[GH-ISSUE #31] AttributeError: 'QuantLinear' object has no attribute 'weight' #25

Open
opened 2026-10-26 21:10:25 +00:00 by bsakaguchi · 2 comments
Owner

Originally created by @n1ck-guo on GitHub (Jan 6, 2025).
Original GitHub issue: https://github.com/deepseek-ai/DeepSeek-VL2/issues/31

The commit faf18023 introduce a bug for quantized deepseek-vl2 model.

AttributeError: 'QuantLinear' object has no attribute 'weight'.

It point to modeling_deepseek.py Line 888.

Old version does not have any problem.

test model address: https://huggingface.co/OPEA/deepseek-vl2-int4-sym-gptq-inc

Originally created by @n1ck-guo on GitHub (Jan 6, 2025). Original GitHub issue: https://github.com/deepseek-ai/DeepSeek-VL2/issues/31 The commit faf18023 introduce a bug for quantized deepseek-vl2 model. AttributeError: 'QuantLinear' object has no attribute 'weight'. It point to modeling_deepseek.py Line 888. Old version does not have any problem. test model address: https://huggingface.co/OPEA/deepseek-vl2-int4-sym-gptq-inc
Author
Owner

@dprokhorov17 commented on GitHub (Jan 9, 2025):

Hey @n1ck-guo , I've already faced the same issue. Check out this one: https://huggingface.co/OPEA/deepseek-vl2-int4-sym-gptq-inc/discussions/3

<!-- gh-comment-id:2579918118 --> @dprokhorov17 commented on GitHub (Jan 9, 2025): Hey @n1ck-guo , I've already faced the same issue. Check out this one: https://huggingface.co/OPEA/deepseek-vl2-int4-sym-gptq-inc/discussions/3
Author
Owner

@n1ck-guo commented on GitHub (Jan 10, 2025):

Hey @n1ck-guo , I've already faced the same issue. Check out this one: https://huggingface.co/OPEA/deepseek-vl2-int4-sym-gptq-inc/discussions/3

We uploaded the OPEA/deepseek-vl2-int4-sym-gptq-inc model and found this problem. We hope DeepSeek-VL2 could solve this in the future version so that we do not have to use the specific version for quantized model.

<!-- gh-comment-id:2581567380 --> @n1ck-guo commented on GitHub (Jan 10, 2025): > Hey @n1ck-guo , I've already faced the same issue. Check out this one: https://huggingface.co/OPEA/deepseek-vl2-int4-sym-gptq-inc/discussions/3 We uploaded the OPEA/deepseek-vl2-int4-sym-gptq-inc model and found this problem. We hope DeepSeek-VL2 could solve this in the future version so that we do not have to use the specific version for quantized model.
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#25
No description provided.