高级用法

约 768 字大约 3 分钟

2025-03-16

注

请确保在 LLaMA-Factory 目录下执行下述命令。

示例

LoRA 微调

（增量）预训练

llamafactory-cli train examples/train_lora/llama3_lora_pretrain.yaml

指令监督微调

llamafactory-cli train examples/train_lora/llama3_lora_sft.yaml

多模态指令监督微调

llamafactory-cli train examples/train_lora/llava1_5_lora_sft.yaml
llamafactory-cli train examples/train_lora/qwen2vl_lora_sft.yaml

DPO/ORPO/SimPO 训练

llamafactory-cli train examples/train_lora/llama3_lora_dpo.yaml

多模态 DPO/ORPO/SimPO 训练

llamafactory-cli train examples/train_lora/qwen2vl_lora_dpo.yaml

奖励模型训练

llamafactory-cli train examples/train_lora/llama3_lora_reward.yaml

PPO 训练

llamafactory-cli train examples/train_lora/llama3_lora_ppo.yaml

KTO 训练

llamafactory-cli train examples/train_lora/llama3_lora_kto.yaml

预处理数据集

对于大数据集有帮助，在配置中使用 tokenized_path 以加载预处理后的数据集。

llamafactory-cli train examples/train_lora/llama3_preprocess.yaml

在 MMLU/CMMLU/C-Eval 上评估

llamafactory-cli eval examples/train_lora/llama3_lora_eval.yaml

多机指令监督微调

FORCE_TORCHRUN=1 NNODES=2 NODE_RANK=0 MASTER_ADDR=192.168.0.1 MASTER_PORT=29500 llamafactory-cli train examples/train_lora/llama3_lora_sft.yaml
FORCE_TORCHRUN=1 NNODES=2 NODE_RANK=1 MASTER_ADDR=192.168.0.1 MASTER_PORT=29500 llamafactory-cli train examples/train_lora/llama3_lora_sft.yaml

使用 DeepSpeed ZeRO-3 平均分配显存

FORCE_TORCHRUN=1 llamafactory-cli train examples/train_lora/llama3_lora_sft_ds3.yaml

使用 Ray 在 4 张 GPU 上微调

USE_RAY=1 llamafactory-cli train examples/train_lora/llama3_lora_sft_ray.yaml

QLoRA 微调

基于 4/8 比特 Bitsandbytes/HQQ/EETQ 量化进行指令监督微调（推荐）

llamafactory-cli train examples/train_qlora/llama3_lora_sft_otfq.yaml

在 NPU 上基于 4 比特 Bitsandbytes 量化进行指令监督微调

llamafactory-cli train examples/train_qlora/llama3_lora_sft_bnb_npu.yaml

基于 4/8 比特 GPTQ 量化进行指令监督微调

llamafactory-cli train examples/train_qlora/llama3_lora_sft_gptq.yaml

基于 4 比特 AWQ 量化进行指令监督微调

llamafactory-cli train examples/train_qlora/llama3_lora_sft_awq.yaml

基于 2 比特 AQLM 量化进行指令监督微调

llamafactory-cli train examples/train_qlora/llama3_lora_sft_aqlm.yaml

全参数微调

在单机上进行指令监督微调

FORCE_TORCHRUN=1 llamafactory-cli train examples/train_full/llama3_full_sft.yaml

在多机上进行指令监督微调

FORCE_TORCHRUN=1 NNODES=2 NODE_RANK=0 MASTER_ADDR=192.168.0.1 MASTER_PORT=29500 llamafactory-cli train examples/train_full/llama3_full_sft.yaml
FORCE_TORCHRUN=1 NNODES=2 NODE_RANK=1 MASTER_ADDR=192.168.0.1 MASTER_PORT=29500 llamafactory-cli train examples/train_full/llama3_full_sft.yaml

多模态指令监督微调

FORCE_TORCHRUN=1 llamafactory-cli train examples/train_full/qwen2vl_full_sft.yaml

合并 LoRA 适配器与模型量化

合并 LoRA 适配器

注：请勿使用量化后的模型或 quantization_bit 参数来合并 LoRA 适配器。

llamafactory-cli export examples/merge_lora/llama3_lora_sft.yaml

使用 AutoGPTQ 量化模型

llamafactory-cli export examples/merge_lora/llama3_gptq.yaml

保存 Ollama 配置文件

llamafactory-cli export examples/merge_lora/llama3_full_sft.yaml

推理 LoRA 模型

使用 vLLM+TP 批量推理

python scripts/vllm_infer.py --model_name_or_path path_to_merged_model --dataset alpaca_en_demo

使用命令行对话框

llamafactory-cli chat examples/inference/llama3_lora_sft.yaml

使用浏览器对话框

llamafactory-cli webchat examples/inference/llama3_lora_sft.yaml

启动 OpenAI 风格 API

llamafactory-cli api examples/inference/llama3_lora_sft.yaml

杂项

使用 GaLore 进行全参数训练

llamafactory-cli train examples/extras/galore/llama3_full_sft.yaml

使用 APOLLO 进行全参数训练

llamafactory-cli train examples/extras/apollo/llama3_full_sft.yaml

使用 BAdam 进行全参数训练

llamafactory-cli train examples/extras/badam/llama3_full_sft.yaml

使用 Adam-mini 进行全参数训练

llamafactory-cli train examples/extras/adam_mini/qwen2_full_sft.yaml

LoRA+ 微调

llamafactory-cli train examples/extras/loraplus/llama3_lora_sft.yaml

PiSSA 微调

llamafactory-cli train examples/extras/pissa/llama3_lora_sft.yaml

深度混合微调

llamafactory-cli train examples/extras/mod/llama3_full_sft.yaml

LLaMA-Pro 微调

bash examples/extras/llama_pro/expand.sh
llamafactory-cli train examples/extras/llama_pro/llama3_freeze_sft.yaml

FSDP+QLoRA 微调

bash examples/extras/fsdp_qlora/train.sh

计算 BLEU 和 ROUGE 分数

llamafactory-cli train examples/extras/nlg_eval/llama3_lora_predict.yaml