See axolotl config

axolotl version: 0.12.2

base_model: google/gemma-3-12b-it

load_in_4bit: true

# gemma3 doesn't seem to play nice with ddp
ddp_find_unused_parameters: true

# huggingface repo
chat_template: gemma3
eot_tokens:
  - <end_of_turn>
datasets:
  - path: sam2ai/en-or-hi-ml-bn-translation
    type: chat_template
    field_messages: conversations
    message_property_mappings:
      role: from
      content: value
    roles:
      assistant:
        - gpt
      user:
        - human


dataset_prepared_path: last_run_prepared
val_set_size: 0.01
output_dir: ./outputs/gemma-3-12b-wat2025-qlora

adapter: qlora
lora_model_dir:

sequence_len: 2048
sample_packing: true


lora_r: 32
lora_alpha: 16
lora_dropout: 0.05
lora_target_modules: 'model.language_model.layers.[\d]+.(mlp|cross_attn|self_attn).(up|down|gate|q|k|v|o)_proj'

wandb_project: gemma3-en-indic-wat2025
wandb_entity:
wandb_watch:
wandb_name: gemma3-12b-qlora
wandb_log_model:

gradient_accumulation_steps: 4
micro_batch_size: 2
num_epochs: 1
optimizer: adamw_bnb_8bit
lr_scheduler: cosine
learning_rate: 0.0002

bf16: true
fp16:
tf32: false

gradient_checkpointing: true
gradient_checkpointing_kwargs:
  use_reentrant: false
logging_steps: 1
flash_attention: true
eager_attention:

warmup_ratio: 0.1
evals_per_epoch: 1
saves_per_epoch: 1
weight_decay: 0.0

# save_first_step: true  # uncomment this to validate checkpoint saving works with your config

outputs/gemma-3-12b-wat2025-qlora

This model is a fine-tuned version of google/gemma-3-12b-it on the sam2ai/en-or-hi-ml-bn-translation dataset.

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 0.0002
train_batch_size: 2
eval_batch_size: 2
seed: 42
distributed_type: multi-GPU
num_devices: 8
gradient_accumulation_steps: 4
total_train_batch_size: 64
total_eval_batch_size: 16
optimizer: Use OptimizerNames.ADAMW_BNB with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
lr_scheduler_type: cosine
lr_scheduler_warmup_steps: 45
training_steps: 457