verachen/Fun-ASR
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

add more training details

Browse Source
This commit is contained in:
pengzhendong
2025-12-29 17:19:24 +08:00
parent 95de8cfe20
commit 1fd7785b29
3 changed files with 5 additions and 13 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
docs/finetune.md
View File

@ -53,6 +53,8 @@ Modify the `audio_encoder_conf.freeze`, `audio_adaptor_conf.freeze`, and `llm_co
Set the `freeze` parameter of the modules to be fine-tuned to false(by default, only the LLM is fine-tuned). Set the `freeze` parameter of the modules to be fine-tuned to false(by default, only the LLM is fine-tuned).
For more detailed parameters, refer to: [SenseVoice Model Training and Testing](https://github.com/modelscope/FunASR/blob/main/docs/tutorial/README.md#Model%20Training%20and%20Testing)
``` ```
bash finetune.sh bash finetune.sh
``` ```

2
docs/fintune_zh.md
View File

@ -57,6 +57,8 @@ python tools/scp2jsonl.py \
将需要微调的模块 `freeze` 设置成 `false`（默认只微调 llm）。 将需要微调的模块 `freeze` 设置成 `false`（默认只微调 llm）。
更多参数细节参考：[SenseVoice 模型训练与测试](https://github.com/modelscope/FunASR/blob/main/docs/tutorial/README_zh.md#%E6%A8%A1%E5%9E%8B%E8%AE%AD%E7%BB%83%E4%B8%8E%E6%B5%8B%E8%AF%95)
``` ```
bash finetune.sh bash finetune.sh
``` ```

14
finetune.sh
View File

@ -8,17 +8,8 @@ export CUDA_VISIBLE_DEVICES="0"
gpu_num=$(echo $CUDA_VISIBLE_DEVICES | awk -F "," '{print NF}') gpu_num=$(echo $CUDA_VISIBLE_DEVICES | awk -F "," '{print NF}')
# model_name from model_hub, or model_dir in local path # model_name from model_hub, or model_dir in local path
## option 1, download model automatically
model_name_or_model_dir="FunAudioLLM/Fun-ASR-Nano-2512" model_name_or_model_dir="FunAudioLLM/Fun-ASR-Nano-2512"
## option 2, download model by git
#local_path_root=${workspace}/modelscope_models
#mkdir -p ${local_path_root}/${model_name_or_model_dir}
#git clone https://www.modelscope.cn/${model_name_or_model_dir}.git ${local_path_root}/${model_name_or_model_dir}
#model_name_or_model_dir=${local_path_root}/${model_name_or_model_dir}
# data dir, which contains: train.json, val.json # data dir, which contains: train.json, val.json
train_data=${workspace}/data/train_example.jsonl train_data=${workspace}/data/train_example.jsonl
val_data=${workspace}/data/val_example.jsonl val_data=${workspace}/data/val_example.jsonl
@ -39,7 +30,6 @@ DISTRIBUTED_ARGS="
--master_addr ${MASTER_ADDR:-127.0.0.1} \ --master_addr ${MASTER_ADDR:-127.0.0.1} \
--master_port ${MASTER_PORT:-26669} --master_port ${MASTER_PORT:-26669}
" "
echo $DISTRIBUTED_ARGS echo $DISTRIBUTED_ARGS
# funasr trainer path # funasr trainer path
@ -70,7 +60,5 @@ ${train_tool} \
++optim_conf.lr=0.0002 \ ++optim_conf.lr=0.0002 \
++audio_encoder_conf.freeze=true \ ++audio_encoder_conf.freeze=true \
++audio_adaptor_conf.freeze=true \ ++audio_adaptor_conf.freeze=true \
++llm_conf.freeze=true \ ++llm_conf.freeze=false \
++llm_conf.use_lora=true \
++llm_conf.lora_conf.freeze_lora=false \
++output_dir="${output_dir}" &> ${log_file} ++output_dir="${output_dir}" &> ${log_file}