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..531262dfb0686f8c2de0dd7bc4743c7cfc2a6943 --- /dev/null +++ b/configs/config_1.14G_dp8_tp8_pp1_acc128_mbs2_seq2048_zero1_tpmodeRED_vocab32k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.14G_dp8_tp8_pp1_acc128_mbs2_seq2048_zero1_tpmodeRED_vocab32k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 2048 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 8 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 8 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 128 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 2 + sequence_length: 2048 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp128_tp2_pp1_acc8_mbs2_seq2048_zero1_tpmodeALL_vocab131k.yaml b/configs/config_1.34G_dp128_tp2_pp1_acc8_mbs2_seq2048_zero1_tpmodeALL_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..847d84e74a8fd7fc0d7e41099dfdc78acc649af5 --- /dev/null +++ b/configs/config_1.34G_dp128_tp2_pp1_acc8_mbs2_seq2048_zero1_tpmodeALL_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp128_tp2_pp1_acc8_mbs2_seq2048_zero1_tpmodeALL_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 2048 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 128 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 2 + tp_linear_async_communication: false + tp_mode: ALL_REDUCE +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 8 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 2 + sequence_length: 2048 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp16_tp16_pp1_acc1_mbs2_seq32768_zero1_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp16_tp16_pp1_acc1_mbs2_seq32768_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..4d966e800b0b166705ea93a47e6b9dcca15786d5 --- /dev/null +++ b/configs/config_1.34G_dp16_tp16_pp1_acc1_mbs2_seq32768_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp16_tp16_pp1_acc1_mbs2_seq32768_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 32768 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 16 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 16 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 1 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 2 + sequence_length: 32768 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp16_tp32_pp1_acc32_mbs1_seq2048_zero1_tpmodeALL_vocab131k.yaml b/configs/config_1.34G_dp16_tp32_pp1_acc32_mbs1_seq2048_zero1_tpmodeALL_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..b5030dc17aceee43d780e689b2dad2ce69613d80 --- /dev/null +++ b/configs/config_1.34G_dp16_tp32_pp1_acc32_mbs1_seq2048_zero1_tpmodeALL_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp16_tp32_pp1_acc32_mbs1_seq2048_zero1_tpmodeALL_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 2048 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 16 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 32 + tp_linear_async_communication: false + tp_mode: ALL_REDUCE +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 32 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 1 + sequence_length: 2048 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp16_tp4_pp1_acc4_mbs32_seq2048_zero1_tpmodeALL_vocab131k.yaml b/configs/config_1.34G_dp16_tp4_pp1_acc4_mbs32_seq2048_zero1_tpmodeALL_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..3112d35868c2c2126618fa8e9ea7ea206c6403c7 --- /dev/null +++ b/configs/config_1.34G_dp16_tp4_pp1_acc4_mbs32_seq2048_zero1_tpmodeALL_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp16_tp4_pp1_acc4_mbs32_seq2048_zero1_tpmodeALL_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 2048 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 16 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 4 + tp_linear_async_communication: false + tp_mode: ALL_REDUCE +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 4 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 32 + sequence_length: 2048 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp1_tp16_pp2_acc1_mbs256_seq4096_zero0_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp1_tp16_pp2_acc1_mbs256_seq4096_zero0_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..b937ad1dafe1dc7a18c23b930d4e10f663362c69 --- /dev/null +++ b/configs/config_1.34G_dp1_tp16_pp2_acc1_mbs256_seq4096_zero0_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp1_tp16_pp2_acc1_mbs256_seq4096_zero0_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 1 + expert_parallel_size: 1 + pp: 2 + pp_engine: 1f1b + tp: 16 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 1 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 256 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp1_tp1_pp8_acc7_mbs2_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp1_tp1_pp8_acc7_mbs2_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..5d02647e7fdf730f330e2188a9f19900a9184945 --- /dev/null +++ b/configs/config_1.34G_dp1_tp1_pp8_acc7_mbs2_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2_pp.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp1_tp1_pp8_acc7_mbs2_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 1 + expert_parallel_size: 1 + pp: 8 + pp_engine: 1f1b + tp: 1 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 7 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 2 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp256_tp2_pp1_acc2_mbs1_seq8192_zero1_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp256_tp2_pp1_acc2_mbs1_seq8192_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..979afc95230969418bc1eb2dfe3f546219f7826a --- /dev/null +++ b/configs/config_1.34G_dp256_tp2_pp1_acc2_mbs1_seq8192_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp256_tp2_pp1_acc2_mbs1_seq8192_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 8192 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 256 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 2 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 2 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 1 + sequence_length: 8192 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp2_tp256_pp1_acc2_mbs128_seq8192_zero1_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp2_tp256_pp1_acc2_mbs128_seq8192_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..144cbffa3ea41f583c6d5ead421d6d93caa58296 --- /dev/null +++ b/configs/config_1.34G_dp2_tp256_pp1_acc2_mbs128_seq8192_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp2_tp256_pp1_acc2_mbs128_seq8192_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 8192 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 2 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 256 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 2 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 128 + sequence_length: 8192 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp2_tp256_pp1_acc64_mbs4_seq8192_zero1_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp2_tp256_pp1_acc64_mbs4_seq8192_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..5f65d2e4fc84953d411ff92973e6fbc72d6d3115 --- /dev/null +++ b/configs/config_1.34G_dp2_tp256_pp1_acc64_mbs4_seq8192_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp2_tp256_pp1_acc64_mbs4_seq8192_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 8192 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 2 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 256 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 64 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 4 + sequence_length: 8192 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp2_tp4_pp1_acc128_mbs8_seq2048_zero1_tpmodeALL_vocab131k.yaml b/configs/config_1.34G_dp2_tp4_pp1_acc128_mbs8_seq2048_zero1_tpmodeALL_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..ebfdb38d7f22dc058208b82f5feee65db7dad161 --- /dev/null +++ b/configs/config_1.34G_dp2_tp4_pp1_acc128_mbs8_seq2048_zero1_tpmodeALL_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp2_tp4_pp1_acc128_mbs8_seq2048_zero1_tpmodeALL_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 2048 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 2 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 4 + tp_linear_async_communication: false + tp_mode: ALL_REDUCE +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 128 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 8 + sequence_length: 2048 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp2_tp4_pp1_acc4_mbs64_seq8192_zero1_tpmodeALL_vocab131k.yaml b/configs/config_1.34G_dp2_tp4_pp1_acc4_mbs64_seq8192_zero1_tpmodeALL_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..5118b0faba81b83a4c6a87d4a10c5ae7243c204f --- /dev/null +++ b/configs/config_1.34G_dp2_tp4_pp1_acc4_mbs64_seq8192_zero1_tpmodeALL_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp2_tp4_pp1_acc4_mbs64_seq8192_zero1_tpmodeALL_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 8192 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 2 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 4 + tp_linear_async_communication: false + tp_mode: ALL_REDUCE +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 4 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 64 + sequence_length: 8192 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp2_tp8_pp1_acc1_mbs16_seq32768_zero1_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp2_tp8_pp1_acc1_mbs16_seq32768_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..b32dfe6711f45b460eec95573e8faacc33d07c05 --- /dev/null +++ b/configs/config_1.34G_dp2_tp8_pp1_acc1_mbs16_seq32768_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp2_tp8_pp1_acc1_mbs16_seq32768_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 32768 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 2 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 8 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 1 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 16 + sequence_length: 32768 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp32_tp16_pp1_acc4_mbs2_seq4096_zero0_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp32_tp16_pp1_acc4_mbs2_seq4096_zero0_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..c50a8c64381b07e781d2a5dcbc0932ef2a53a651 --- /dev/null +++ b/configs/config_1.34G_dp32_tp16_pp1_acc4_mbs2_seq4096_zero0_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp32_tp16_pp1_acc4_mbs2_seq4096_zero0_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 32 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 16 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 4 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 2 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp32_tp2_pp1_acc1_mbs4_seq32768_zero1_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp32_tp2_pp1_acc1_mbs4_seq32768_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..0143f9b355c6e3cb13b49e64c138d7bce2ae902f --- /dev/null +++ b/configs/config_1.34G_dp32_tp2_pp1_acc1_mbs4_seq32768_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp32_tp2_pp1_acc1_mbs4_seq32768_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 32768 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 32 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 2 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 1 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 4 + sequence_length: 32768 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp4_tp16_pp2_acc2_mbs32_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp4_tp16_pp2_acc2_mbs32_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..65f94de0a5dad974b39d13458833dd34ddf3a08a --- /dev/null +++ b/configs/config_1.34G_dp4_tp16_pp2_acc2_mbs32_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp4_tp16_pp2_acc2_mbs32_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 4 + expert_parallel_size: 1 + pp: 2 + pp_engine: 1f1b + tp: 16 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 2 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 32 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp4_tp16_pp4_acc64_mbs1_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp4_tp16_pp4_acc64_mbs1_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..0135c22e0245061ce6cdcbb64b38b1178572fb5e --- /dev/null +++ b/configs/config_1.34G_dp4_tp16_pp4_acc64_mbs1_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp4_tp16_pp4_acc64_mbs1_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 4 + expert_parallel_size: 1 + pp: 4 + pp_engine: 1f1b + tp: 16 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 64 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 1 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp4_tp2_pp1_acc1_mbs128_seq2048_zero1_tpmodeALL_vocab131k.yaml b/configs/config_1.34G_dp4_tp2_pp1_acc1_mbs128_seq2048_zero1_tpmodeALL_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..52961014316f96cdbd97d604998480381307fa6a --- /dev/null +++ b/configs/config_1.34G_dp4_tp2_pp1_acc1_mbs128_seq2048_zero1_tpmodeALL_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp4_tp2_pp1_acc1_mbs128_seq2048_zero1_tpmodeALL_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 2048 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 4 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 2 + tp_linear_async_communication: false + tp_mode: ALL_REDUCE +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 1 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 128 + sequence_length: 2048 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp4_tp32_pp1_acc64_mbs2_seq2048_zero1_tpmodeALL_vocab131k.yaml b/configs/config_1.34G_dp4_tp32_pp1_acc64_mbs2_seq2048_zero1_tpmodeALL_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..0c301c9e518cd22929f6e16f46274591802578af --- /dev/null +++ b/configs/config_1.34G_dp4_tp32_pp1_acc64_mbs2_seq2048_zero1_tpmodeALL_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp4_tp32_pp1_acc64_mbs2_seq2048_zero1_tpmodeALL_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 2048 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 4 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 32 + tp_linear_async_communication: false + tp_mode: ALL_REDUCE +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 64 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 2 + sequence_length: 2048 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp4_tp8_pp1_acc32_mbs16_seq2048_zero1_tpmodeALL_vocab131k.yaml b/configs/config_1.34G_dp4_tp8_pp1_acc32_mbs16_seq2048_zero1_tpmodeALL_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..2d6b172aaeaa4503ba7d2f3c7e48b3ca16bde2f1 --- /dev/null +++ b/configs/config_1.34G_dp4_tp8_pp1_acc32_mbs16_seq2048_zero1_tpmodeALL_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp4_tp8_pp1_acc32_mbs16_seq2048_zero1_tpmodeALL_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 2048 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 4 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 8 + tp_linear_async_communication: false + tp_mode: ALL_REDUCE +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 32 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 16 + sequence_length: 2048 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp64_tp1_pp1_acc2_mbs2_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp64_tp1_pp1_acc2_mbs2_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e7d8db0e28fe662071f92a9bf9b9a2c1b7ee5906 --- /dev/null +++ b/configs/config_1.34G_dp64_tp1_pp1_acc2_mbs2_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp64_tp1_pp1_acc2_mbs2_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 64 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 1 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 2 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 2 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp8_tp16_pp1_acc16_mbs4_seq8192_zero1_tpmodeALL_vocab131k.yaml b/configs/config_1.34G_dp8_tp16_pp1_acc16_mbs4_seq8192_zero1_tpmodeALL_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..8b442673e5091345e821af7b67ec0abf45c17b5c --- /dev/null +++ b/configs/config_1.34G_dp8_tp16_pp1_acc16_mbs4_seq8192_zero1_tpmodeALL_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp8_tp16_pp1_acc16_mbs4_seq8192_zero1_tpmodeALL_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 8192 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 8 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 16 + tp_linear_async_communication: false + tp_mode: ALL_REDUCE +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 16 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 4 + sequence_length: 8192 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp8_tp1_pp1_acc256_mbs1_seq2048_zero1_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp8_tp1_pp1_acc256_mbs1_seq2048_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..c4040782d1b12cca329ff73a7884a1bc081f28ce --- /dev/null +++ b/configs/config_1.34G_dp8_tp1_pp1_acc256_mbs1_seq2048_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp8_tp1_pp1_acc256_mbs1_seq2048_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 2048 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 8 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 1 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 256 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 1 + sequence_length: 2048 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp8_tp1_pp2_acc2_mbs16_seq4096_zero0_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp8_tp1_pp2_acc2_mbs16_seq4096_zero0_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..7111c788d6db516caefbe7a95664ccbac9391917 --- /dev/null +++ b/configs/config_1.34G_dp8_tp1_pp2_acc2_mbs16_seq4096_zero0_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp8_tp1_pp2_acc2_mbs16_seq4096_zero0_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 8 + expert_parallel_size: 1 + pp: 2 + pp_engine: 1f1b + tp: 1 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 2 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 16 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp8_tp64_pp1_acc2_mbs32_seq8192_zero1_tpmodeRED_vocab131k.yaml b/configs/config_1.34G_dp8_tp64_pp1_acc2_mbs32_seq8192_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..818ec9198d517a8f289e4a3a1ad277d533ab8d27 --- /dev/null +++ b/configs/config_1.34G_dp8_tp64_pp1_acc2_mbs32_seq8192_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp8_tp64_pp1_acc2_mbs32_seq8192_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 