yuyu5333 230fb55899 [Performance] Decode Offload improves the long texts performance 100% through dynamic block offload. (#17216) ...

Co-authored-by: zhangheng <hzh0425@apache.org>

2026-03-08 17:16:53 +08:00

..

4-gpu-models

[Qwen3.5] Enable MTP spec_v2 and add test for nvidia/Qwen3.5-397B-A17B-NVFP4 (#19391)

2026-03-04 14:01:25 -08:00

8-gpu-models

[Nightly] Replace MiniMax-M2 with MiniMax-M2.5 (#20083)

2026-03-07 01:15:34 -08:00

amd

[AMD] [Z-Image-Turbo Day 0] Add Z-Image-Turbo nightly test for AMD GPUs (#19733)

2026-03-05 08:36:17 -08:00

ascend

[NPU]grok2 model support (#17119)

2026-03-03 10:24:10 +08:00

attention

fix flaky test for test_kda_kernels (#19864)

2026-03-04 22:47:29 +08:00

backends

…

bench_fn

[Benchmark] Fix generated_shared_prefix attribute naming and remove args dependency (#19363)

2026-02-25 18:45:54 -08:00

constrained_decoding

…

core

[CI] Add GPT-OSS test for SM120 (#20056)

2026-03-06 11:43:04 -08:00

debug_utils

Enhance error resilience in dump comparator (#19685)

2026-03-02 19:08:35 +08:00

disaggregation

[Performance] Decode Offload improves the long texts performance 100% through dynamic block offload. (#17216)

2026-03-08 17:16:53 +08:00

distributed

[Qwen3.5] Support Qwen3.5 Pipeline Parallelism (#19670)

2026-03-07 23:34:08 +08:00

dllm

[DLLM] Add initial radix cache support (#18724)

2026-03-04 23:24:09 -08:00

ep

chore: bump mooncake version to 0.3.9 (#18316)

2026-02-07 17:30:01 +08:00

eval

Fix nightly VLM accuracy: gemma3n TP fixes + removal, latency thresholds (#19401)

2026-02-27 14:24:02 -08:00

function_call

[Tool Call] Stream DeepSeek-V3.2 function call parameters in JSON format. (#16091)

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

hicache

[Benchmark] Remove re-exports from bench_serving.py (#19130)

2026-02-21 14:30:30 -08:00

kernels

[DSv32] [GLM5] Improve Model Quality by Avoiding FP32 Precision Loss in weights_proj (#19041)

2026-02-22 16:20:51 +08:00

layers

Fix mamba2 mixer ci test (#19658)

2026-03-02 14:55:15 +08:00

lora

[Bug] Fix lora tp bug on H200 (#19769)

2026-03-04 20:11:02 -08:00

metrics

[CI] Temporarily disable flaky test_priority_metrics on CUDA (#20075)

2026-03-06 20:34:45 -08:00

mla

Relax flaky test thresholds for MLA DeepSeek V3 and AutoRound (#20068)

2026-03-06 17:26:28 -08:00

model_loading

Fix/nemotron mtp quantaized (#19433)

2026-03-03 01:07:46 -08:00

models

[AMD] Fix AMD CI - stage-b-small-1-gpu-amd (partition 7) (#20028)

2026-03-06 23:49:14 -08:00

moe

Merge stage-c-test-large-4-gpu suites into partitioned suites (#18325)

2026-02-06 15:32:33 -08:00

openai_server

fix: Properly return abort error for streaming requests if the abort is triggered by scheduler (#19357)

2026-03-03 17:18:15 -08:00

ops

[AMD] CI - Add MI35x nightly/PR tests for kv-cache-fp8 and allreduce-fusion (DeepSeek) (#19834)

2026-03-05 07:09:57 -08:00

parser

Add KimiK2Detector with tool interruption support (#19696)

2026-03-03 14:04:49 +08:00

perf

[TestFix] change LoRA tests to use NVIDIA adapter instead of Nutanix (#19642)

2026-03-02 12:55:41 -08:00

piecewise_cuda_graph

Piecewise Cuda Graph set default (#16331)

2026-03-02 23:18:07 +08:00

profiling

Migrate 4-GPU/8-GPU workflow jobs to stage-c and add CI registry decorators (#17299)

2026-01-31 22:37:22 -08:00

quant

Relax flaky test thresholds for MLA DeepSeek V3 and AutoRound (#20068)

2026-03-06 17:26:28 -08:00

radix_cache

Piecewise Cuda Graph set default (#16331)

2026-03-02 23:18:07 +08:00

rl

[feat] Enhance lora_update_weight_from_tensor for RL training (#19314)

2026-03-04 18:10:42 -08:00

rotary

feat: Support mrope_section with rope_type: "yarn" (#13313)

2026-02-14 22:51:44 +08:00

sampling

Fix flaky penalty tests by using higher temperature for effect comparison (#18380)

2026-02-12 21:08:37 +08:00

scheduler

Fix: Reject requests with a duplicate request ID which can cause server crash/hang (#19035)

2026-03-03 00:33:25 -08:00

sessions

[CI] Re-enable streaming session tests with correctness and stress coverage (#20084)

2026-03-07 18:18:11 -08:00

spec

[Spec] Refactor NaN/OOB checks to async maybe_detect_* with env-var control (#19899)

2026-03-05 13:51:05 -08:00

stress

…

tokenizer

…

utils

Decouple scheduler log printing from metrics collection (#20107)

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

vlm

[VLM] Replace conv3d proj with linear for GLM4V (#20033)

2026-03-07 22:50:47 -08:00

test_hybrid_dp_ep_tp_mtp.py

…

test_srt_backend.py

…