[Question]: Can't load the model for 'gpt-cpm-small-cn-distill'

[lao-xu]

请提出你的问题

from paddlenlp.transformers import GPTChineseTokenizer, GPTLMHeadModel

model_name = "gpt-cpm-small-cn-distill"

tokenizer = GPTChineseTokenizer.from_pretrained(model_name)
model = GPTLMHeadModel.from_pretrained(model_name)

why say "OSError: Can't load the model for 'gpt-cpm-small-cn-distill'. If you were trying to load it from 'https://paddlenlp.bj.bcebos.com/'"?

[w5688414]

已修复，或者回退到2.6版本试一下。
#8253

[lao-xu]

已修复，或者回退到2.6版本试一下。 #8253

还是不行吧。
from paddlenlp.transformers import GPTChineseTokenizer, GPTLMHeadModel

model_name = "gpt-cpm-small-cn-distill"

tokenizer = GPTChineseTokenizer.from_pretrained(model_name)
model = GPTLMHeadModel.from_pretrained(model_name)
model.eval()

inputs = "花间一壶酒，独酌无相亲。举杯邀明月，"
inputs_ids = tokenizer(inputs)["input_ids"]
inputs_ids = paddle.to_tensor(inputs_ids, dtype="int64").unsqueeze(0)

outputs, _ = model.generate(input_ids=inputs_ids, max_length=10, decode_strategy="greedy_search", use_fast=True)

result = tokenizer.convert_ids_to_string(outputs[0].numpy().tolist())

print("Model input:", inputs)
print("Result:", result)

---

Traceback (most recent call last)
File /opt/conda/envs/python35-paddle120-env/lib/python3.10/site-packages/paddlenlp/transformers/model_utils.py:1581, in PretrainedModel._resolve_model_file_path(cls, pretrained_model_name_or_path, from_hf_hub, from_aistudio, cache_dir, subfolder, config, convert_from_torch, use_safetensors, variant)
1579 if pretrained_model_name_or_path in cls.pretrained_init_configuration:
1580 # fetch the weight url from the pretrained_resource_files_map
-> 1581 resource_file_url = cls.pretrained_resource_files_map["model_state"][
1582 pretrained_model_name_or_path
1583 ]
1584 resolved_archive_file = cached_file(
1585 resource_file_url, _add_variant(PADDLE_WEIGHTS_NAME, variant), **cached_file_kwargs
1586 )

KeyError: 'model_state'

During handling of the above exception, another exception occurred:

OSError Traceback (most recent call last)
Cell In[3], line 6
3 model_name = "gpt-cpm-small-cn-distill"
5 tokenizer = GPTChineseTokenizer.from_pretrained(model_name)
----> 6 model = GPTLMHeadModel.from_pretrained(model_name)
7 model.eval()
9 inputs = "花间一壶酒，独酌无相亲。举杯邀明月，"

File /opt/conda/envs/python35-paddle120-env/lib/python3.10/site-packages/paddlenlp/transformers/model_utils.py:2116, in PretrainedModel.from_pretrained(cls, pretrained_model_name_or_path, *args, **kwargs)
2113 use_keep_in_fp32_modules = False
2115 # resolve model_weight file
-> 2116 resolved_archive_file, sharded_metadata, is_sharded = cls._resolve_model_file_path(
2117 pretrained_model_name_or_path,
2118 cache_dir=cache_dir,
2119 subfolder=subfolder,
2120 from_hf_hub=from_hf_hub,
2121 from_aistudio=from_aistudio,
2122 config=config,
2123 convert_from_torch=convert_from_torch,
2124 use_safetensors=use_safetensors,
2125 variant=variant,
2126 )
2128 # load pt weights early so that we know which dtype to init the model under
2129 if not is_sharded and state_dict is None:
2130 # Time to load the checkpoint

File /opt/conda/envs/python35-paddle120-env/lib/python3.10/site-packages/paddlenlp/transformers/model_utils.py:1638, in PretrainedModel._resolve_model_file_path(cls, pretrained_model_name_or_path, from_hf_hub, from_aistudio, cache_dir, subfolder, config, convert_from_torch, use_safetensors, variant)
1636 logger.info(e)
1637 # For any other exception, we throw a generic error.
-> 1638 raise EnvironmentError(
1639 f"Can't load the model for '{pretrained_model_name_or_path}'. If you were trying to load it"
1640 " from 'https://paddlenlp.bj.bcebos.com/'"
1641 )
1643 if is_local:
1644 logger.info(f"Loading weights file {archive_file}")

OSError: Can't load the model for 'gpt-cpm-small-cn-distill'. If you were trying to load it from 'https://paddlenlp.bj.bcebos.com/'

[w5688414]

我测了几次都没有问题：

[lao-xu]

我在aistduio上跑的，一直报这个错，这云环境有关系吗？

[w5688414]

查看一下你的paddle版本，可以装develop版本试一下

[yangquanbiubiu]

我是在昆仑芯R200上跑的，也碰到了同样的问题，请问你解决了吗

[lao-xu]

我是在昆仑芯R200上跑的，也碰到了同样的问题，请问你解决了吗
没有，我升级了paddlenlp版本还是不行，不知道为什么

[w5688414]

按照如下的命令试一下：

pip uninstall paddlenlp
git clone https://github.com/PaddlePaddle/PaddleNLP.git
cd PaddleNLP
pip install -e .

[github-actions]

This issue is stale because it has been open for 60 days with no activity. 当前issue 60天内无活动，被标记为stale。