mirror of
https://github.com/wassname/alignment-handbook.git
synced 2026-06-27 17:47:01 +08:00
wassname
51 lines
1.6 KiB
YAML
51 lines
1.6 KiB
YAML
# Model arguments
|
|
model_name_or_path: Qwen/Qwen3-4B-Base
|
|
model_revision: main
|
|
torch_dtype: bfloat16
|
|
attn_implementation: flash_attention_2
|
|
|
|
# Data training arguments
|
|
tokenizer_name_or_path: Qwen/Qwen3-4B # Custom tokenizer with <|im_start|> and <|im_end|> tokens
|
|
# chat_template: "{% for message in messages %}\n{% if message['role'] == 'user' %}\n{{ '<|user|>\n' + message['content'] + eos_token }}\n{% elif message['role'] == 'system' %}\n{{ '<|system|>\n' + message['content'] + eos_token }}\n{% elif message['role'] == 'assistant' %}\n{{ '<|assistant|>\n' + message['content'] + eos_token }}\n{% endif %}\n{% if loop.last and add_generation_prompt %}\n{{ '<|assistant|>' }}\n{% endif %}\n{% endfor %}"
|
|
dataset_mixer:
|
|
wassname/v2ray_4chan_formatted: 0.8
|
|
wassname/ultrachat_200k_filtered: 0.2
|
|
dataset_splits:
|
|
- train_sft
|
|
- test_sft
|
|
preprocessing_num_workers: 12
|
|
|
|
# SFT trainer config
|
|
bf16: true
|
|
do_eval: true
|
|
eval_strategy: steps
|
|
eval_steps: 200
|
|
gradient_accumulation_steps: 32
|
|
gradient_checkpointing: true
|
|
gradient_checkpointing_kwargs:
|
|
use_reentrant: False
|
|
hub_model_id: Qwen3-4B-4chan
|
|
hub_strategy: every_save
|
|
learning_rate: 2.0e-04
|
|
log_level: info
|
|
logging_steps: 5
|
|
logging_strategy: steps
|
|
lr_scheduler_type: cosine
|
|
max_seq_length: 2048
|
|
max_steps: -1
|
|
num_train_epochs: 3
|
|
output_dir: /workspace/checkpoints_new/Qwen3-4B-4chan
|
|
run_name: Qwen3-4B-4chan
|
|
overwrite_output_dir: true
|
|
per_device_eval_batch_size: 3
|
|
per_device_train_batch_size: 2
|
|
push_to_hub: false
|
|
remove_unused_columns: true
|
|
report_to:
|
|
- wandb
|
|
save_strategy: "steps"
|
|
save_steps: 1000000
|
|
save_total_limit: 1
|
|
seed: 42
|
|
warmup_ratio: 0.1
|