Skip to content

[pull] main from NVIDIA:main#1452

Merged
pull[bot] merged 6 commits into
LarryXFly:mainfrom
NVIDIA:main
Jul 6, 2026
Merged

[pull] main from NVIDIA:main#1452
pull[bot] merged 6 commits into
LarryXFly:mainfrom
NVIDIA:main

Conversation

@pull

@pull pull Bot commented Jul 6, 2026

Copy link
Copy Markdown

See Commits and Changes for more details.


Created by pull[bot] (v2.0.0-alpha.4)

Can you help keep this open source service alive? 💖 Please sponsor : )

2ez4bz and others added 6 commits July 6, 2026 09:58
* Why?

Gemma 4 diverged from the Hugging Face reference for multimodal
attention and MoE routing, while chunked prefill repeated avoidable
encoder and mask work. Incremental decoding could also fail on invalid
byte prefixes, and padded KV page tables added unnecessary overhead.

* What?

This commit:

- aligns Gemma 4 masks and routing with Hugging Face, streamlines mask
  construction, and reuses multimodal embeddings across prefill chunks.
- recovers invalid HF decode streams and limit FlashInfer and KV cache
  page processing to each sequence's live block count.
- tweaks unit tests against the HF reference implementation to tighten
  bounds by an order of magnitude.

Signed-off-by: William Zhang <133824995+2ez4bz@users.noreply.github.com>
Signed-off-by: Yuanjing Xue <197832395+yuanjingx87@users.noreply.github.com>
Signed-off-by: Yuanjing Xue <197832395+yuanjingx87@users.noreply.github.com>
Co-authored-by: tburt-nv <195370667+tburt-nv@users.noreply.github.com>
Signed-off-by: Yibin Li <109242046+yibinl-nvidia@users.noreply.github.com>
Signed-off-by: Chien-Chun Hung <2679986+chienchunhung@users.noreply.github.com>
…15524)

Signed-off-by: Chien-Chun Hung <2679986+chienchunhung@users.noreply.github.com>
@pull pull Bot locked and limited conversation to collaborators Jul 6, 2026
@pull pull Bot added the ⤵️ pull label Jul 6, 2026
@pull pull Bot merged commit a0c406f into LarryXFly:main Jul 6, 2026
Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants