[FixBug] Improve exception string in `tensorizer.py` (#31680)

maang committed 10 months ago

[Misc] fix this log format not space (#32112)

rongfu.leng committed 10 months ago

[CI/Build] Separate out flaky responses API tests (#32110)

Cyrus Leung committed 10 months ago

[Misc] Make `scipy` as optional audio/benchmark dependency (#32096)

Isotr0py committed 10 months ago

[KVConnector] OffloadingConnector: Fix bug in handling of preemptions (#29870)

Or Ozeri committed 10 months ago

[Hardware][AMD][CI][Bugfix] Fix AMD Quantization test group (#31713)

Matt committed 10 months ago

[CPU][BugFix] Disable AOT Compile for CPU (#32037)

Fadi Arafeh committed 10 months ago

make assume_32_bit_indexing configurable (#32044)

Laith Sakka committed 10 months ago

[MTP][GLM][Bugfix] Fixed .weight_scale loading logic that dropped MTP prediction accuracy with fp8+mtp (#32101)

Andy Liu committed 10 months ago

[Benchmark][2/2] Use spline interpolation to tune SLA variables (#32095)

Cyrus Leung committed 10 months ago

[BugFix] Wait for compute before offloading KV to CPU (#31341)

Or Ozeri committed 10 months ago

[Bugfix] Fix Qwen3-VL-Reranker model loading for sequence classification (#32089)

RickyChen / 陳昭儒 committed 10 months ago

[MISC] Add strict contiguity check for FlashInfer attention tensors (#32008)

Vadim Gimpelson committed 10 months ago

[Bugfix][Quantization] Ensure input contiguity in per_token_quant_int8 (#31637)

Vensen committed 10 months ago

Revert "[Kernels][FI] Skip trtllm attention when num_kv_heads=1 (#308… (#31617)

shyeh25 committed 10 months ago

[BugFix] scheduler: Fix resuming of preempted requests after async load (#31583)

Or Ozeri committed 10 months ago

fused_moe_kernel - cast accumulator after applying router weights (#32002)

gnovack committed 10 months ago

[LoRA][Perf] Improve FusedMoE LoRA performance for small rank (#32019)

Xin Yang committed 10 months ago

[Kernel] Optimize Sliding Window Attention in 3D Triton Kernel (#31984)

jvlunteren committed 10 months ago

[Quant] Support MXFP4 W4A16 for compressed-tensors dense models (#31926)

Michael Goin committed 10 months ago