epoch 1
Browse files- configs/minotaur-13b.yml +143 -0
- pytorch_model.bin +1 -1
configs/minotaur-13b.yml
ADDED
|
@@ -0,0 +1,143 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
base_model: huggyllama/llama-13b
|
| 2 |
+
base_model_config: huggyllama/llama-13b
|
| 3 |
+
model_type: LlamaForCausalLM
|
| 4 |
+
tokenizer_type: LlamaTokenizer
|
| 5 |
+
load_in_8bit: false
|
| 6 |
+
load_in_4bit: false
|
| 7 |
+
gptq: false
|
| 8 |
+
strict: false
|
| 9 |
+
push_dataset_to_hub: winglian
|
| 10 |
+
hf_use_auth_token: true
|
| 11 |
+
datasets:
|
| 12 |
+
- path: winglian/evals
|
| 13 |
+
data_files:
|
| 14 |
+
- hf/ARC-Challenge.jsonl
|
| 15 |
+
- hf/ARC-Easy.jsonl
|
| 16 |
+
- hf/riddle_sense.jsonl
|
| 17 |
+
- hf/piqa.jsonl
|
| 18 |
+
type: explainchoice:chat
|
| 19 |
+
- path: winglian/evals
|
| 20 |
+
data_files:
|
| 21 |
+
- hf/gsm8k.jsonl
|
| 22 |
+
- hf/winogrande.jsonl
|
| 23 |
+
type: alpaca_chat.load_qa
|
| 24 |
+
- path: winglian/evals
|
| 25 |
+
data_files:
|
| 26 |
+
- custom/n_task.jsonl
|
| 27 |
+
- custom/misconceptions.jsonl
|
| 28 |
+
- custom/context_insensitivity.jsonl
|
| 29 |
+
type: alpaca_chat
|
| 30 |
+
- path: camel-ai/math
|
| 31 |
+
type: alpaca_chat.load_camel_ai
|
| 32 |
+
- path: camel-ai/biology
|
| 33 |
+
type: alpaca_chat.load_camel_ai
|
| 34 |
+
- path: camel-ai/physics
|
| 35 |
+
type: alpaca_chat.load_camel_ai
|
| 36 |
+
- path: camel-ai/chemistry
|
| 37 |
+
type: alpaca_chat.load_camel_ai
|
| 38 |
+
- path: winglian/evals
|
| 39 |
+
data_files:
|
| 40 |
+
- custom/in_context_qa.jsonl
|
| 41 |
+
type: context_qa
|
| 42 |
+
- path: winglian/evals
|
| 43 |
+
data_files:
|
| 44 |
+
- custom/in_context_qa.jsonl
|
| 45 |
+
type: context_qa.load_404
|
| 46 |
+
- path: winglian/evals
|
| 47 |
+
data_files:
|
| 48 |
+
- custom/jokes_explained_500up.jsonl
|
| 49 |
+
type: sharegpt_jokes
|
| 50 |
+
- path: winglian/evals
|
| 51 |
+
data_files:
|
| 52 |
+
- custom/classify-self-chat.sharegpt.jsonl
|
| 53 |
+
- custom/coding-self-chat.sharegpt.jsonl
|
| 54 |
+
- custom/prose-gpt4.sharegpt.jsonl
|
| 55 |
+
- custom/prose-rewrite-gpt4.sharegpt.jsonl
|
| 56 |
+
type: sharegpt_simple.load_role
|
| 57 |
+
- path: winglian/evals
|
| 58 |
+
data_files:
|
| 59 |
+
- openai/tldr.jsonl
|
| 60 |
+
type: summarizetldr:chat
|
| 61 |
+
- path: winglian/evals
|
| 62 |
+
data_files:
|
| 63 |
+
- hellaswag/hellaswag.jsonl
|
| 64 |
+
type: explainchoice:chat
|
| 65 |
+
- path: metaeval/ScienceQA_text_only
|
| 66 |
+
type: concisechoice:chat
|
| 67 |
+
- path: teknium/GPT4-LLM-Cleaned
|
| 68 |
+
type: alpaca_chat
|
| 69 |
+
- path: teknium/GPTeacher-General-Instruct
|
| 70 |
+
data_files: gpt4-instruct-similarity-0.6-dataset.json
|
| 71 |
+
type: gpteacher:chat
|
| 72 |
+
- path: QingyiSi/Alpaca-CoT
|
| 73 |
+
data_files:
|
| 74 |
