fix: make GLM sparse indexer decode top-k deterministic - #4

Open

yhl-amd wants to merge 8 commits into

feat/dsv4-pro-openai-anthropic-responses-apifrom

fix/glm52-deterministic-indexer-topk

Commits on Jul 13, 2026

[Frontend] Abort engine request on client disconnect (free leaked KV)

yhl-amd
and
claude
committed
2274cf5
[Frontend] Detect client disconnect via ASGI event, not polling

yhl-amd
and
claude
committed
af0ae45
[Frontend] Extend disconnect-abort to fan-out/multimodal; narrow cancel except

yhl-amd
and
claude
committed
2556e6f
style: apply black formatting to api_server.py

yhl-amd
and
claude
committed
e9ebbd6
[Frontend] DeepSeek-V4 native OpenAI/Anthropic/Responses API + DSML tool parser

yhl-amd
and
claude
committed
505d2b8
style: apply black formatting

yhl-amd
and
claude
committed
999168e

Commits on Jul 14, 2026

fix: make GLM sparse indexer decode top-k deterministic
yhl-amd
committed
14cf46a
docs: document vLLM CUDA top-k reference
yhl-amd
committed
32d1dc4