[GH-ISSUE #84] 使用fim后human eval分数很低? #71

Open
opened 2026-10-26 21:03:04 +00:00 by bsakaguchi · 4 comments
Owner

Originally created by @nullxjx on GitHub (Dec 26, 2023).
Original GitHub issue: https://github.com/deepseek-ai/DeepSeek-Coder/issues/84

不使用fim,跑出来的分数跟官方给的比较接近,但是使用fim后分数很低,低了8个点,以下是我使用的fim格式,请问有啥问题吗?

def get_full_prompt(model_name, prompt_prefix, prompt, suffix="\n"):
    """
    get_full_prompt 通过匹配模型名,使用不同的FIM格式来准备prompt
    """
    output = prompt
    if "star-coder" in model_name.lower():
        ...
    elif "deep-seek" in model_name.lower() or "deepseek" in model_name.lower():
        template = Template("<|fim▁begin|>${preprefix}${prefix}<|fim▁hole|>${suffix}<|fim▁end|>")
        output = template.substitute(preprefix=prompt_prefix, prefix=prompt, suffix=suffix)
    else:
        if prompt_prefix:
            output = prompt_prefix + output
    return output
Originally created by @nullxjx on GitHub (Dec 26, 2023). Original GitHub issue: https://github.com/deepseek-ai/DeepSeek-Coder/issues/84 不使用fim,跑出来的分数跟官方给的比较接近,但是使用fim后分数很低,低了8个点,以下是我使用的fim格式,请问有啥问题吗? ``` def get_full_prompt(model_name, prompt_prefix, prompt, suffix="\n"): """ get_full_prompt 通过匹配模型名,使用不同的FIM格式来准备prompt """ output = prompt if "star-coder" in model_name.lower(): ... elif "deep-seek" in model_name.lower() or "deepseek" in model_name.lower(): template = Template("<|fim▁begin|>${preprefix}${prefix}<|fim▁hole|>${suffix}<|fim▁end|>") output = template.substitute(preprefix=prompt_prefix, prefix=prompt, suffix=suffix) else: if prompt_prefix: output = prompt_prefix + output return output ```
Author
Owner

@guoday commented on GitHub (Jan 26, 2024):

看起来格式有点问题,请看这个readme
https://github.com/deepseek-ai/DeepSeek-Coder#2-code-insertion

<!-- gh-comment-id:1911294985 --> @guoday commented on GitHub (Jan 26, 2024): 看起来格式有点问题,请看这个readme https://github.com/deepseek-ai/DeepSeek-Coder#2-code-insertion
Author
Owner

@xidongyuandong commented on GitHub (Feb 29, 2024):

看起来格式有点问题,请看这个readme https://github.com/deepseek-ai/DeepSeek-Coder#2-code-insertion

看起来跟官方文档是一致的,他这种 fim格式问题出在哪里?

<!-- gh-comment-id:1970501507 --> @xidongyuandong commented on GitHub (Feb 29, 2024): > 看起来格式有点问题,请看这个readme https://github.com/deepseek-ai/DeepSeek-Coder#2-code-insertion 看起来跟官方文档是一致的,他这种 fim格式问题出在哪里?
Author
Owner

@guoday commented on GitHub (Feb 29, 2024):

"<|fim▁begin|>${preprefix}${prefix}<|fim▁hole|>${suffix}<|fim▁end|>"

"<|fim▁begin|>${prefix}<|fim▁hole|>${suffix}<|fim▁end|>"

看起来格式有点问题,请看这个readme https://github.com/deepseek-ai/DeepSeek-Coder#2-code-insertion

看起来跟官方文档是一致的,他这种 fim格式问题出在哪里?

"<|fim▁begin|>${prefix}<|fim▁hole|>${suffix}<|fim▁end|>" 多了${preprefix}

<!-- gh-comment-id:1970537447 --> @guoday commented on GitHub (Feb 29, 2024): > "<|fim▁begin|>${preprefix}${prefix}<|fim▁hole|>${suffix}<|fim▁end|>" "<|fim▁begin|>${prefix}<|fim▁hole|>${suffix}<|fim▁end|>" > > 看起来格式有点问题,请看这个readme https://github.com/deepseek-ai/DeepSeek-Coder#2-code-insertion > > 看起来跟官方文档是一致的,他这种 fim格式问题出在哪里? "<|fim▁begin|>${prefix}<|fim▁hole|>${suffix}<|fim▁end|>" 多了${preprefix}
Author
Owner

@xidongyuandong commented on GitHub (Feb 29, 2024):

ˇ

低了8个点,

@guoday 这个会有这么大影响么? 模型应该有足够的泛化能力吧

<!-- gh-comment-id:1970541875 --> @xidongyuandong commented on GitHub (Feb 29, 2024): ˇ > 低了8个点, @guoday 这个会有这么大影响么? 模型应该有足够的泛化能力吧
Sign in to join this conversation.
No milestone
No project
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference
deepseek-ai/DeepSeek-Coder#71
No description provided.