[Feature]: DeepSeek-V4-Flash-Vision-Exp

[halexan]

🚀 The feature, motivation and pitch

Support DeepSeek-V4-Flash-Vision-Exp

Alternatives

No response

Additional context

No response

Before submitting a new issue...

• Make sure you already searched for relevant issues, and asked the chatbot living at the bottom right corner of the documentation page, which can answer lots of frequently asked questions.

[wtdcode]

I would assume nothing needs to change? My cards are occupied with other issues at this moment.

[halexan]

I will try it later with v0.11.1.

[wtdcode]

I will try it later with v0.11.1.

I just reviewed their inference sample. We need to update template parsing at least.

[wtdcode]

I will try it later with v0.11.1.

It won't work and needs new support.

[halexan]

I'll waiting for your great job @wtdcode

[wtdcode]

https://github.com/wtdcode/vllm-backport/tree/dsv4-vision-exp has an intial support (but I need it does not cuase regressions).

[leflakk]

https://github.com/wtdcode/vllm-backport/tree/dsv4-vision-exp has an intial support (but I need it does not cuase regressions).

Do you plan to also correct at same time the regression you mentionned? Thank you so much for your work!

[wtdcode]

https://github.com/wtdcode/vllm-backport/tree/dsv4-vision-exp has an intial support (but I need it does not cuase regressions).

Do you plan to also correct at same time the regression you mentionned? Thank you so much for your work!

Your issue is not related to the regression I mentioned. It is probably due to just out of vram. My cards are not free at this moment but I will have a try when they are free.

[maglat]

Cant wait for your update. DS4F still is my daily runner. With vision it will become much smarter

[zuuky]

Cant wait for your update，thanks for your great job