GPT-OSS¶

3 Steps to LoRA Fine-tuning for GPT-OSS¶

1. Install LLaMA-Factory and transformers¶

git clone --depth 1 https://github.com/hiyouga/LLaMA-Factory.git
cd LLaMA-Factory
pip install -e .
pip install -r requirements/metrics.txt --no-build-isolation
pip install "transformers==4.55.0"

2. Train GPT-OSS on a single GPU (requires VRAM > 44 GB, multi-GPU supported)¶

llamafactory-cli train examples/train_lora/gpt_lora_sft.yaml

3. Merge LoRA Weights¶

llamafactory-cli export --model_name_or_path openai/gpt-oss-20b \
    --adapter_name_or_path saves/gpt-20b/lora/sft \
    --export_dir gpt_merged

Chat with the Fine-tuned Model¶

llamafactory-cli chat --model_name_or_path gpt_merged --template gpt --skip_special_tokens False

Full Fine-tuning Script¶

### model
model_name_or_path: openai/gpt-oss-20b
trust_remote_code: true

### method
stage: sft
do_train: true
finetuning_type: full
deepspeed: examples/deepspeed/ds_z3_config.json

### dataset
dataset: identity,alpaca_en_demo
template: gpt
cutoff_len: 2048
max_samples: 1000
overwrite_cache: true
preprocessing_num_workers: 16
dataloader_num_workers: 4

### output
output_dir: saves/gpt-20b/lora/sft
logging_steps: 10
save_steps: 500
plot_loss: true
overwrite_output_dir: true
save_only_model: false
report_to: none  # choices: [none, wandb, tensorboard, swanlab, mlflow]

### train
per_device_train_batch_size: 1
gradient_accumulation_steps: 8
learning_rate: 1.0e-4
num_train_epochs: 3.0
lr_scheduler_type: cosine
warmup_ratio: 0.1
bf16: true
ddp_timeout: 180000000
resume_from_checkpoint: null

### eval
# eval_dataset: alpaca_en_demo
# val_size: 0.1
# per_device_eval_batch_size: 1
# eval_strategy: steps
# eval_steps: 500

Fine-tune the Model via Web UI: