[Bugfix] Skip VllmConfig re-validation for with_hf_config submodel views (#58212)

Nick Hill committed 29 days ago

[ROCm] Use silu_and_mul_with_clamp's torch._C op (#52052)

Tres committed 30 days ago

[ROCm][Model][Bugfix] Enable GLM-5.2-MXFP4 on the deepseek_v32 path and fix sparse attention correctness (#51915)

Jack Hu committed 30 days ago

[Bugfix][ROCm] Use the platform FP8 range in the concat MLA q test (#58153)

stefankoncarevic committed 30 days ago

[ROCm][Build][The Rock] Bump Triton version to 3.8.x tip-of-tree with source build in The Rock image (#58006)

rasmith committed 30 days ago

[Perf] Remove CPU-GPU sync in heterogeneous vocabulary speculative decoding (#57396)

Michael Lapshin committed 30 days ago

[CI] Shard (H200 MIG 18GB) Spec Decode Draft Model across whole-directory replicas (#58193)

Thang Nguyen committed 30 days ago

[CI] Split LM Eval TurboQuant KV Cache into per-config jobs (#57113)

Thang Nguyen committed 30 days ago

[Core] Disable JIT warmup in eager mode (#58197)

Misha Goin committed 30 days ago

[Bugfix] Backport Inductor custom-op pattern matching fix (#58189)

Misha Goin committed 30 days ago

[CI] Build the torch-nightly image on Ubuntu 24.04 (#58204)

Andrey Talman committed 30 days ago

[Kernel] Remove AllSpark INT8 W8A16 GEMM backend (#58001)

Misha Goin committed 30 days ago

[Core] structured generation mode for DiffusionGemma model (Jev-like) (#57250)

Matt Mastracci committed 30 days ago

[Refactor] Remove dead code multiple places (#58002)

Wentao Ye committed 30 days ago

[Bugfix][Model][Spec Decode] Defer disposable GLM MTP head (#55442)

lucamotz committed 30 days ago

[Bugfix][V1] Read ModelState max_model_len from model config (#58149)

hcl committed 30 days ago

[ROCm][CI] Add GELU activation for AiterExperts in the modular-kernel coverage (#58030)

Divakar Verma committed 30 days ago

[Bugfix][V1] Honor enable_jit_warmup for V2 kernel warmup (#55146)

pengyihang committed 30 days ago

[MRV2] Release weight offloader on shutdown (#57834)

Taneem Ibrahim committed 30 days ago

[Bugfix] batch_invariant: keep non-AllReduce collectives enabled on NCCL >= 2.31 (#58179)

Guanxin Li committed 30 days ago