8192 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 8 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 64 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 2 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 32 + sequence_length: 8192 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp8_tp64_pp1_acc8_mbs2_seq32768_zero1_tpmodeALL_vocab131k.yaml b/configs/config_1.34G_dp8_tp64_pp1_acc8_mbs2_seq32768_zero1_tpmodeALL_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..d3ec561c8dee6597fdb926697a9ce821b1bd3b76 --- /dev/null +++ b/configs/config_1.34G_dp8_tp64_pp1_acc8_mbs2_seq32768_zero1_tpmodeALL_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp8_tp64_pp1_acc8_mbs2_seq32768_zero1_tpmodeALL_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 32768 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 8 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 64 + tp_linear_async_communication: false + tp_mode: ALL_REDUCE +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 8 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 2 + sequence_length: 32768 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_1.34G_dp8_tp8_pp1_acc1_mbs16_seq32768_zero1_tpmodeALL_vocab131k.yaml b/configs/config_1.34G_dp8_tp8_pp1_acc1_mbs16_seq32768_zero1_tpmodeALL_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..1eead566aa95456a9ddc73e4c978591f92850871 --- /dev/null +++ b/configs/config_1.34G_dp8_tp8_pp1_acc1_mbs16_seq32768_zero1_tpmodeALL_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 1.34G_dp8_tp8_pp1_acc1_mbs16_seq32768_zero1_tpmodeALL_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 2048 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 32768 + num_attention_heads: 32 + num_hidden_layers: 16 + num_key_value_heads: 8 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 8 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 8 + tp_linear_async_communication: false + tp_mode: ALL_REDUCE +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 1 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 16 + sequence_length: 32768 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_3.27G_dp32_tp8_pp2_acc1_mbs1_seq2048_zero0_tpmodeRED_l28_h3072_heads24.yaml b/configs/config_3.27G_dp32_tp8_pp2_acc1_mbs1_seq2048_zero0_tpmodeRED_l28_h3072_heads24.yaml new file mode 100644 index 0000000000000000000000000000000000000000..4293340bbf60c05ffab7fb87dc8bb68a520ca019 --- /dev/null +++ b/configs/config_3.27G_dp32_tp8_pp2_acc1_mbs1_seq2048_zero0_tpmodeRED_l28_h3072_heads24.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: bench_seqlen.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 3.27G_dp32_tp8_pp2_acc1_mbs1_seq2048_zero0_tpmodeRED_l28_h3072_heads24 + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 3072 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 2048 + num_attention_heads: 24 + num_hidden_layers: 28 + num_key_value_heads: 24 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 32 + expert_parallel_size: 1 + pp: 2 + pp_engine: 1f1b + tp: 8 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 1 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 1 + sequence_length: 2048 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_3.57G_dp16_tp4_pp2_acc16_mbs1_seq4096_zero0_tpmodeRED_vocab131k.yaml b/configs/config_3.57G_dp16_tp4_pp2_acc16_mbs1_seq4096_zero0_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..2b7f65d680cde6b485b260ec3eaa4c2d7fd0080d --- /dev/null +++ b/configs/config_3.57G_dp16_tp4_pp2_acc16_mbs1_seq4096_zero0_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 3.57G_dp16_tp4_pp2_acc16_mbs1_seq4096_zero0_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 3072 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 28 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 16 + expert_parallel_size: 1 + pp: 2 + pp_engine: 1f1b + tp: 4 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 16 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 1 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_3.57G_dp1_tp1_pp8_acc7_mbs3_seq4096_zero0_tpmodeALL_vocab131k_cache.yaml b/configs/config_3.57G_dp1_tp1_pp8_acc7_mbs3_seq4096_zero0_tpmodeALL_vocab131k_cache.yaml new file mode 100644 index 0000000000000000000000000000000000000000..f317357d8f936e62b4182d9f2332cc92e38b0ca1 --- /dev/null +++ b/configs/config_3.57G_dp1_tp1_pp8_acc7_mbs3_seq4096_zero0_tpmodeALL_vocab131k_cache.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 3.57G_dp1_tp1_pp8_acc7_mbs3_seq4096_zero0_tpmodeALL_vocab131k_cache + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 3072 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 28 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 1 + expert_parallel_size: 1 + pp: 8 + pp_engine: 1f1b + tp: 1 + tp_linear_async_communication: false + tp_mode: ALL_REDUCE +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 7 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 3 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_3.57G_dp2_tp16_pp1_acc128_mbs1_seq4096_zero0_tpmodeRED_vocab131k.yaml b/configs/config_3.57G_dp2_tp16_pp1_acc128_mbs1_seq4096_zero0_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..4a382384f914c1ebfed4e46dc02b4ab59dd7bf5b --- /dev/null +++ b/configs/config_3.57G_dp2_tp16_pp1_acc128_mbs1_seq4096_zero0_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 3.57G_dp2_tp16_pp1_acc128_mbs1_seq4096_zero0_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 3072 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 28 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 2 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 16 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 128 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 1 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_3.57G_dp2_tp16_pp2_acc128_mbs1_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_3.57G_dp2_tp16_pp2_acc128_mbs1_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..845f5f3da60ae9ad230cff302537a5199917c474 --- /dev/null +++ b/configs/config_3.57G_dp2_tp16_pp2_acc128_mbs1_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 