ColossalAI icon indicating copy to clipboard operation
ColossalAI copied to clipboard

[BUG]: 运行chatgpt推理示例报错

Open peng06051126 opened this issue 2 years ago • 4 comments

🐛 Describe the bug

(https://github.com/hpcaitech/ColossalAI/tree/main/applications/ChatGPT)/examples/ 运行inference.py 抛出OSError: image

Environment

No response

peng06051126 avatar Mar 01 '23 06:03 peng06051126

Bot detected the issue body's language is not English, translate it automatically. 👯👭🏻🧑‍🤝‍🧑👫🧑🏿‍🤝‍🧑🏻👩🏾‍🤝‍👨🏿👬🏿


Title: [BUG]: Run the chatgpt reasoning example and report an error

Issues-translate-bot avatar Mar 01 '23 06:03 Issues-translate-bot

meet the same issue.

hurun avatar Mar 01 '23 08:03 hurun

一样

JingxinLee avatar Mar 01 '23 09:03 JingxinLee

Bot detected the issue body's language is not English, translate it automatically. 👯👭🏻🧑‍🤝‍🧑👫🧑🏿‍🤝‍🧑🏻👩🏾‍🤝‍👨🏿👬🏿


Same

Issues-translate-bot avatar Mar 01 '23 09:03 Issues-translate-bot

Hi @peng06051126 @hurun @JingxinLee Thanks for reminding. Sorry for the trouble. We have fixed it.

binmakeswell avatar Mar 02 '23 03:03 binmakeswell

Please refer to the screenshot to check if there is any problem with the inference code. I see that you only modified README.md.

peng06051126 avatar Mar 02 '23 05:03 peng06051126

Please refer to the screenshot to check if there is any problem with the inference code. I see that you only modified README.md.

Sorry, We are also fixing the bug of code.

binmakeswell avatar Mar 02 '23 07:03 binmakeswell

still error:

size mismatch for transformer.ln_f.weight: copying a param with shape torch.Size([768]) from checkpoint, the shape in current model is torch.Size([64]). size mismatch for transformer.ln_f.bias: copying a param with shape torch.Size([768]) from checkpoint, the shape in current model is torch.Size([64]). size mismatch for lm_head.weight: copying a param with shape torch.Size([50257, 768]) from checkpoint, the shape in current model is torch.Size([250880, 64]).

params: parser.add_argument('--model', default='bloom', choices=['gpt2', 'bloom', 'opt']) parser.add_argument('--pretrain', type=str, default='./actor_checkpoint_prompts.pt')

wac81 avatar Mar 02 '23 16:03 wac81

same error

JingxinLee avatar Mar 03 '23 03:03 JingxinLee