+
- Chain-of-Thought/formatted_cot_data/aqua_train.json
|
| 75 |
+
- Chain-of-Thought/formatted_cot_data/creak_train.json
|
| 76 |
+
- Chain-of-Thought/formatted_cot_data/ecqa_train.json
|
| 77 |
+
- Chain-of-Thought/formatted_cot_data/esnli_train.json
|
| 78 |
+
- Chain-of-Thought/formatted_cot_data/qasc_train.json
|
| 79 |
+
- Chain-of-Thought/formatted_cot_data/qed_train.json
|
| 80 |
+
- Chain-of-Thought/formatted_cot_data/sensemaking_train.json
|
| 81 |
+
- Chain-of-Thought/formatted_cot_data/strategyqa_train.json
|
| 82 |
+
- GPTeacher/Roleplay/formatted_roleplay-similarity_0.6-instruct-dataset.json
|
| 83 |
+
type: alpaca_chat
|
| 84 |
+
- path: ehartford/WizardLM_alpaca_evol_instruct_70k_unfiltered
|
| 85 |
+
type: alpaca_chat
|
| 86 |
+
- path: ehartford/wizard_vicuna_70k_unfiltered
|
| 87 |
+
type: sharegpt:chat
|
| 88 |
+
dataset_prepared_path: last_run_prepared
|
| 89 |
+
val_set_size: 0.02
|
| 90 |
+
adapter:
|
| 91 |
+
lora_model_dir:
|
| 92 |
+
sequence_len: 2048
|
| 93 |
+
max_packed_sequence_len: 2048
|
| 94 |
+
lora_r:
|
| 95 |
+
lora_alpha:
|
| 96 |
+
lora_dropout:
|
| 97 |
+
lora_target_modules:
|
| 98 |
+
lora_target_linear: true
|
| 99 |
+
lora_fan_in_fan_out:
|
| 100 |
+
wandb_project: minotaur-13b
|
| 101 |
+
wandb_watch:
|
| 102 |
+
wandb_run_id:
|
| 103 |
+
wandb_log_model:
|
| 104 |
+
output_dir: ./minotaur-13b
|
| 105 |
+
gradient_accumulation_steps: 1
|
| 106 |
+
micro_batch_size: 12
|
| 107 |
+
num_epochs: 3
|
| 108 |
+
optimizer: adamw_bnb_8bit
|
| 109 |
+
torchdistx_path:
|
| 110 |
+
lr_scheduler: cosine
|
| 111 |
+
learning_rate: 0.00003
|
| 112 |
+
train_on_inputs: false
|
| 113 |
+
group_by_length: true
|
| 114 |
+
bf16: true
|
| 115 |
+
fp16: false
|
| 116 |
+
tf32: true
|
| 117 |
+
gradient_checkpointing: true
|
| 118 |
+
early_stopping_patience:
|
| 119 |
+
resume_from_checkpoint:
|
| 120 |
+
local_rank:
|
| 121 |
+
logging_steps: 1
|
| 122 |
+
xformers_attention: true
|
| 123 |
+
flash_attention:
|
| 124 |
+
gptq_groupsize:
|
| 125 |
+
gptq_model_v1:
|
| 126 |
+
warmup_steps: 20
|
| 127 |
+
eval_steps: 38
|
| 128 |
+
save_steps: 56
|
| 129 |
+
load_best_model_at_end: false
|
| 130 |
+
debug:
|
| 131 |
+
deepspeed:
|
| 132 |
+
weight_decay: 0.001
|
| 133 |
+
fsdp:
|
| 134 |
+
- full_shard
|
| 135 |
+
- auto_wrap
|
| 136 |
+
fsdp_config:
|
| 137 |
+
fsdp_offload_params: true
|
| 138 |
+
fsdp_transformer_layer_cls_to_wrap: LlamaDecoderLayer
|
| 139 |
+
special_tokens:
|
| 140 |
+
bos_token: "<s>"
|
| 141 |
+
eos_token: "</s>"
|
| 142 |
+
unk_token: "<unk>"
|
| 143 |
+
|
pytorch_model.bin
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
size 26031937261
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:cfa57f5707da4a8627874b0a8cbabe73230d3585e5857fb7e15a4b15e8399d9e
|
| 3 |
size 26031937261
|