3.57G_dp2_tp16_pp2_acc128_mbs1_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 3072 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 28 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 2 + expert_parallel_size: 1 + pp: 2 + pp_engine: 1f1b + tp: 16 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 128 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 1 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_3.57G_dp2_tp16_pp4_acc8_mbs16_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_3.57G_dp2_tp16_pp4_acc8_mbs16_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..6d6d4a6e08c3981c2060b2995668213e522888ff --- /dev/null +++ b/configs/config_3.57G_dp2_tp16_pp4_acc8_mbs16_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 3.57G_dp2_tp16_pp4_acc8_mbs16_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 3072 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 28 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 2 + expert_parallel_size: 1 + pp: 4 + pp_engine: 1f1b + tp: 16 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 8 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 16 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_3.57G_dp4_tp2_pp2_acc16_mbs4_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_3.57G_dp4_tp2_pp2_acc16_mbs4_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..10e01c28fa58f17f29c64ecaffa71543c8fa92c1 --- /dev/null +++ b/configs/config_3.57G_dp4_tp2_pp2_acc16_mbs4_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 3.57G_dp4_tp2_pp2_acc16_mbs4_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 3072 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 28 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 4 + expert_parallel_size: 1 + pp: 2 + pp_engine: 1f1b + tp: 2 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 16 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 4 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_3.57G_dp4_tp8_pp1_acc4.0_mbs16_seq4096_zero0_tpmodeRED_vocab131k.yaml b/configs/config_3.57G_dp4_tp8_pp1_acc4.0_mbs16_seq4096_zero0_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..81dbfd88cc0a0c127c028e96695a722cb7abe016 --- /dev/null +++ b/configs/config_3.57G_dp4_tp8_pp1_acc4.0_mbs16_seq4096_zero0_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 3.57G_dp4_tp8_pp1_acc4.0_mbs16_seq4096_zero0_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 3072 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 28 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 4 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 8 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 4.0 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 16 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_3.57G_dp64_tp2_pp1_acc1_mbs4_seq4096_zero0_tpmodeRED_vocab131k.yaml b/configs/config_3.57G_dp64_tp2_pp1_acc1_mbs4_seq4096_zero0_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..04854d6fa02cb2c04f29c007bdcffcae50bbc18f --- /dev/null +++ b/configs/config_3.57G_dp64_tp2_pp1_acc1_mbs4_seq4096_zero0_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 3.57G_dp64_tp2_pp1_acc1_mbs4_seq4096_zero0_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 3072 + initializer_range: 0.02 + intermediate_size: 8192 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 28 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: true + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 64 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 2 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 1 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 4 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_37.8G_dp32_tp4_pp4_acc1_mbs1_seq2048_zero0_tpmodeRED_l80_h8192_heads64.yaml b/configs/config_37.8G_dp32_tp4_pp4_acc1_mbs1_seq2048_zero0_tpmodeRED_l80_h8192_heads64.yaml new file mode 100644 index 0000000000000000000000000000000000000000..4fb649e6c3c0c43d6722184a7791591759ddea53 --- /dev/null +++ b/configs/config_37.8G_dp32_tp4_pp4_acc1_mbs1_seq2048_zero0_tpmodeRED_l80_h8192_heads64.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: bench_seqlen.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 37.8G_dp32_tp4_pp4_acc1_mbs1_seq2048_zero0_tpmodeRED_l80_h8192_heads64 + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 8192 + initializer_range: 0.02 + intermediate_size: 131072 + is_llama_config: true + max_position_embeddings: 2048 + num_attention_heads: 64 + num_hidden_layers: 80 + num_key_value_heads: 64 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 32768 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 32 + expert_parallel_size: 1 + pp: 4 + pp_engine: 1f1b + tp: 4 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 1 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 1 + sequence_length: 2048 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_469G_dp16_tp32_pp1_acc4_mbs4_seq4096_zero0_tpmodeRED_vocab131k.yaml b/configs/config_469G_dp16_tp32_pp1_acc4_mbs4_seq4096_zero0_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..c82a5eb9601b52bf3fe7a227a12990e80b2419de --- /dev/null +++ b/configs/config_469G_dp16_tp32_pp1_acc4_mbs4_seq4096_zero0_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 469G_dp16_tp32_pp1_acc4_mbs4_seq4096_zero0_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 16384 + initializer_range: 0.02 + intermediate_size: 53248 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 128 + num_hidden_layers: 126 + num_key_value_heads: 128 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 16 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 32 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 4 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 4 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_469G_dp16_tp4_pp2_acc1_mbs16_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_469G_dp16_tp4_pp2_acc1_mbs16_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..4c3327c165f462f5969c0c9cbedc7af037fc5b56 --- /dev/null +++ b/configs/config_469G_dp16_tp4_pp2_acc1_mbs16_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 469G_dp16_tp4_pp2_acc1_mbs16_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 