Update finetune.py

hnluo

2023-07-21 ac1ed30ed2dfe17d49ab4fbf20d9d000865e948c

Update finetune.py

 egs_modelscope/asr/paraformer/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-pytorch/finetune.py

@@ -33,10 +33,10 @@
    params.batch_bins = 2000                       # batch size，如果dataset_type="small"，batch_bins单位为fbank特征帧数，如果dataset_type="large"，batch_bins单位为毫秒，
    params.max_epoch = 20                           # 最大训练轮数
    params.lr = 0.0002                             # 设置学习率
    init_param = []
    freeze_param = []
    ignore_init_mismatch = True
    use_lora = False
    init_param = []                                 # 初始模型路径，默认加载modelscope模型初始化，例如: ["checkpoint/20epoch.pb"]
    freeze_param = []                               # 模型参数freeze, 例如: ["encoder"]
    ignore_init_mismatch = True                     # 是否忽略模型参数初始化不匹配
    use_lora = False                                # 是否使用lora进行模型微调
    params.param_dict = {"init_param":init_param, "freeze_param": freeze_param, "ignore_init_mismatch": ignore_init_mismatch}
    if use_lora:
        enable_lora = True