Files

leavelet e3cfba8441 CP HiCache trace: key KV-value hashes by host-slot fingerprint + zero-KV flag

The first KV-value pass couldn't compare (reload hashes keyed by the merged
load op's node id, backup hashes per node -> no join). Fix: emit the host-slot
range fingerprint at both backup and reload (host slots are the stable identity
across node id / splits / merge), and the analyzer matches on
(rank, layer, host_fp). Also add a nonzero-byte count (nz): real KV is never
all-zero, so an nz==0 slice is uninitialized/zero KV -- a direct, self-contained
corruption flag that needs no cross-stage matching. This targets the observed
'0|0|0...' garbage on large-cached reloads (store never filled the backed-up
pages, or the round-trip lost them). Gated at SGLANG_CP_HICACHE_KV_TRACE=2.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

2026-06-17 02:29:00 +00:00

benchmark

[Benchmark] use flashinfer bench_gpu_time instead of triton do_bench (#20305 )

2026-03-12 04:04:30 +00:00

cli

Simplify server startup output (#20885 )

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

eval

Fix none-comparison (E711) warnings (#19745 )

2026-03-06 16:15:21 -08:00

jit_kernel

ci: unify PR test suite naming (#21187 )

2026-03-23 00:18:45 -07:00

lang

[CPU] Add frontend support for Gemma (#12590 )

2026-03-18 23:02:26 -07:00

multimodal_gen

[diffusion] CI: auto-skip diffusion tests when required pipeline class is missing from diffusers (#21139 )

2026-03-23 12:15:21 +08:00

srt

CP HiCache trace: key KV-value hashes by host-slot fingerprint + zero-KV flag

2026-06-17 02:29:00 +00:00

test

ci: remove IS_BLACKWELL env var; auto-detect Blackwell (#21118 )

2026-03-22 23:44:48 -07:00

__init__.py

[diffusion][llm] macOS support (#19549 )

2026-03-10 13:11:07 -07:00

_mps_stub.py

[diffusion][llm] macOS support (#19549 )

2026-03-10 13:11:07 -07:00

_triton_stub.py

[diffusion][llm] macOS support (#19549 )

2026-03-10 13:11:07 -07:00

bench_offline_throughput.py

Add Ray actor support for scheduler process management (DP=1) (#17684 )

2026-03-05 13:21:23 -08:00

bench_one_batch_server.py

Upload nightly test metrics to GH artifacts (#17696 )

2026-01-25 14:35:14 -08:00

bench_one_batch.py

Fix correctness test issue for bench_one_batch (#20650 )

2026-03-15 20:05:36 -07:00

bench_serving.py

[Benchmark] Add sglang-embedding backend to bench_serving (#20017 )

2026-03-11 13:13:16 -07:00

check_env.py

Upgrade transformers==5.3.0 (#17784 )

2026-03-18 13:50:43 -07:00

compile_deep_gemm.py

…

global_config.py

…

kernel_api_logging.py

Add SGLang CUDA crash API logging inspired by FlashInfer (#20910 )

2026-03-22 16:39:40 +08:00

launch_server.py

Simplify server startup output (#20885 )

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

profiler.py

Add XPU profiler activity support in benchmark code (#12981 )

2026-03-04 23:22:56 -08:00

README.md

Update python/sglang/README.md (#18045 )

2026-01-31 13:20:41 -08:00

utils.py

[Utils] Move network/socket utilities from common.py to network.py (#20646 )

2026-03-15 20:35:24 -07:00

version.py

…

README.md

Code Structure

eval: The evaluation utilities.
lang: The frontend language.
multimodal_gen: Inference framework for accelerated image/video generation.
srt: The backend engine for running local models. (SRT = SGLang Runtime).
test: The test utilities.
api.py: The public APIs.
bench_offline_throughput.py: Benchmark the performance in the offline mode.
bench_one_batch.py: Benchmark the latency of running a single static batch without a server.
bench_one_batch_server.py: Benchmark the latency of running a single batch with a server.
bench_serving.py: Benchmark online serving with dynamic requests.
check_env.py: Check the environment variables and dependencies.
global_config.py: The global configs and constants.
launch_server.py: The entry point for launching a local server.
profiler.py: The profiling entry point to send profile requests.
utils.py: Common utilities.
version.py: Version info.