16384 + initializer_range: 0.02 + intermediate_size: 53248 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 128 + num_hidden_layers: 126 + num_key_value_heads: 128 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 16 + expert_parallel_size: 1 + pp: 2 + pp_engine: 1f1b + tp: 4 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 1 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 16 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_469G_dp2_tp2_pp2_acc16_mbs8_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_469G_dp2_tp2_pp2_acc16_mbs8_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..7a97961ccb375aefa0ab0ca640b01f955abbad90 --- /dev/null +++ b/configs/config_469G_dp2_tp2_pp2_acc16_mbs8_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 469G_dp2_tp2_pp2_acc16_mbs8_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 16384 + initializer_range: 0.02 + intermediate_size: 53248 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 128 + num_hidden_layers: 126 + num_key_value_heads: 128 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 2 + expert_parallel_size: 1 + pp: 2 + pp_engine: 1f1b + tp: 2 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 16 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 8 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_469G_dp2_tp2_pp32_acc64_mbs2_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_469G_dp2_tp2_pp32_acc64_mbs2_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..41ff7fa29b0e358b1220ed3df71d1b0fa2eded7f --- /dev/null +++ b/configs/config_469G_dp2_tp2_pp32_acc64_mbs2_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 469G_dp2_tp2_pp32_acc64_mbs2_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 16384 + initializer_range: 0.02 + intermediate_size: 53248 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 128 + num_hidden_layers: 126 + num_key_value_heads: 128 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 2 + expert_parallel_size: 1 + pp: 32 + pp_engine: 1f1b + tp: 2 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 64 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 2 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_469G_dp2_tp32_pp4_acc32_mbs4_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_469G_dp2_tp32_pp4_acc32_mbs4_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..005599df76c1bd193c34e373f4282fed8d059eb3 --- /dev/null +++ b/configs/config_469G_dp2_tp32_pp4_acc32_mbs4_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 469G_dp2_tp32_pp4_acc32_mbs4_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 16384 + initializer_range: 0.02 + intermediate_size: 53248 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 128 + num_hidden_layers: 126 + num_key_value_heads: 128 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 2 + expert_parallel_size: 1 + pp: 4 + pp_engine: 1f1b + tp: 32 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 32 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 4 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_8.86G_dp16_tp16_pp2_acc2_mbs8_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_8.86G_dp16_tp16_pp2_acc2_mbs8_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..ec25b17cd70ad786b1b86b408499a8d9932c2c1b --- /dev/null +++ b/configs/config_8.86G_dp16_tp16_pp2_acc2_mbs8_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 8.86G_dp16_tp16_pp2_acc2_mbs8_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 4096 + initializer_range: 0.02 + intermediate_size: 14336 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 32 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 16 + expert_parallel_size: 1 + pp: 2 + pp_engine: 1f1b + tp: 16 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 2 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 8 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_8.86G_dp1_tp16_pp1_acc4_mbs64_seq4096_zero0_tpmodeRED_vocab131k.yaml b/configs/config_8.86G_dp1_tp16_pp1_acc4_mbs64_seq4096_zero0_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..3f1f43c48e6941bbe7ae95d492c2120015b3aeef --- /dev/null +++ b/configs/config_8.86G_dp1_tp16_pp1_acc4_mbs64_seq4096_zero0_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 8.86G_dp1_tp16_pp1_acc4_mbs64_seq4096_zero0_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 4096 + initializer_range: 0.02 + intermediate_size: 14336 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 32 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 1 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 16 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 4 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 64 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_8.86G_dp1_tp16_pp2_acc2_mbs128_seq4096_zero0_tpmodeRED_vocab131k.yaml b/configs/config_8.86G_dp1_tp16_pp2_acc2_mbs128_seq4096_zero0_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..eda293d37be2caaf62063a61d38e806fbcd033d9 --- /dev/null +++ b/configs/config_8.86G_dp1_tp16_pp2_acc2_mbs128_seq4096_zero0_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 8.86G_dp1_tp16_pp2_acc2_mbs128_seq4096_zero0_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 4096 + initializer_range: 0.02 + intermediate_size: 14336 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 32 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 1 + expert_parallel_size: 1 + pp: 2 + pp_engine: 1f1b + tp: 16 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 2 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 128 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_8.86G_dp1_tp4_pp4_acc16_mbs16_seq4096_zero0_tpmodeRED_vocab131k.yaml b/configs/config_8.86G_dp1_tp4_pp4_acc16_mbs16_seq4096_zero0_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e03249c2baf53972348f6cd35b56d4c5d0893e92 --- /dev/null +++ b/configs/config_8.86G_dp1_tp4_pp4_acc16_mbs16_seq4096_zero0_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 8.86G_dp1_tp4_pp4_acc16_mbs16_seq4096_zero0_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 4096 + initializer_range: 0.02 + intermediate_size: 14336 