search

MooreThreads / MooER

MooER: Open-sourced LLM for audio understanding trained on 80,000 hours of data
Other
128 stars 7 forks source link

no module torchaudio,please update the requirments.txt #12

Closed seetimee closed 1 month ago

seetimee commented 1 month ago

File "demo/app.py", line 80, in this_model, this_tokenizer = mooer_model.init_model( File "/data2/stephen/语音/MooER/src/mooer/models/mooer_model.py", line 24, in init_model llm = setup_llm(model_config, train_config, peft_config) File "/data2/stephen/语音/MooER/src/mooer/models/mooer_model.py", line 140, in setup_llm elif train_config.use_peft: AttributeError: 'NoneType' object has no attribute 'use_peft'

seetimee commented 1 month ago

The "moore-5k" is operating normally, but the "moore-80kv2" is not functioning properly.

lzl-mt commented 1 month ago

File "demo/app.py", line 80, in this_model, this_tokenizer = mooer_model.init_model( File "/data2/stephen/语音/MooER/src/mooer/models/mooer_model.py", line 24, in init_model llm = setup_llm(model_config, train_config, peft_config) File "/data2/stephen/语音/MooER/src/mooer/models/mooer_model.py", line 140, in setup_llm elif train_config.use_peft: AttributeError: 'NoneType' object has no attribute 'use_peft'

can you check your src/mooer/configs/asr_config.py? You should set your own lora_dir, or start as python app.py --lora_dir XXX. The app.py script will not, by default, reach this line because of the logic present here.

lzl-mt commented 1 month ago

e.g., python demo/app.py --task asr --cmvn_path /configs/am.mvn --encoder_path /jfs/zhenlin.liang/model/paraformer-encoder.pth --llm_path /jfs/zhenlin.liang/model/Qwen2-7B-Instruct --adapter_path /root/MooER/output/save_models/adapter_project.pt --lora_dir /root/MooER/output/save_models/lora/new_llm. you can check whether loading lora_dir from XXX in your log.

seetimee commented 1 month ago

After specifying the lora_dir and adapter_path clearly, it was fixed.