微调的各种问题 #40
Replies: 19 comments 9 replies
-
lora微调后的模型,怎样通过vllm部署调用? |
Beta Was this translation helpful? Give feedback.
-
128k微调大概需要多少A100 |
Beta Was this translation helpful? Give feedback.
-
lora微调的时候,{'loss': 0.0, 'grad_norm': 0.0, 'learning_rate': 0.0004983333333333334, 'epoch': 0.06} loss和grad_norm一直都是0 |
Beta Was this translation helpful? Give feedback.
-
你好,如何使用微调后的模型进行对话 |
Beta Was this translation helpful? Give feedback.
-
这个错是什么原因呢? python finetune.py data/AdvertiseGen/ /mnt/workspace/panyun/models/THUDM/glm-4-9b-chat configs/lora.yaml |
Beta Was this translation helpful? Give feedback.
-
8*80GA00,8k文本长度lora能调动吗?ds zero2或者zero3 |
Beta Was this translation helpful? Give feedback.
-
想问一下GLM-4-9B模型是否支持在4张T4卡共64G显存配置的服务器上实现微调,感谢,因为包括我使用多模态模型还有微调时都会出现运行错误 |
Beta Was this translation helpful? Give feedback.
-
我使用仓库中默认的lora设置,在10w条数据上使用lora微调,loss在下降到1.1附近后,模型不再收敛了,要如何解决呢? |
Beta Was this translation helpful? Give feedback.
-
请问微调之后模型输出停不下来,在后半段会一直重复某段话反复输出,有什么问题影响到吗? |
Beta Was this translation helpful? Give feedback.
-
请问单卡单机运行GLM4v的lora时,如何让gpu为指定的cuda而不是cuda:0呢 |
Beta Was this translation helpful? Give feedback.
-
想询问下GLM4和GLM2、3版本的性能差异来源。 |
Beta Was this translation helpful? Give feedback.
-
进行微调时使用lora.yaml可以进行从从保存点进行微调,但是使用ptuning_v2.yaml会出现如下图问题。代码使用的demo代码 |
Beta Was this translation helpful? Give feedback.
-
TypeError: FinetuningConfig.init() got an unexpected keyword argument |
Beta Was this translation helpful? Give feedback.
-
|
Beta Was this translation helpful? Give feedback.
-
微调过后,训练样本的问题回答的很好,但是通用能力回答的问题都很简短。这个如何提高? |
Beta Was this translation helpful? Give feedback.
-
各种微调的问题,请在此提问
Beta Was this translation helpful? Give feedback.
All reactions