See axolotl config

axolotl version: 0.12.2

base_model: Qwen/Qwen3-32B
# Automatically upload checkpoint and final model to HF
hub_model_id: sam2ai/qwen3-32b-en-indic-mt

  #plugins:
  #- axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin
strict: false

chat_template: qwen3
datasets:
  - path: sam2ai/en-oriya-translation
    type: chat_template
    field_messages: conversations
    message_property_mappings:
      role: from
      content: value
    roles:
      assistant:
        - gpt
      user:
        - human

val_set_size: 0.0
output_dir: ./outputs/Qwen3/Qwen3-32b-wat25
dataset_prepared_path: last_run_prepared

sequence_len: 1096
sample_packing: true
eval_sample_packing: true


load_in_4bit: true
adapter: qlora
lora_r: 16
lora_alpha: 32
lora_target_modules:
  - q_proj
  - k_proj
  - v_proj
  - o_proj
  - down_proj
  - up_proj
lora_mlp_kernel: true
lora_qkv_kernel: true
lora_o_kernel: true

wandb_project: QWEN3-WAT2025
wandb_entity:
wandb_watch:
wandb_name: Qwen3-27B-en-indic-mt
wandb_log_model:

gradient_accumulation_steps: 4
micro_batch_size: 2
num_epochs: 1
optimizer: adamw_torch_4bit
lr_scheduler: cosine
learning_rate: 0.0002

bf16: auto
tf32: false

gradient_checkpointing: offload
gradient_checkpointing_kwargs:
  use_reentrant: false
resume_from_checkpoint:
logging_steps: 1
flash_attention: true

warmup_ratio: 0.1
evals_per_epoch: 4
saves_per_epoch: 1
weight_decay: 0.0
special_tokens:

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

qwen3-32b-en-indic-mt

This model is a fine-tuned version of Qwen/Qwen3-32B on the sam2ai/en-oriya-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_TORCH_4BIT 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: 274
• training_steps: 2745

Training results

Framework versions

• PEFT 0.17.0
• Transformers 4.55.2
• Pytorch 2.7.0+gitf717b2a
• Datasets 4.0.0
• Tokenizers 0.21.4