Yuan Luo 343a7ac652 [GDN] Fuse GDN kkt + solve_tril into one kernel (#21411) ...

Co-authored-by: luoyuan.luo <luoyuan.luo@antgroup.com>

2026-03-29 12:02:07 +08:00

..

batch_invariant_ops

[Auto Sync] Update batch_invariant_ops.py (20260221) (#19098)

2026-02-20 18:27:31 -08:00

batch_overlap

[AMD][TBO] Fix mori ep dual stream accuracy (#19888)

2026-03-18 02:00:55 -07:00

checkpoint_engine

update pre-commit config (#18860)

2026-02-16 00:18:31 +08:00

compilation

fix: piecewise_cuda_graph get correct qo_indptr (#21452)

2026-03-28 15:57:29 -07:00

configs

docs: improve CI and testing documentation (#21202)

2026-03-23 10:48:50 -07:00

connector

Remove sync points in mamba cache + prefill cudagraph plumbing for DP (#19639)

2026-03-02 15:03:42 +08:00

constrained

[MUSA] apply_vocab_mask support musa device (#21296)

2026-03-25 21:00:58 -07:00

debug_utils

Add e2e demo test in dump comparator (#21031)

2026-03-20 22:41:01 +08:00

disaggregation

[Security] 1/N: Bind ZMQ sockets to localhost to prevent unauthenticated remote access (#21435)

2026-03-26 23:33:49 -07:00

distributed

Support Qwen3 MoE context parallel (#18233)

2026-03-22 01:27:20 -07:00

dllm

Decouple scheduler log printing from metrics collection (#20107)

2026-03-07 22:09:10 -08:00

elastic_ep

[5/N] (Elastic EP) Use GPU P2P to exchange expert weights during EPLB as much as possible (#12068)

2026-03-16 18:40:58 +08:00

entrypoints

feat: enable CUDA graph and timestamp for the whisper model(#21190)

2026-03-29 01:46:03 +08:00

eplb

[5/N] (Elastic EP) Use GPU P2P to exchange expert weights during EPLB as much as possible (#12068)

2026-03-16 18:40:58 +08:00

function_call

[feat] Enhance Kimi-K2/K2.5 function call and reasoning detection (#19552)

2026-03-19 12:57:57 -07:00

grpc

[gRPC] Extract gRPC servicer into standalone package (#20478)

2026-03-13 09:13:29 -07:00

hardware_backend

[MLX] Add native MLX execution backend for Apple Silicon Mac (#20342)

2026-03-26 00:09:17 -07:00

layers

[GDN] Fuse GDN kkt + solve_tril into one kernel (#21411)

2026-03-29 12:02:07 +08:00

lora

[1/n] lora support - Auto detect lora target modules (#21439)

2026-03-27 16:08:36 -07:00

managers

[VLM] Optimize ShmPointerMMData for multi-pickle safety and deferred unwrap (#21465)

2026-03-28 23:11:12 +08:00

mem_cache

[NVIDIA] Enable automatic NUMA configuration (#19452)

2026-03-27 18:44:13 -07:00

model_executor

fix: piecewise_cuda_graph get correct qo_indptr (#21452)

2026-03-28 15:57:29 -07:00

model_loader

[Bug Fix] GLM-V / GLM-OCR: field detection for transformers 5.x and MTP omission fix (#21134)

2026-03-23 13:19:48 -07:00

models

feat: enable CUDA graph and timestamp for the whisper model(#21190)

2026-03-29 01:46:03 +08:00

multimodal

feat: enable CUDA graph and timestamp for the whisper model(#21190)

2026-03-29 01:46:03 +08:00

multiplex

Revert "[Scheduler] Decouple maybe_send_health_check_signal from process_batch_result" (#20259)

2026-03-10 01:58:48 -07:00

observability

Simplify server startup output (#20885)

2026-03-19 13:11:37 -07:00

parser

Add Mistral Small 4 (Pixtral) support (#20708)

2026-03-18 14:15:32 -07:00

ray

[Fix] RayEngine multi-node: co-locate rank0 scheduler with Engine and fix CUDA device setting (#20722)

2026-03-19 10:27:16 -07:00

sampling

Add adjusted_filter_batch (#21260)

2026-03-26 10:59:05 +08:00

speculative

[Spec][Ngram] 3/N: Fix synchronization issues in Ngram.cpp (#21186)

2026-03-23 00:37:07 -07:00

tokenizer

[Piecewise Cuda Graph] rename, refactor and add more logging (#13675)

2025-11-21 13:28:38 +08:00

utils

Patch transformers is_base_mistral in CI to avoid HF 429 rate limiting (#21586)

2026-03-27 22:19:36 -07:00

weight_sync

[RL] Allow passing tensors of different dtypes for FlattenedTensorBucket (#13413)

2025-11-19 17:24:48 +08:00

constants.py

Support releasing CUDA graph memory when paused (#7873)

2025-10-28 14:40:50 +08:00

environ.py

Support skip-softmax attention (#19089)

2026-03-28 15:55:48 -07:00

server_args_config_parser.py

[ConfigArgumentMerger] Improve ConfigArgumentMerger compatibility with external callers (#17051)

2026-01-16 15:32:42 +08:00

server_args.py

feat: enable CUDA graph and timestamp for the whisper model(#21190)

2026-03-29 01:46:03 +08:00