[Model Bash] DeepSeek R1 BF16 Min Latency QKV A GEMM (0.5% E2E Speedup) (#34758)

Robert Shaw committed 9 months ago

Add unit tests for fp8 output fusion of triton_attn (#34228)

Burkhard Ringlein committed 9 months ago

[Model Runner V2] Avoid prepare prefill kernel launch overhead (#34780)

Nick Hill committed 9 months ago

[Renderer] Deprecate code paths for old input processing (#34775)

Cyrus Leung committed 9 months ago

[Quantization] - Added uses_meta_device_weights to quant config (#34645)

Asaf Joseph Gardin committed 9 months ago

[Bugfix] fix activation in cpu_fused_moe_torch call (#34696)

Marek Michalowski committed 9 months ago

[Bugfix] Fix prefix creation for Qwen3.5 (#34723)

Michael Goin committed 9 months ago

[Bugfix] Fix quant RMS norm fusion for quantization with TMA-aligned scales (#33255)

ElizaWszola committed 9 months ago

[Model Runner V2] A bit more PP simplification (#34766)

Nick Hill committed 9 months ago

[CI/Build] Remove use of `skip_v1` (#34699)

Cyrus Leung committed 9 months ago

[ROCm][CI] Removed hard-coded attn backend requirement for Qwen VL (#34753)

Andreas Karatzas committed 9 months ago

[Core] Fix SSRF bypass via backslash-@ URL parsing inconsistency (#34743)

Russell Bryant committed 9 months ago

[torch.compile] Turn on silu+fp4 quant fusion by default for O1+ (#34718)

Luka Govedič committed 9 months ago

[BugFix] [Build] fix string literals comparison in indexer_k_quant_and_cache calling site (#34653)

Hongxia Yang committed 9 months ago

[Renderer] Move MM Hash parsing into Renderer (#34711)

Cyrus Leung committed 9 months ago

[CI] Remove unused precompiled wheel args from image build (#34767)

Amr Mahdi committed 9 months ago

[Attention] Refactor `check_and_update_config` (#33600)

Matthew Bonanni committed 9 months ago

[Feature] Decode Context Parallel support for GPU model runner v2 (#34179)

Wentao Ye committed 9 months ago

[Model Runner V2] Further simplification for PP (#34724)

Woosuk Kwon committed 9 months ago

[Kernel] Triton-based Top-k and Top-p sampler kernels (#33538)

Jongseok Park committed 9 months ago