DeepSpeed的json配置讲解:ds_config_zero3.json

ds_config_zero3.json文件内容如下:

{
   
   
    "fp16": {
   
   
        "enabled": "auto",
        "loss_scale": 0,
        "loss_scale_window": 1000,</
The easiest method is to set these DeepSpeed config values to &#39;auto&#39;. [rank1]: Traceback (most recent call last): [rank1]: File "/data1/users/heyu/find_size_and_weight/train711.py", line 625, in <module> [rank1]: train() [rank1]: File "/data1/users/heyu/find_size_and_weight/train711.py", line 617, in train [rank1]: train_result = trainer.train() [rank1]: File "/data1/users/heyu/uv_env/pyhy/lib/python3.10/site-packages/transformers/trainer.py", line 2240, in train [rank1]: return inner_training_loop( [rank1]: File "/data1/users/heyu/uv_env/pyhy/lib/python3.10/site-packages/transformers/trainer.py", line 2322, in _inner_training_loop [rank1]: self.optimizer, self.lr_scheduler = deepspeed_init(self, num_training_steps=max_steps) [rank1]: File "/data1/users/heyu/uv_env/pyhy/lib/python3.10/site-packages/transformers/integrations/deepspeed.py", line 444, in deepspeed_init [rank1]: hf_deepspeed_config.trainer_config_finalize(args, model, num_training_steps) [rank1]: File "/data1/users/heyu/uv_env/pyhy/lib/python3.10/site-packages/transformers/integrations/deepspeed.py", line 268, in trainer_config_finalize [rank1]: raise ValueError( [rank1]: ValueError: Please correct the following DeepSpeed config values that mismatch TrainingArguments values: [rank1]: - ds scheduler.params.warmup_max_lr=0.0001 vs hf learning_rate=1e-05 [rank1]: The easiest method is to set these DeepSpeed config values to &#39;auto&#39;. 2025-07-11 16:02:32,818 - ERROR - Training failed: Please correct the following DeepSpeed config values that mismatch TrainingArguments values: - ds scheduler.params.warmup_max_lr=0.0001 vs hf learning_rate=1e-05 The easiest method is to set these DeepSpeed config values to &#39;auto&#39;.上述代码报错,修改
最新发布
07-12
评论
添加红包

请填写红包祝福语或标题

红包个数最小为10个

红包金额最低5元

当前余额3.43前往充值 >
需支付:10.00
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包

打赏作者

Ven%

你的鼓励将是我创作的最大动力!

¥1 ¥2 ¥4 ¥6 ¥10 ¥20
扫码支付:¥1
获取中
扫码支付

您的余额不足,请更换扫码支付或充值

打赏作者

实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、付费专栏及课程。

余额充值