在参考run.sh微调cosyvoice 2.0时，发现qwen_pretrain_path是空的

[hixiaoxiong]

请教一下：

1. cosyvoice 2.0近期会出类似老版本examples/magicdata-read/cosyvoice里的微调脚本吗？
2. 我在参考老版本的微调脚本时，发现qwen_pretrain_path是空的，会报错，是需要自己下载qwen2.5-0.5B模型吗（如果需要，应该下载哪个版本？有预训练版和instruct版）？能直接用CosyVoice2-0.5B模型下的llm.pt吗？（试过把qwen_pretrain_path指向llm.pt还是报错）

[zchoi]

pretrained_models/CosyVoice2-0.5B/CosyVoice-BlankEN

[hixiaoxiong]

pretrained_models/CosyVoice2-0.5B/CosyVoice-BlankEN

谢谢！加了这个路径之后，examples/magicdata-read/cosyvoice/run.sh里的inference部分能跑通了，但是在跑train llm部分时出现了如下报错：“NotImplementedError: Module [Qwen2LM] is missing the required "forward" function”。我的训练脚本如下所示，基本是直接照搬examples/magicdata-read/cosyvoice/run.sh，不知道是不是哪里配置错了？感谢各位大神帮忙看一下！

export CUDA_VISIBLE_DEVICES="0"
num_gpus=$(echo $CUDA_VISIBLE_DEVICES | awk -F "," '{print NF}')
job_id=1986
dist_backend="nccl"
num_workers=2
prefetch=100
train_engine=torch_ddp
if [ ${stage} -le 5 ] && [ ${stop_stage} -ge 5 ]; then
  echo "Run train. We only support llm traning for now. If your want to train from scratch, please use conf/cosyvoice.fromscratch.yaml"
  if [ $train_engine == 'deepspeed' ]; then
    echo "Notice deepspeed has its own optimizer config. Modify conf/ds_stage2.json if necessary"
  fi
  cp data/train/parquet/data.list data/train.data.list
  cp data/dev/parquet/data.list data/dev.data.list
  for model in llm flow; do
    torchrun --nnodes=1 --nproc_per_node=$num_gpus \
        --rdzv_id=$job_id --rdzv_backend="c10d" --rdzv_endpoint="localhost:0" \
      cosyvoice/bin/train.py \
      --train_engine $train_engine \
      --config conf/cosyvoice.yaml \
      --train_data data/train.data.list \
      --cv_data data/dev.data.list \
      --model $model \
      --checkpoint $pretrained_model_dir/$model.pt \
      --model_dir `pwd`/exp/cosyvoice/$model/$train_engine \
      --tensorboard_dir `pwd`/tensorboard/cosyvoice/$model/$train_engine \
      --ddp.dist_backend $dist_backend \
      --num_workers ${num_workers} \
      --prefetch ${prefetch} \
      --pin_memory \
      --deepspeed_config ./conf/ds_stage2.json \
      --deepspeed.save_states model+optimizer
  done
fi

[wjddd]

pretrained_models/CosyVoice2-0.5B/CosyVoice-BlankEN

谢谢！加了这个路径之后，examples/magicdata-read/cosyvoice/run.sh里的inference部分能跑通了，但是在跑train llm部分时出现了如下报错：“NotImplementedError: Module [Qwen2LM] is missing the required "forward" function”。我的训练脚本如下所示，基本是直接照搬examples/magicdata-read/cosyvoice/run.sh，不知道是不是哪里配置错了？感谢各位大神帮忙看一下！

export CUDA_VISIBLE_DEVICES="0"
num_gpus=$(echo $CUDA_VISIBLE_DEVICES | awk -F "," '{print NF}')
job_id=1986
dist_backend="nccl"
num_workers=2
prefetch=100
train_engine=torch_ddp
if [ ${stage} -le 5 ] && [ ${stop_stage} -ge 5 ]; then
  echo "Run train. We only support llm traning for now. If your want to train from scratch, please use conf/cosyvoice.fromscratch.yaml"
  if [ $train_engine == 'deepspeed' ]; then
    echo "Notice deepspeed has its own optimizer config. Modify conf/ds_stage2.json if necessary"
  fi
  cp data/train/parquet/data.list data/train.data.list
  cp data/dev/parquet/data.list data/dev.data.list
  for model in llm flow; do
    torchrun --nnodes=1 --nproc_per_node=$num_gpus \
        --rdzv_id=$job_id --rdzv_backend="c10d" --rdzv_endpoint="localhost:0" \
      cosyvoice/bin/train.py \
      --train_engine $train_engine \
      --config conf/cosyvoice.yaml \
      --train_data data/train.data.list \
      --cv_data data/dev.data.list \
      --model $model \
      --checkpoint $pretrained_model_dir/$model.pt \
      --model_dir `pwd`/exp/cosyvoice/$model/$train_engine \
      --tensorboard_dir `pwd`/tensorboard/cosyvoice/$model/$train_engine \
      --ddp.dist_backend $dist_backend \
      --num_workers ${num_workers} \
      --prefetch ${prefetch} \
      --pin_memory \
      --deepspeed_config ./conf/ds_stage2.json \
      --deepspeed.save_states model+optimizer
  done
fi

请问您微调成功了吗？