[BUG] fix hidden states nan for hybrid attention models (#45849)

shanjiaz committed 5 months ago

[Bugfix][Model] Validate DefaultModelLoader / LoadConfig and fail with clear errors (#45196)

Ting SUN committed 5 months ago

[Bugfix][Gemma4] Render reasoning on assistant turns without tool_calls (#45867)

Luciano Martins committed 5 months ago

[Refactor] Remove dead quantization code and tests (#45454)

Wentao Ye committed 5 months ago

[Bugfix] MiniMax-M3 (AMD): add packed_modules_mapping and pass swiglu… (#45794)

wangjiaxin99 committed 5 months ago

[feature] MiniMax-M3-MXFP4 support added (#45896)

qli88 committed 5 months ago

[DSV4 Perf] Optimize dsv4 cudagraph by reducing `eager_break_during_capture`, 26.8% ~ 27.9% E2E TTFT improvement (#45309)

Wentao Ye committed 5 months ago

[Log] Update deepgemm log (#45857)

Wentao Ye committed 5 months ago

[Bugfix] Pass TP group to FlashInfer all-reduce fusion (#45917)

danisereb committed 5 months ago

[Bugfix][PD] Fix DSV4 disaggregated serving (#45831)

zhanqiuhu committed 5 months ago

[Kernel][Helion][1/N] Add Helion kernel for rms_norm_dynamic_per_token_quant (#34432)

Xiaohong (Sean) Chen committed 5 months ago

[DSv4 Perf] DSv4 flashinfer sparse index cache for metadata, 2%~4% TTFT improvement (#45863)

Wentao Ye committed 5 months ago

Feature: Enable Flashinfer non-gated MoE bf16 (#43853)

amirkl94 committed 5 months ago

[ROCm][AITER][Quark] Tag per-channel FP8 weights as PER_CHANNEL so AITER pre-shuffled GEMM is selected (#44626)

xaguilar-amd committed 5 months ago

[quant][autoround]Refactor INC quantization into package with INCScheme orchestrator (#40601)

Yi Liu committed 5 months ago

[XPU] Fix test_logprobs_e2e import error: pin lm-eval[api]>=0.4.12 (#44469)

Chaojun Zhang committed 5 months ago

[CI/Build] Avoid duplicate ViT CG test introduced by accident (#45654)

Isotr0py committed 5 months ago

[MM][Perf][CG] Support ViT full CUDA graph for Kimi-VL (#41992)

Oğuzhan KIR committed 5 months ago

[ROCm][Quant] Minimax-M3: Enable fp8_per_channel for bf16 weights on mi300x (#45854)

Hongxia Yang committed 5 months ago

docs, kv_offloading: add docs for selective offload (#45279)

Angelo Ruocco committed 5 months ago