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 32 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 1 + expert_parallel_size: 1 + pp: 4 + pp_engine: 1f1b + tp: 4 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 16 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 16 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_8.86G_dp2_tp1_pp8_acc64_mbs2_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_8.86G_dp2_tp1_pp8_acc64_mbs2_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..db7e776392bcd5a76b6e3fe465c05302a07840d6 --- /dev/null +++ b/configs/config_8.86G_dp2_tp1_pp8_acc64_mbs2_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 8.86G_dp2_tp1_pp8_acc64_mbs2_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 4096 + initializer_range: 0.02 + intermediate_size: 14336 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 32 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 2 + expert_parallel_size: 1 + pp: 8 + pp_engine: 1f1b + tp: 1 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 64 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 2 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_8.86G_dp2_tp8_pp2_acc4_mbs32_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_8.86G_dp2_tp8_pp2_acc4_mbs32_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..ff5aaaee70556c86d645087e1864fa825763f988 --- /dev/null +++ b/configs/config_8.86G_dp2_tp8_pp2_acc4_mbs32_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 8.86G_dp2_tp8_pp2_acc4_mbs32_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 4096 + initializer_range: 0.02 + intermediate_size: 14336 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 32 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 2 + expert_parallel_size: 1 + pp: 2 + pp_engine: 1f1b + tp: 8 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 4 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 32 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_8.86G_dp32_tp4_pp2_acc8_mbs1_seq4096_zero0_tpmodeRED_vocab131k.yaml b/configs/config_8.86G_dp32_tp4_pp2_acc8_mbs1_seq4096_zero0_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..c76e5f626fa0eed1c6d4d0d12f4888a7352fdd2e --- /dev/null +++ b/configs/config_8.86G_dp32_tp4_pp2_acc8_mbs1_seq4096_zero0_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 8.86G_dp32_tp4_pp2_acc8_mbs1_seq4096_zero0_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 4096 + initializer_range: 0.02 + intermediate_size: 14336 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 32 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 32 + expert_parallel_size: 1 + pp: 2 + pp_engine: 1f1b + tp: 4 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 8 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 1 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_8.86G_dp32_tp8_pp1_acc128_mbs1_seq4096_zero0_tpmodeRED_vocab131k.yaml b/configs/config_8.86G_dp32_tp8_pp1_acc128_mbs1_seq4096_zero0_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..361c9ac7c9c8ae65ca22089270e04b122f97102c --- /dev/null +++ b/configs/config_8.86G_dp32_tp8_pp1_acc128_mbs1_seq4096_zero0_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 8.86G_dp32_tp8_pp1_acc128_mbs1_seq4096_zero0_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 4096 + initializer_range: 0.02 + intermediate_size: 14336 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 32 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 32 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 8 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 128 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 1 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_8.86G_dp4_tp2_pp2_acc1_mbs64_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_8.86G_dp4_tp2_pp2_acc1_mbs64_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e359d63ffcb3c2cbcae2e38528f9cfc7b18aaf5f --- /dev/null +++ b/configs/config_8.86G_dp4_tp2_pp2_acc1_mbs64_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 8.86G_dp4_tp2_pp2_acc1_mbs64_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 4096 + initializer_range: 0.02 + intermediate_size: 14336 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 32 + num_hidden_layers: 32 + num_key_value_heads: 32 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 4 + expert_parallel_size: 1 + pp: 2 + pp_engine: 1f1b + tp: 2 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 1 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 64 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_80G_dp16_tp2_pp2_acc1_mbs16_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_80G_dp16_tp2_pp2_acc1_mbs16_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..d8fecb520a4a82263137e61f342841725205ca56 --- /dev/null +++ b/configs/config_80G_dp16_tp2_pp2_acc1_mbs16_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 80G_dp16_tp2_pp2_acc1_mbs16_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 8192 + initializer_range: 0.02 + intermediate_size: 28672 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 64 + num_hidden_layers: 80 + num_key_value_heads: 64 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 16 + expert_parallel_size: 1 + pp: 2 + pp_engine: 1f1b + tp: 2 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 1 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 16 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_80G_dp16_tp8_pp1_acc8_mbs2_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_80G_dp16_tp8_pp1_acc8_mbs2_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..fdc45d4a49630746f03b4dbf8410839be344b731 --- /dev/null +++ b/configs/config_80G_dp16_tp8_pp1_acc8_mbs2_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 80G_dp16_tp8_pp1_acc8_mbs2_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 8192 + initializer_range: 0.02 + intermediate_size: 28672 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 64 + num_hidden_layers: 80 + num_key_value_heads: 64 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 16 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 8 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 8 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 