COMMITS
August 4, 2026
I
Enable the Offloading options on DeepSeek-V4-Flash (#727)
Itay Etelis committed
F
Update MiniMax-M3-MXFP8 for Blackwell (GB300) (#734)
Faradawn Yang committed
I
Enable the Offloading options on Inkling-Small (#731)
Itay Etelis committed
P
Add 2x RTX Pro 6000 recipe for Mimimax 2.7 NVFP4 (#674)
prodney-dev committed
B
[ROCm] Add MiniMax-H3 support on MI355X (#737)
BadrBasowid committed
X
[ROCm] Verify MiniMax H3 on MI300X (#732)
Xunzhuo committed
F
Update MiniMax-M3-NVFP4 for B300 Cutlass MSA decode (#735)
Faradawn Yang committed
August 3, 2026
F
[AMD][Kimi-K3] Enable fused gated RMSNorm (#733)
Fangzhou Ai committed
I
Enable the Offloading options on full-attention recipes (#706)
Itay Etelis committed
P
Gemma-4 26B-A4B recipe update (#716)
prodney-dev committed
Z
Add DiffusionGemma 26B-A4B-IT CPU (Intel Xeon 6) recipe (#726)
zhao, zhenhui committed
R
Add MiniMaxAI/MiniMax-H3 recipe (#725)
Roy Wang committed
I
Add OffloadingConnector options to the KV Offload row (#704)
Itay Etelis committed
August 2, 2026
F
Minimax M3 disagg (#687)
Faradawn Yang committed
K
Kimi-K3: add mla_prefill_backend TRTLLM_RAGGED to attention-config (#718)
Kyle Sayers committed
P
Nemotron 3 Super recipe for DGX Station (#721)
prodney-dev committed
August 1, 2026
Y
Update deepseek-v4-flash-0731 Docker image version to v0.25.0 to fix unStablize bug (#723)
yiminghub2024 committed
July 31, 2026
R
R
Add DeepSeek-V4-Flash 0731 and DSpark variants (#717)
Roy Wang committed
R
fix: upgrade next to 15.5.22 (CVE-2026-45109 + 9 more) (#715)
Roy Wang committed
P
Add/Update DGX Spark recipes (#713)
prodney-dev committed
F
Add B200 multi-node TP+PP profile to Kimi-K3 (#695)
functionstackx committed
P
RTX 5090 recipes (#663)
prodney-dev committed
July 30, 2026
W
Add TML Inkling Small recipe (#712)
Woosuk Kwon committed
F
[ROCm] Promote tuned Kimi-K2.5 MXFP4 MoE env vars into extra_env (#703)
Fangzhou Ai committed
Y
Update H20 config & Best Practices in Kimi-K3.yaml (#711)
yiminghub2024 committed
R
Update K3 Hopper config (#710)
Roy Wang committed
R
R
Render one command per node in multi-node command blocks (#708)
Roy Wang committed
R
Update K3 hopper config (#707)
Roy Wang committed