[GH-ISSUE #106] TypeError: LlamaAttention.forward() missing 1 required positional argument: 'position_embeddings' #87

Open
opened 2026-10-26 21:11:42 +00:00 by bsakaguchi · 2 comments
Owner

Originally created by @zczhang-sjtu on GitHub (Mar 2, 2025).
Original GitHub issue: https://github.com/deepseek-ai/DeepSeek-VL2/issues/106

swift finetune deepseek vl2 tiny

Image

Image

transformers==4.49.0
torch==2.6.0

Has anyone experienced a similar problem?

Originally created by @zczhang-sjtu on GitHub (Mar 2, 2025). Original GitHub issue: https://github.com/deepseek-ai/DeepSeek-VL2/issues/106 swift finetune deepseek vl2 tiny ![Image](https://github.com/user-attachments/assets/52bf3faf-b0b2-4397-8879-0f366d7f2e84) ![Image](https://github.com/user-attachments/assets/460559f7-3619-45d0-828e-53ad9122a13b) transformers==4.49.0 torch==2.6.0 Has anyone experienced a similar problem?
Author
Owner

@sssimpleboy commented on GitHub (Mar 10, 2025):

遇到了。 降低了 transformers 版本之后可以训练了。 但是我觉得不是长久之计

<!-- gh-comment-id:2709317103 --> @sssimpleboy commented on GitHub (Mar 10, 2025): 遇到了。 降低了 transformers 版本之后可以训练了。 但是我觉得不是长久之计
Author
Owner

@bytedancer996 commented on GitHub (Aug 6, 2025):

遇到了。 降低了 transformers 版本之后可以训练了。 但是我觉得不是长久之计

自己把老版本的实现拷贝过来解决了,还是期待官方维护更新一下

<!-- gh-comment-id:3159248458 --> @bytedancer996 commented on GitHub (Aug 6, 2025): > 遇到了。 降低了 transformers 版本之后可以训练了。 但是我觉得不是长久之计 自己把老版本的实现拷贝过来解决了,还是期待官方维护更新一下
Sign in to join this conversation.
No labels
pull-request
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-VL2#87
No description provided.