fzyzcjy 9d6029fb92 Fix TokenizerManager bottleneck for offline generation (#16456)

2026-01-05 16:28:25 +08:00

..

batch_invariant_ops

…

batch_overlap

[Feature] Xiaomi MiMo-V2-Flash day0 support (#15207)

2025-12-19 11:40:07 +08:00

checkpoint_engine

…

compilation

Custom All Reduce for Piecewise Cuda Graph (#15356)

2025-12-25 23:55:15 +08:00

configs

Skip local attn init metadata for mimo swa model (#16349)

2026-01-04 22:38:36 +08:00

connector

…

constrained

Adding tool calling and reasoning parser support for Intern-S1 (#14866)

2025-12-16 00:01:58 -08:00

debug_utils

Support multiple engines for router simulation in schedule simulator (#16368)

2026-01-04 11:46:05 +08:00

disaggregation

EVS Framework: Support NemotronH_Nano_VL_V2 (#14051)

2026-01-05 16:18:07 +08:00

distributed

[distributed] Clean up MoE groups in destroy_model_parallel (#15345)

2026-01-02 14:45:14 +08:00

dllm

Tiny extract ModelRunnerOutput (#15400)

2025-12-18 22:18:45 +08:00

elastic_ep

…

entrypoints

[VLM] Support request level max_dynamic_patch for OpenAI request (#16268)

2026-01-04 13:04:43 +08:00

eplb

Support EPLB balancedness prometheus metric without GPU->CPU synchronize (#15401)

2025-12-18 22:24:23 +08:00

function_call

Fix DeepSeekV31's structural tag trigger (#13394)

2025-12-31 21:13:52 +08:00

grpc

Revert "[grpc] update api to scheduler in grpc request manager" (#16387)

2026-01-04 22:05:39 -08:00

hardware_backend/npu

[DeepSeek v3.2] opt Context Parallelism: support fused moe, multi batch and fp8 kvcache (#13959)

2026-01-02 23:49:14 +08:00

layers

[JIT kernel] Apply jit per_tensor_quant_fp8 kernel (#15836)

2026-01-05 10:15:00 +08:00

lora

[LoRA] Torch native backend: rework implementation and updated tests (#15187)

2025-12-30 11:48:51 +08:00

managers

Fix TokenizerManager bottleneck for offline generation (#16456)

2026-01-05 16:28:25 +08:00

mem_cache

EVS Framework: Support NemotronH_Nano_VL_V2 (#14051)

2026-01-05 16:18:07 +08:00

metrics

Tiny add explanations to realtime token metric labels (#16009)

2026-01-03 12:36:55 +08:00

model_executor

Super tiny fix main code (#16432)

2026-01-05 10:34:25 +08:00

model_loader

Download missing shards in model weights files when not in CI (#16211)

2025-12-31 20:42:34 +08:00

models

EVS Framework: Support NemotronH_Nano_VL_V2 (#14051)

2026-01-05 16:18:07 +08:00

multimodal

EVS Framework: Support NemotronH_Nano_VL_V2 (#14051)

2026-01-05 16:18:07 +08:00

multiplex

Fix deprecation warning of sre_parse for python 3.13 (#16247)

2025-12-31 21:25:09 -08:00

parser

[VLM] Support request level max_dynamic_patch for OpenAI request (#16268)

2026-01-04 13:04:43 +08:00

sampling

Support returning customized info (#16144)

2026-01-02 15:32:18 +08:00

speculative

[SPEC_V2] Fix Acclen drop when enabling DP Attention for Spec-Overlap (#16310)

2026-01-05 11:39:23 +08:00

tokenizer

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

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

tracing

SGLang Tracing: fix attribute errors (header extraction & bootstrap span closing) (#15693)

2025-12-27 22:44:31 -06:00

utils

Tiny speed up kimi detokenizer by 10x (#16427)

2026-01-05 09:12:05 +08:00

weight_sync

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

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

constants.py

…

environ.py

Fix TokenizerManager bottleneck for offline generation (#16456)

2026-01-05 16:28:25 +08:00

server_args_config_parser.py

Fix parse args from file(#13911) (#14085)

2026-01-01 11:37:33 +08:00

server_args.py

Tiny add --log-requests-target (#16338)

2026-01-03 17:28:27 +08:00