ValueError: Can't find config.json at './best_ckpt/'

OneStepAndTwoSteps commented 1 year ago

您好,我在用您给的代码进行微调的时候,发现在最后调用模型,用 LoraArguments 读取 /best_ckpt/config.json 文件的时候,即使相关目录下面存在 config.json 文件,但是最终还是报"ValueError: Can't find config.json at './best_ckpt/'" 的错误:

lora_args = LoraArguments.from_pretrained('./best_ckpt/') ValueError: Can't find config.json at './best_ckpt/'

不知道是什么原因导致,以下是 config.json 文件的内容,您遇到过这样的问题吗,或者您知道可能是什么原因导致的吗,期待您的回复.

{ "architectures": [ "ChatGLMModel" ], "auto_map": { "AutoConfig": "configuration_chatglm.ChatGLMConfig", "AutoModel": "modeling_chatglm.ChatGLMForConditionalGeneration", "AutoModelForSeq2SeqLM": "modeling_chatglm.ChatGLMForConditionalGeneration" }, "bos_token_id": 150004, "eos_token_id": 150005, "hidden_size": 4096, "initializer_range": 0.02, "initializer_weight": false, "inner_hidden_size": 16384, "layernorm_epsilon": 1e-05, "max_sequence_length": 2048, "model_type": "chatglm", "num_attention_heads": 32, "num_layers": 28, "pad_token_id": 20003, "position_encoding_2d": true, "pre_seq_len": null, "precision": 16, "prefix_projection": false, "quantization_bit": 0, "return_dict": false, "task_specific_params": { "learning_rate": 2e-05, "learning_rate_for_task": 2e-05 }, "torch_dtype": "float16", "transformers_version": "4.27.4", "use_cache": true, "vocab_size": 150528 }

hikariming commented 1 year ago

感觉就是路径的问题，不然用绝对路径试一下？

OneStepAndTwoSteps commented 1 year ago

在colab上运行的，指明了 /content/best_ckpt/ 还是会出现同样的问题。重新运行了好几次了也没能解决。sad

hikariming commented 1 year ago

好神奇，我没遇到过，下周我们打算重启研究，到时候我看看捏

Nicole1130 commented 1 year ago

同问，期待解答 @hikariming

Shkklt commented 1 year ago

./anaconda3/lib/python3.8/site-packages/deep_training/nlp/models/lora/configuration.py 下面默认一个CONFIG_NAME ="adapter_config.json" 而不是config.json. 我改了，但是编译文件没生效。。

Nicole1130 commented 1 year ago

chatglm_finetuning原作[ssbuild]大佬给了解答

———————————————————————————————————————————————— (https://github.com/ssbuild/chatglm_finetuning/issues/149#issuecomment-1501317978) lora 按照loss保存模型已注释，因此保存的模型路径在 last_ckpt文件夹，配置文件在best_ckpt , 复制best_ckpt config.json 到 last_ckpt, 修改infor_lora_finetune 代码路径即可。有时间会更新下路径 ————————————————————————————————————————————————

hikariming / chat-dataset-baseline

ValueError: Can't find config.json at './best_ckpt/' #57