[ROCm][CI][AITER Coverage] Harden MoE sorting-backend/dispatch env-var test matrix (#58393)

Divakar Verma committed 27 days ago

[Bugfix][LogitsProcessor] Validate ':' separator in custom logits processor FQCN (#56020)

김재억 committed 27 days ago

[Perf][DSv4.1] Restore the fused query RMSNorm + MXFP8 quantization path (#57679)

Juntian Liu committed 27 days ago

[Bugfix][Quantization] Refresh online NVFP4 scales before reload post-processing (#57954)

Matej Sirovatka committed 27 days ago

[Minimax-M3][Perf] Use triton_mrope for vision tower + int64 offset fix for triton_mrope (#58526)

Thien Tran committed 27 days ago

[Kimi-K3][Perf] Dispatch GEMM for vision patch embedder (#58527)

Thien Tran committed 27 days ago

[Bugfix][MRV2] Treat padded prompt tails as spec-decode rows for hybrid models (#58434)

Nick Hill committed 27 days ago

[CI] Run DFlash2 NVFP4 acceptance test on B200; skip it on H200 35GB MIG (#58496)

Kevin H. Luu committed 27 days ago

[Perf] Batch Mamba2 prefill SSM state saves, removing GPU<->CPU syncs (#49371)

Samuel Kim committed 27 days ago

[Bugfix] Accept EOS after grammar finish in outlines backend; reject json_object at validation (#57743)

SIDDARTHA REDDY committed 27 days ago

[Bugfix] GLM-5.3-Flash: launch the kpool paged MQA logits in the varlen mode its schedule was built with (#55270)

Yifan Qiao committed 27 days ago

[Bugfix][Frontend] Count reasoning tokens for Harmony, DeepSeek-V3 and Step3 parsers (#58626)

Samyabrata Maji committed 27 days ago

[Bugfix][KVConnector] Finalize saves on steps without a forward (#57775)

Yifan Qiao committed 27 days ago

Release prompt_embeds tensor when its InputBatch slot is freed (#57988)

Khushali Desai committed 27 days ago

[KV Connector] Fix DecodeBench fp8 fill values and add a startup fill mode (#58472)

Zijing Liu committed 27 days ago

[Perf] Parallelize registered CUDA Triton kernel warmup at startup (#58582)

Misha Goin committed 27 days ago

[Bugfix][Mamba] Restore prompt-tail prefix-cache hits with MTP (#58368)

Netanel Haber committed 27 days ago

[Bugfix][Quantization] Give LM heads standard linear metadata (#58444)

Misha Goin committed 27 days ago

[ROCm][CI] Mirror the DSv4-Flash disaggregated DP EP group on MI355 (#58558)

stefankoncarevic committed 27 days ago

[Bugfix][Core] Keep every multimodal feature in the partial-block KV event (#58288)

Sense_wang committed 27 days ago