feat(deepseek4): serve image requests in the concurrent batch - #760

Merged

davide221 merged 16 commits into

mainfrom

feat/ds4v-batched-images

Sep 24, 2026

Commits on Sep 23, 2026

feat(qwen35): speculative decoding for image requests

mrciffa
and
claude
committed
bf2826e
feat(deepseek4): DSpark speculative decoding for image requests

mrciffa
and
claude
committed
04ce6a1
feat(deepseek4): run the vision encoder on a second GPU and stream images into prefill

mrciffa
and
claude
committed
457cf8a
feat(qwen35): serve image requests in the concurrent batch

mrciffa
and
claude
committed
bfe6f4d
Merge remote-tracking branch 'lucebox/feat/qwen35-image-spec-decode' into feat/ds4v-batched-images
mrciffa
committed
805067b
Merge remote-tracking branch 'lucebox/feat/qwen35-batched-images' into feat/ds4v-batched-images
mrciffa
committed
a0ee4ad
feat(deepseek4): serve image requests in the concurrent batch

mrciffa
and
claude
committed
9cf92ab
perf(deepseek4): prefill concurrent image requests in one shared pass

mrciffa
and
claude
committed
d2adbeb

Commits on Sep 24, 2026

fix(vision): address review on image speculative decode

mrciffa
and
claude
committed
c2b02a3
fix(qwen35): claim the slot before encoding a batched image request

mrciffa
and
claude
committed
87bce9e
fix(deepseek4): keep exceptions inside the image stream thread

mrciffa
and
claude
committed
88e54cf
fix(deepseek4): claim the slot before encoding a batched image request

mrciffa
and
claude
committed
14e2385
Merge branches 'fix754', 'fix758' and 'fix759' into fix760
mrciffa
committed
a518c7c
fix(vision): answer 503 when the image request gate is full

mrciffa
and
claude
committed
9297fe9
feat(deepseek4): queue image requests instead of refusing them

mrciffa
and
claude
committed
5720b52
Merge lucebox/main into feat/ds4v-batched-images
mrciffa
committed
a36227f