[BUG/Help] <ChatGLM-6B做P-tuning-v2时无法保留权重参数pytorch_model.bin>

[SKURA502]

Is there an existing issue for this?

• I have searched the existing issues

Current Behavior

日志中显示成功保存参数pytorch_model.bin：

但文档中找不到模型权重参数：

运行inference.py时报错：

Expected Behavior

No response

Steps To Reproduce

train.sh:
PRE_SEQ_LEN=128
LR=2e-2
NUM_GPUS=1

CUDA_VISIBLE_DEVICES=1 python main.py
--do_train
--train_file /home/ns/chatbot/ChatGLM2-6B/ptuning/Chinese-medical-dialogue-data-master/train.json
--validation_file /home/ns/chatbot/ChatGLM2-6B/ptuning/Chinese-medical-dialogue-data-master/dev.json
--preprocessing_num_workers 10
--prompt_column context
--response_column target
--overwrite_cache
--model_name_or_path /home/ns/chatbot/ChatGLM2-6B/chatglm2-6b
--output_dir output/adgen-chatglm2-6b-pt-$PRE_SEQ_LEN-$LR
--overwrite_output_dir
--max_source_length 64
--max_target_length 128
--per_device_train_batch_size 1
--per_device_eval_batch_size 1
--gradient_accumulation_steps 16
--predict_with_generate
--max_steps 3000
--logging_steps 10
--save_steps 10
--learning_rate $LR
--pre_seq_len $PRE_SEQ_LEN
--quantization_bit 4

Environment

- OS:win11
- Python:Python 3.8.16
- Transformers:4.36.2
- PyTorch:2.1.2
- CUDA Support (`python -c "import torch; print(torch.cuda.is_available())"`) : True

Anything else?

No response

[annpion]

transformers==4.30.2