Revert "Add GlmOcrConfig for GLM-OCR model type recognition" (#35512)

Harry Mellor committed 8 months ago

[compile] Cleanup: Remove unnecessary +rms_norm forcing for sequence parallelism (#35410)

Jason Li committed 8 months ago

[Bugfix] Add missing activation attr to RMSNormGated (#35423)

Tib committed 8 months ago

Flashinfer cuDNN backend for Qwen3 VL ViT attention (#34580)

Max Hu committed 8 months ago

[Bugfix] Replace assert with ValueError for response_format validation in completions endpoint (#35456)

Umut Polat committed 8 months ago

[Bugfix] Use 'sum' reduction instead of 'avg' in Async TP reduce-scatter (#33088)

Wang Xingran committed 8 months ago

[Model Performance] Add Qwen3MoE tuned MoE configs for H200 (#35457)

Chengyi Nie committed 8 months ago

[Bug] correct out dtype of rms_norm_gated native path (#35369)

zofia committed 8 months ago

[Bugfix] disable allreduce_rms_fusion by default when pp size > 1 (#35424)

Jiangyun Zhu committed 8 months ago

[BugFix] Repo utils debug print patch (#35434)

Daniel Huang committed 8 months ago

[Bug] Fix outdated links in source code (#35314)

Wentao Ye committed 8 months ago

use 'max_active_experts' for moe lora input size (#33197)

gnovack committed 8 months ago

[Bugfix] Fix Qwen3NextForCausalLM packed_modules_mapping (#35413)

Jee Jee Li committed 8 months ago

[Misc] Move `GPUModelRunner.prepare_kernel_block_sizes` to utils (#35400)

Nicolò Lucchesi committed 8 months ago

[Core]Extract is_last_rank in Ray for tpu to override (#33012)

Chenyaaang committed 8 months ago

[compile] Invalidate cache for cpu flags (#35119)

Angela Yi committed 8 months ago

[Bugfix] Emit reasoning_part events in simple streaming path for Resp… (#35184)

daniel-salib committed 8 months ago

[CI] Actually run tests/kernels/quantization/test_block_fp8.py in CI (#34274)

Michael Goin committed 8 months ago

[Performance] Cublas Bf16 Gate with Fp32 Output (#35121)

roikoren755 committed 8 months ago

[Update] Use FlashInfer fast_decode_plan directly instead of replication (#34687)

Andrii Skliar committed 8 months ago