[None][chore] Test CI on feature branch#16804

Closed

brb-nv wants to merge 27 commits into

NVIDIA:feat/m3_with_msafrom

brb-nv:user/brb/test-ci-on-feat-branch

Commits on Jul 23, 2026

[https://nvbugs/6435642][fix] handle session reuse worker registration (#16444)

chienchunhung
authored and
brb-nv
committed
[TRTLLMINF-99][infra] Add SLURM frontend failover to L0 (#15674)

dpitman-nvda
authored and
brb-nv
committed
[None][infra] Handle SBSA image tags for x86 CI agents (#16503)

chzblych
authored and
brb-nv
committed
[https://nvbugs/6403920][fix] Move allreduce strategy test to smoke, use full-warp hidden_size (#16389)

MrGeva
authored and
brb-nv
committed
[TRTLLM-14019][feat] Add MiniMax-M3 MSA sparse attention backend
brb-nv
committed
[TRTLLM-13969][feat] Support MiniMax M3 for Disaggregated Serving

peihu-nv
authored and
brb-nv
committed
[None][bench] Add per-layer and intra-layer nvtx markers
brb-nv
committed
[None][perf] Deduplicate MSA plan builds in eager mode
brb-nv
committed
perf: avoid Index-K cache materialization for MSA

peihu-nv
authored and
brb-nv
committed
[None][perf] Avoid paged MSA K/V materialization during prefill

peihu-nv
authored and
brb-nv
committed
[None][perf] Fuse index-q/index-k projections in MinimaxM3
brb-nv
committed
[None][perf] Fused GEMM + SwiGLU-OAI
brb-nv
committed
[None][perf] Fuse QK-norm + RoPE and overlap qkv/idx_qk math
brb-nv
committed
[TRTLLM-14093][feat] One-model Eagle3 speculative decoding for MiniMax-M3 on the MSA backend (#4)

zheyuf
authored and
brb-nv
committed
[None][perf] AllReduce + ResidualAdd + RMSNorm
brb-nv
committed
[None][perf] Fuse MiniMax-M3 MoE routing

peihu-nv
authored and
brb-nv
committed
[TRTLLM-14255][fix] migrate MiniMax M3 to loader v2 for TP8 support

peihu-nv
authored and
brb-nv
committed
[None][perf] Remove blocking cudaStreamSynchronize (#16680)
brb-nv
committed
[None][perf] Fuse MiniMax-M3 MSA block selection (#16681)

peihu-nv
authored and
brb-nv
committed
[None][feat] add BaseMultimodalDummyInputsBuilder to MiniMaxM3VL input processor (#16689)

authored and
brb-nv
committed
[None][test] Add InferenceMAX-style GSM8K accuracy eval mode (#16711)

zheyuf
authored and
brb-nv
committed
[None][perf] Use FlashInfer MXFP8 GEMM for MiniMax-M3 decode (#16695)

peihu-nv
authored and
brb-nv
committed
[None][perf] Fold q/k/v quantization into qknorm_rope_fused kernel & remove contiguous (#16699)
brb-nv
committed
[TRTLLMINF-232][infra] Trim CI to M3-only stages on feat/m3_with_msa branch (#16722)

ZhanruiSunCh
authored and
brb-nv
committed
[None][perf] Remove spurious sync in sparse fmha forward (#16771)
brb-nv
committed
[None][perf] Use FP8 MiniMax-M3 MSA indexer QK (#16742)

peihu-nv
authored and
brb-nv
committed
[None][chore] dummy commit
brb-nv
committed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

[None][chore] Test CI on feature branch#16804

[None][chore] Test CI on feature branch#16804
brb-nv wants to merge 27 commits into
NVIDIA:feat/m3_with_msafrom
brb-nv:user/brb/test-ci-on-feat-branch

Commits on Jul 23, 2026

[https://nvbugs/6435642][fix] handle session reuse worker registration (#16444)

[TRTLLMINF-99][infra] Add SLURM frontend failover to L0 (#15674)

[None][infra] Handle SBSA image tags for x86 CI agents (#16503)

[https://nvbugs/6403920][fix] Move allreduce strategy test to smoke, use full-warp hidden_size (#16389)

[TRTLLM-14019][feat] Add MiniMax-M3 MSA sparse attention backend

[TRTLLM-13969][feat] Support MiniMax M3 for Disaggregated Serving

[None][bench] Add per-layer and intra-layer nvtx markers

[None][perf] Deduplicate MSA plan builds in eager mode

perf: avoid Index-K cache materialization for MSA

[None][perf] Avoid paged MSA K/V materialization during prefill

[None][perf] Fuse index-q/index-k projections in MinimaxM3

[None][perf] Fused GEMM + SwiGLU-OAI

[None][perf] Fuse QK-norm + RoPE and overlap qkv/idx_qk math

[TRTLLM-14093][feat] One-model Eagle3 speculative decoding for MiniMax-M3 on the MSA backend (#4)

[None][perf] AllReduce + ResidualAdd + RMSNorm

[None][perf] Fuse MiniMax-M3 MoE routing

[TRTLLM-14255][fix] migrate MiniMax M3 to loader v2 for TP8 support

[None][perf] Remove blocking cudaStreamSynchronize (#16680)

[None][perf] Fuse MiniMax-M3 MSA block selection (#16681)

[None][feat] add BaseMultimodalDummyInputsBuilder to MiniMaxM3VL input processor (#16689)

[None][test] Add InferenceMAX-style GSM8K accuracy eval mode (#16711)

[None][perf] Use FlashInfer MXFP8 GEMM for MiniMax-M3 decode (#16695)

[None][perf] Fold q/k/v quantization into qknorm_rope_fused kernel & remove contiguous (#16699)

[TRTLLMINF-232][infra] Trim CI to M3-only stages on feat/m3_with_msa branch (#16722)

[None][perf] Remove spurious sync in sparse fmha forward (#16771)

[None][perf] Use FP8 MiniMax-M3 MSA indexer QK (#16742)

[None][chore] dummy commit