-
Notifications
You must be signed in to change notification settings - Fork 4.3k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Feature/support qwenvl glm4-v phi3-v(conflict resolving) #4377
base: main
Are you sure you want to change the base?
Conversation
终于还差一个image的padding处理就能做好训练支持了。 |
@hiyouga 改的比较多捏,有空帮忙看看这个实现思路行不行。谢谢。 |
成功跑了训练。 |
暂时先不要引入更多模型,把现有的三个模型完善好,控制diff🤗 @marko1616 |
oaky,现在确实在测试别的训练模式。 |
TODO
|
你说的这个版本应该是可以运行的。glm4v没有使用任何新特性好像。如果遇到问题欢迎随时报告。
Get Outlook for Android<https://aka.ms/AAb9ysg>
…________________________________
From: Coding Steven ***@***.***>
Sent: Thursday, July 11, 2024 11:37:17 AM
To: hiyouga/LLaMA-Factory ***@***.***>
Cc: marko1616 ***@***.***>; Mention ***@***.***>
Subject: Re: [hiyouga/LLaMA-Factory] Feature/support qwenvl glm4-v phi3-v(tested) (PR #4377)
glm4v官方的demo要求的transformer版本低于您的transformer版本要求,导致无法兼容,我现在是4.40.2,请问下您的版本?
―
Reply to this email directly, view it on GitHub<#4377 (comment)>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/AKZ2M5MWLAEZCHBEZBIAKUDZLX4W3AVCNFSM6AAAAABJSALHYGVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDEMRRHE2TGMZQGA>.
You are receiving this because you were mentioned.Message ID: ***@***.***>
|
在sft训练时,eval步骤出现问题: ... File ".../glm-4v-9b/modeling_chatglm.py", line 1035, in forward File ".../glm-4v-9b/modeling_chatglm.py", line 835, in get_masks input_ids: inputs: 可能的原因:
在 sft/workflow.py 文件 59 行处设置了 pad_to_multiple_of=8,如果 attention_mask 长度不为 8 的倍数,则末尾需要用 0 填充,在 modeling_chatglm.py 第 1035 行处需要计算 full_attention_mask,eval 时不会在之前步骤重新计算 attention_mask,导致 inputs_embeds 和 attention_mask 长度不匹配,从而出现错误。 可能的修复: |
真的非常感谢你指出了这个问题与详细的分析,我会更加仔细的查看对应的功能实现,并尽快给出可用的commit。 |
补充: |
方便加个聊天方式交流一下吗?(非常感谢你完成了这些测试,我今天就会进行修正) |
行,我的qq是227154737 |
OK加了 |
What does this PR do?
Fixes #4375
Before submitting