2 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_80G_dp1_tp32_pp4_acc8_mbs32_seq4096_zero0_tpmodeRED_vocab131k.yaml b/configs/config_80G_dp1_tp32_pp4_acc8_mbs32_seq4096_zero0_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..833434f5078f255e4960edeaef42d230a20c3821 --- /dev/null +++ b/configs/config_80G_dp1_tp32_pp4_acc8_mbs32_seq4096_zero0_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 80G_dp1_tp32_pp4_acc8_mbs32_seq4096_zero0_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 8192 + initializer_range: 0.02 + intermediate_size: 28672 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 64 + num_hidden_layers: 80 + num_key_value_heads: 64 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 1 + expert_parallel_size: 1 + pp: 4 + pp_engine: 1f1b + tp: 32 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 8 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 32 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_80G_dp1_tp4_pp16_acc32_mbs8_seq4096_zero0_tpmodeRED_vocab131k.yaml b/configs/config_80G_dp1_tp4_pp16_acc32_mbs8_seq4096_zero0_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..5bfaea42593aa7da53a6e9d54ed314c2a60e54aa --- /dev/null +++ b/configs/config_80G_dp1_tp4_pp16_acc32_mbs8_seq4096_zero0_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 80G_dp1_tp4_pp16_acc32_mbs8_seq4096_zero0_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 8192 + initializer_range: 0.02 + intermediate_size: 28672 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 64 + num_hidden_layers: 80 + num_key_value_heads: 64 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 1 + expert_parallel_size: 1 + pp: 16 + pp_engine: 1f1b + tp: 4 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 32 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 8 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_80G_dp2_tp8_pp1_acc1_mbs128_seq4096_zero0_tpmodeRED_vocab131k.yaml b/configs/config_80G_dp2_tp8_pp1_acc1_mbs128_seq4096_zero0_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..17a005f523e4b2f2df55dda784bc992a9e39c608 --- /dev/null +++ b/configs/config_80G_dp2_tp8_pp1_acc1_mbs128_seq4096_zero0_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 80G_dp2_tp8_pp1_acc1_mbs128_seq4096_zero0_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 8192 + initializer_range: 0.02 + intermediate_size: 28672 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 64 + num_hidden_layers: 80 + num_key_value_heads: 64 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 0 +parallelism: + dp: 2 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 8 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 1 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 128 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_80G_dp2_tp8_pp2_acc1_mbs128_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_80G_dp2_tp8_pp2_acc1_mbs128_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..f2f8a5d70884b3794d997fca176004d52dd3a04c --- /dev/null +++ b/configs/config_80G_dp2_tp8_pp2_acc1_mbs128_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 80G_dp2_tp8_pp2_acc1_mbs128_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 8192 + initializer_range: 0.02 + intermediate_size: 28672 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 64 + num_hidden_layers: 80 + num_key_value_heads: 64 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 2 + expert_parallel_size: 1 + pp: 2 + pp_engine: 1f1b + tp: 8 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 1 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 128 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100 diff --git a/configs/config_80G_dp64_tp4_pp1_acc1_mbs4_seq4096_zero1_tpmodeRED_vocab131k.yaml b/configs/config_80G_dp64_tp4_pp1_acc1_mbs4_seq4096_zero1_tpmodeRED_vocab131k.yaml new file mode 100644 index 0000000000000000000000000000000000000000..53f15a93502df44b264f77db92463facf41f9bd1 --- /dev/null +++ b/configs/config_80G_dp64_tp4_pp1_acc1_mbs4_seq4096_zero1_tpmodeRED_vocab131k.yaml @@ -0,0 +1,91 @@ +checkpoints: + checkpoint_interval: 10000 + checkpoints_path: checkpoints + checkpoints_path_is_shared_file_system: false + resume_checkpoint_path: null + save_initial_state: false +data_stages: +- data: + dataset: null + num_loading_workers: 1 + seed: 42 + name: Stable Training Stage + start_training_step: 1 +general: + benchmark_csv_path: benchmark/results/bench_final2.csv + consumed_train_samples: null + ignore_sanity_checks: true + project: debug + run: 80G_dp64_tp4_pp1_acc1_mbs4_seq4096_zero1_tpmodeRED_vocab131k + seed: 42 + step: null +lighteval: null +logging: + iteration_step_info_interval: 1 + log_level: info + log_level_replica: info +model: + ddp_bucket_cap_mb: 25 + dtype: bfloat16 + init_method: + std: 0.02 + make_vocab_size_divisible_by: 1 + model_config: + bos_token_id: 0 + eos_token_id: 0 + hidden_act: silu + hidden_size: 8192 + initializer_range: 0.02 + intermediate_size: 28672 + is_llama_config: true + max_position_embeddings: 4096 + num_attention_heads: 64 + num_hidden_layers: 80 + num_key_value_heads: 64 + pad_token_id: null + pretraining_tp: 1 + rms_norm_eps: 1.0e-05 + rope_scaling: null + tie_word_embeddings: false + use_cache: true + vocab_size: 131072 +optimizer: + accumulate_grad_in_fp32: true + clip_grad: 1.0 + learning_rate_scheduler: + learning_rate: 0.0003 + lr_decay_starting_step: null + lr_decay_steps: 13 + lr_decay_style: cosine + lr_warmup_steps: 2 + lr_warmup_style: linear + min_decay_lr: 1.0e-05 + optimizer_factory: + adam_beta1: 0.9 + adam_beta2: 0.95 + adam_eps: 1.0e-08 + name: adamW + torch_adam_is_fused: true + weight_decay: 0.01 + zero_stage: 1 +parallelism: + dp: 64 + expert_parallel_size: 1 + pp: 1 + pp_engine: 1f1b + tp: 4 + tp_linear_async_communication: true + tp_mode: REDUCE_SCATTER +profiler: null +tokenizer: + tokenizer_max_length: null + tokenizer_name_or_path: robot-test/dummy-tokenizer-wordlevel + tokenizer_revision: null +tokens: + batch_accumulation_per_replica: 1 + limit_test_batches: 0 + limit_val_batches: 0 + micro_batch_size: 4 + sequence_length: 4096 + train_steps: 100 + val_check_interval: 100