mirror of
https://github.com/hiyouga/LLaMA-Factory.git
synced 2025-08-04 04:32:50 +08:00
update readme
Former-commit-id: 6c94305e4746c9a735ff62a6428e295d1a67da52
This commit is contained in:
parent
eac2a5b1d3
commit
a9d5b4b68e
@ -451,7 +451,7 @@ CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
|
||||
```
|
||||
|
||||
> [!TIP]
|
||||
> Use `--adapter_name_or_path path_to_sft_checkpoint,path_to_ppo_checkpoint` to infer the fine-tuned model.
|
||||
> Use `--adapter_name_or_path path_to_sft_checkpoint,path_to_ppo_checkpoint` to infer the fine-tuned model if `--create_new_adapter` was enabled.
|
||||
|
||||
> [!WARNING]
|
||||
> Use `--per_device_train_batch_size=1` for LLaMA-2 models in fp16 PPO training.
|
||||
@ -482,7 +482,7 @@ CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
|
||||
```
|
||||
|
||||
> [!TIP]
|
||||
> Use `--adapter_name_or_path path_to_sft_checkpoint,path_to_dpo_checkpoint` to infer the fine-tuned model.
|
||||
> Use `--adapter_name_or_path path_to_sft_checkpoint,path_to_dpo_checkpoint` to infer the fine-tuned model if `--create_new_adapter` was enabled.
|
||||
|
||||
### Distributed Training
|
||||
|
||||
@ -570,7 +570,7 @@ deepspeed --num_gpus 8 src/train_bash.py \
|
||||
### Merge LoRA weights and export model
|
||||
|
||||
```bash
|
||||
CUDA_VISIBLE_DEVICES=0 python src/export_model.py \
|
||||
CUDA_VISIBLE_DEVICES= python src/export_model.py \
|
||||
--model_name_or_path path_to_llama_model \
|
||||
--adapter_name_or_path path_to_checkpoint \
|
||||
--template default \
|
||||
|
@ -450,7 +450,7 @@ CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
|
||||
```
|
||||
|
||||
> [!TIP]
|
||||
> 使用 `--adapter_name_or_path path_to_sft_checkpoint,path_to_ppo_checkpoint` 来进行微调模型的推理。
|
||||
> 如果开启了 `--create_new_adapter`,则使用 `--adapter_name_or_path path_to_sft_checkpoint,path_to_ppo_checkpoint` 来进行微调模型的推理。
|
||||
|
||||
> [!WARNING]
|
||||
> 如果使用 fp16 精度进行 LLaMA-2 模型的 PPO 训练,请使用 `--per_device_train_batch_size=1`。
|
||||
@ -481,7 +481,7 @@ CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
|
||||
```
|
||||
|
||||
> [!TIP]
|
||||
> 使用 `--adapter_name_or_path path_to_sft_checkpoint,path_to_dpo_checkpoint` 来进行微调模型的推理。
|
||||
> 如果开启了 `--create_new_adapter`,则使用 `--adapter_name_or_path path_to_sft_checkpoint,path_to_dpo_checkpoint` 来进行微调模型的推理。
|
||||
|
||||
### 多 GPU 分布式训练
|
||||
|
||||
@ -569,7 +569,7 @@ deepspeed --num_gpus 8 src/train_bash.py \
|
||||
### 合并 LoRA 权重并导出模型
|
||||
|
||||
```bash
|
||||
CUDA_VISIBLE_DEVICES=0 python src/export_model.py \
|
||||
CUDA_VISIBLE_DEVICES= python src/export_model.py \
|
||||
--model_name_or_path path_to_llama_model \
|
||||
--adapter_name_or_path path_to_checkpoint \
|
||||
--template default \
|
||||
|
Loading…
x
Reference in New Issue
Block a user