Ho-Ren (Jack) ChuangandYichen Wang 6d5e16fb1c feat: Add FP4 (E2M1) KV Cache Support for MHA (#12612) ...

Signed-off-by: Ho-Ren (Jack) Chuang <horenchuang@bytedance.com>
Co-authored-by: Yichen Wang <yichen.wang@bytedance.com>

2025-11-14 22:31:35 -08:00

..

batch_invariant_ops

Revert "fallback to triton mm_persistent kernel when deepGemm fail" (#13178)

2025-11-12 22:03:35 -08:00

checkpoint_engine

[Ckpt Engine] feat: new sglang entrypoint support for update (#12216)

2025-10-30 10:39:27 +08:00

compilation

Piecewise Cuda Graph Support for gpt-oss model (#13045)

2025-11-15 09:28:00 +08:00

configs

[Quantization] Support Quark Dense + MoE FP8 & FP8 PTPC (#10485)

2025-11-13 08:16:00 -08:00

connector

[Lint] Add python/sglang to ruff F401 checks and remove unused imports in files (#11685)

2025-10-17 16:49:46 -07:00

constrained

[Feature] New structural tag support (#10691)

2025-10-20 18:25:58 +08:00

debug_utils

Add script to create a model with fewer layers for debugging (#13284)

2025-11-14 22:13:27 +08:00

disaggregation

refactor: remove duplicate function _get_bootstrap_info_from_server (#13277)

2025-11-15 02:48:17 +08:00

distributed

[Auto Sync] Update pynccl_wrapper.py, environ.py, registry.... (20251111) (#13097)

2025-11-13 15:58:47 -08:00

elastic_ep

[2/N] Added the core structure of elastic EP and the eplb algorithm with faulty rank (#10606)

2025-10-22 01:13:31 -07:00

entrypoints

chore: bump flashinfer v0.5.2 (#13242)

2025-11-14 02:47:09 -08:00

eplb

Fix nan in global scaling factor for large scale nvfp4 EP (#13162)

2025-11-12 15:32:23 -08:00

function_call

[Tool Call] Steamline function arguments when tool_choice="auto" for deepseekv31_detector (#11589)

2025-11-13 23:06:02 -08:00

grpc

Add timing metrics for requests (#12646)

2025-11-05 23:07:16 -08:00

layers

Piecewise Cuda Graph Support for gpt-oss model (#13045)

2025-11-15 09:28:00 +08:00

lora

[Ascend] LoRA: adding Ascend LoRA backend with using kernels from sgl_kernel_npu (#12288)

2025-11-12 16:20:34 -08:00

managers

Add more statistics for spec decoding (#13317)

2025-11-15 14:20:28 +08:00

mem_cache

feat: Add FP4 (E2M1) KV Cache Support for MHA (#12612)

2025-11-14 22:31:35 -08:00

metrics

refine stdout logging codes (#13015)

2025-11-12 00:16:14 +08:00

model_executor

feat: Add FP4 (E2M1) KV Cache Support for MHA (#12612)

2025-11-14 22:31:35 -08:00

model_loader

optm(checkpoint-engine): disable multi-thread loading when update weights (#12374)

2025-11-07 21:30:04 +08:00

models

Piecewise Cuda Graph Support for gpt-oss model (#13045)

2025-11-15 09:28:00 +08:00

multimodal

fix: duplicate resize images logic of qwen-vl series models (#12458)

2025-11-12 18:08:40 +08:00

multiplex

[Feature] PD-Multiplexing Context and Scheduler, lazy import spatial. (#12275)

2025-11-01 00:40:01 +08:00

parser

add Kimi k2 reasoning parser (#12702)

2025-11-06 00:37:54 +08:00

sampling

[Bug] OOM (Out-of-Memory) errors for extreme testing scenarios (min_tokens=2) (#11757)

2025-11-01 00:28:41 +08:00

speculative

[Feature] Enable CUDA graph for PD-Multiplexing. (#11595)

2025-11-14 03:39:40 +08:00

tokenizer

…

tracing

[minor] remove debug code in python/sglang/srt/compilation/weak_ref_tensor_jit.py (#13235)

2025-11-14 12:22:24 -08:00

utils

Optimized prefill cache allocation for NPU (#13288)

2025-11-15 02:40:17 +08:00

weight_sync

…

_custom_ops.py

move all get_stream in sgl_kernel to c++ to reduce the launch overhead (#12521)

2025-11-02 13:15:05 -08:00

constants.py

Support releasing CUDA graph memory when paused (#7873)

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

custom_op.py

…

environ.py

[NVIDIA] Fix use case of SGLANG_ENABLE_FLASHINFER_GEMM (#13274)

2025-11-14 12:51:11 -08:00

operations_strategy.py

…

operations.py

Register allgather/reducescatter buffers with symm memory (#12572)

2025-11-04 17:11:36 -08:00

server_args_config_parser.py

[Lint] Add python/sglang to ruff F401 checks and remove unused imports in files (#11685)

2025-10-17 16:49:46 -07:00

server_args.py

Piecewise Cuda Graph Support for gpt-oss model (#13045)

2025-11-15 09:28:00 +08:00

single_batch_overlap.py

[hotfix] Incorrect CombineOverlapArgs in SBO (#12230)

2025-10-28 01:23:06 -07:00

two_batch_overlap.py

Minor code cleanup / improvement for PREBUILT_EXTEND mode (#12948)

2025-11-10 11:59:54 +08:00

warmup.py

…