All pull requests

New pull request

[DNM] Test the llm-d router through srt-slurm
#3741cquil11 opened Oct 5, 2026Collaborator
1
feat(dsv4): refresh GB300 vLLM AgentX DSpark recipes / 更新 GB300 vLLM AgentX DSpark 配方
#3738richardhuo-nv opened Oct 5, 2026Collaborator
3
config(agentx): enable MiniMax-M3 H200 FP8 indexer cache / 配置 MiniMax-M3 H200 FP8 索引器缓存
#3736RohitNagraj opened Oct 5, 2026Collaborator
2
[AMD][MI355X] Cap the DSv4.1 Flash prefill chunk at 8192 on every point / [AMD][MI355X] 将 DSv4.1 Flash 所有数据点的 prefill 分块上限设为 8192
#3735Fangzhou-Ai opened Oct 5, 2026Collaborator
3
GLM-5.3 MI300X/MI325X MoRI AgentX topology search
#3734cquil11 opened Oct 5, 2026Collaborator
[VideoGenX] add Wan serving and deployment evidence / [VideoGenX] 新增 Wan serving 与部署证据
#3733edwingao28 opened Oct 5, 2026Collaborator7 of 10 tasks
1
fix(agentx): export AIPERF_VERSION so the aiperf mmap cache key tracks the pinned revision / 导出 AIPERF_VERSION 使 aiperf mmap 缓存键跟随固定的 aiperf 版本
#3729redhairdragon opened Oct 5, 20263 of 11 tasks
[AMD] Update GLM-5.2 MI355X image to 20261005 daily / [AMD] 更新 GLM-5.2 MI355X 镜像至 20261005 daily
#3728jiejingzhangamd opened Oct 5, 2026Collaborator6 of 7 tasks
2
CollectiveX: capture rank alignment before graph timing
#3727nv-lschneider opened Oct 5, 2026Contributor
2
CollectiveX: enable NCCL EP FP8 decode and remove token cap
#3726nv-lschneider opened Oct 5, 2026Contributor
[AMD] GLM5.2 FP4 MI355X SGLang: enable SGLANG_OPT_USE_TOPK_V2
#3724EricKing626 opened Oct 5, 2026Collaborator
8
[PowerX] Qwen3.5 MI325X AgentX measured-power coverage
#3717edwingao28 opened Oct 4, 2026Collaborator7 of 10 tasks
2
[PowerX] Qwen3.5 MI300X AgentX measured-power coverage
#3716edwingao28 opened Oct 4, 2026Collaborator5 of 11 tasks
2
[PowerX] Kimi-K3 MI355X AgentX measured-power coverage
#3715edwingao28 opened Oct 4, 2026Collaborator7 of 11 tasks
2
[Klaud Cold] Update DSv4 FP4 MI355X ATOM AgentX image to nightly_202610031557 / [Klaud Cold] 将 DSv4 FP4 MI355X ATOM AgentX 镜像更新至 nightly_202610031557
#3712Klaud-Cold opened Oct 4, 2026Collaborator
11
[Klaud Cold] Update MI300X MiniMax-M3 vLLM AgentX image to v0.31.0 / [Klaud Cold] 将 MI300X MiniMax-M3 vLLM AgentX 镜像更新至 v0.31.0
#3710Klaud-Cold opened Oct 4, 2026Collaborator
10
glm5.2 gb300 trt agentx: rc28 + agent bounce + Dynamo --publish-metrics / glm5.2 gb300 trt agentx：rc28 + agent bounce + Dynamo --publish-metrics
#3707richardhuo-nv opened Oct 4, 2026Collaborator
7
[NV] Move GB200 DeepSeek-V4-Pro AgentX disagg to the Mooncake external linker / 将 GB200 DeepSeek-V4-Pro AgentX 分离式配置切换到 Mooncake external linker
#3701weireweire opened Oct 3, 2026Collaborator
4
[AMD][MI355X] Enable INT4 quick reduce for DSv4.1 Flash on the latest ROCm 10 nightly / [AMD][MI355X] 在最新 ROCm 10 nightly 上为 DSv4.1 Flash 启用 INT4 quick reduce
#3697Fangzhou-Ai opened Oct 3, 2026Collaborator
10
[AMD][MI35X] 1003 DSV41 sglang mtp agentic
#36961am9trash opened Oct 3, 2026Collaborator
3
Update GB200 DSV41flash vLLM image and retune TP4/DEP2/DEP4 / 更新 GB200 DSV41flash vLLM 镜像并调优 TP4/DEP2/DEP4
#3692Juntian777 opened Oct 3, 2026Collaborator7 of 11 tasks
4
[AMD][MI355X] Enable DSv4.1 Flash a4w4 and INT4 quick reduce on the latest ROCm 10 nightly / [AMD][MI355X] 在最新 ROCm 10 nightly 上为 DSv4.1 Flash 启用 a4w4 与 INT4 quick reduce
#3690Fangzhou-Ai opened Oct 3, 2026Collaborator
11
[PowerX] accept native temperature artifacts / 兼容原生温度产物
#3685edwingao28 opened Oct 2, 2026Collaborator5 of 11 tasks
1
[PowerX] replace client GPU sampling with native telemetry / [PowerX] 用原生遥测替代客户端 GPU 采样
#3684edwingao28 opened Oct 2, 2026Collaborator8 of 11 tasks
5
fix(kimik3): set MC_ENABLE_DEST_DEVICE_AFFINITY on GB300 agg Mooncake recipes / fix(kimik3)：为 GB300 聚合 Mooncake 配方设置 MC_ENABLE_DEST_DEVICE_AFFINITY
#3683redhairdragon opened Oct 2, 20266 of 11 tasks