Commit Graph
103 Commits
Author SHA1 Message Date
Mohammad Miadh AngkadandMohammad Angkad 7656ea0e74 [CI] python: Move the remaining CUDA build paths off Python 3.10 (#44393)
Co-authored-by: Mohammad Angkad <mohammad.angkad@radixark.ai>
2026-10-19 14:29:00 -07:00
Bowen Wang 3671503e44 [Feat] bench: report goodput and SLO attainment in bench_serving (#43967) 2026-10-18 13:31:48 +08:00
Jiaqi LiandZijie Xia 5569b14d26 fix(benchmark): request token IDs when counting OpenAI prompts (#34511)
Co-authored-by: Zijie Xia <zijie.xia@radixark.ai>
2026-10-13 14:31:45 -07:00
Rain Jiang e98c6d7b25 Rainj me/rust bench (#43950) 2026-10-13 11:07:42 -07:00
Chunan Zeng 3c74f68a9c [Fix] Initialize BF16 GEMM dispatch in the one-batch benchmark (#43702) 2026-10-12 02:53:02 +00:00
Liangsheng Yin 1066decce6 [Scheduler] Replace extend_range with extend_end derived from prefix_len (#42822) 2026-10-07 16:07:16 -07:00
Liangsheng Yin 5349d3e9d2 [Scheduler] Derive prefix KV indices from last_node at allocation; drop Req.prefix_indices (#42825) 2026-10-07 16:03:09 -07:00
Liangsheng Yin e9240b3243 [Scheduler] Track prefill progress as prefix_len (#42824) 2026-10-07 15:59:06 -07:00
Liangsheng Yin 9a09b48da7 [Fix] Add maybe_hand_to_session to the one-batch benchmark TreeCacheNamespace (#42723) 2026-10-05 23:09:41 -07:00
Mohammad Miadh AngkadandMohammad Angkad 0e6d7eba5f [Fix] Count attention-DP ranks in one-batch bench running cap (#42331)
Co-authored-by: Mohammad Angkad <mohammad.angkad@radixark.ai>
2026-10-03 21:04:27 -07:00
Liangsheng Yin 122b59152c [mem_cache] Replace is_chunk_cache / is_tree_cache with supports_prefix_sharing (#42362) 2026-10-03 02:07:55 -07:00
bf2d686a82 Add triton autotune on the Mamba2 SSD kernels (#39130)
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
Co-authored-by: Po-Han Huang (NVIDIA) <53919306+nvpohanh@users.noreply.github.com>
2026-10-01 14:25:37 -07:00
1ba84a4b59 [bench] Take each request's prompt length from the server (#39889)
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
Co-authored-by: Xiaoyu Zhang <1182563586@qq.com>
2026-09-27 23:57:18 +08:00
Cheng Wan 79fec592a8 [Refactor] Read parallel placement in consumers (#40638) 2026-09-23 14:57:25 -07:00
Liangsheng Yin 78980a3b0b [misc] Remove deprecated endpoints, env vars and aliases past two releases (#40795) 2026-09-22 18:14:45 -07:00
Cheng Wan 91c329cc86 Take the model config out of the parallel group build, and finish retiring the parallel getters (#40707) 2026-09-22 10:55:21 -07:00
877a293d6d [Benchmark] Optionally clear HiCache storage between cases (#40659)
Co-authored-by: metamergebot <324680979+metamergebot@users.noreply.github.com>
Co-authored-by: Pengchao Wang <wpc@fb.com>
2026-09-21 21:31:21 -07:00
Cheng Wan acac4dd9d9 [Refactor] Clean up parallel runtime comments (#40632) 2026-09-21 14:32:22 -07:00
Cheng Wan bccf691b22 Bringing the parallel runtime up becomes a phase, not a side effect (#40345) 2026-09-21 12:29:50 -07:00
Cheng Wan 970e946e4f Retire the per-runner parallel record (#40343) 2026-09-21 12:26:40 -07:00
Xiaoyu Zhang cb22f2451e [Cleanup] Deduplicate kernel tests, diffusion fixtures and benchmark helpers (#40265) 2026-09-19 19:45:38 +08:00
Cheng Wan 3a5f52e144 Record a process's placement at publish, not at group build (#40071) 2026-09-18 23:52:37 -07:00
BingjiaWang 7782a2a1c8 [Benchmark] Limit warmup concurrency in serving benchmark (#39398) 2026-09-16 22:58:10 -07:00
84d7604b7e [XPU] weekly simple model enablement 2026/09/14 (#39439)
Co-authored-by: Juan Muneton <102537701+jmunetong@users.noreply.github.com>
Co-authored-by: YangKai0616 <kai.yang@intel.com>
Co-authored-by: devan-carlin <devan-carlin@users.noreply.github.com>
Co-authored-by: Ashwini Rathi <arathi@habana.ai>
Co-authored-by: Ranjan Debnath <ranjan.debnath@intel.com>
Co-authored-by: Juan Muneton <juan.muneton.gallego@intel.com>
Co-authored-by: Amrutha M <amrutha.m@intel.com>
2026-09-17 10:43:04 +08:00
0daa040e48 Support LongBench v2 in one-batch server benchmarks (#39874)
Co-authored-by: yhzhuang <yhzhuang@fb.com>
Co-authored-by: YJYJLee <yejinlee@meta.com>
2026-09-16 19:16:03 -07:00
Dayananda VandClaude Opus 5 100e1cd0d9 [Fix] Spawn, don't fork, the benchmark server process (#34712)
Co-authored-by: Claude Opus 5 <noreply@anthropic.com>
2026-09-16 12:44:09 +08:00
Xingyu Liu 8874c51a96 [Benchmark] Add an opt-out for the token-capacity check (#39284) 2026-09-14 17:17:12 -07:00
Cheng Wan f1a512c51c [Config] msgspec.Struct for the config tier (#38753) 2026-09-09 19:41:19 -07:00
Cheng Wan 53dc77ff4e [Config] One writer for the declaration stash; no exception to the write seal (#38752) 2026-09-09 19:22:06 -07:00
Cheng Wan aaf9a95763 [Config] Round 6.5: a namespace declares what it derives, next to what it derives it from (#38113)
Fifth of five; stacked on #38049. The split gave every namespace a file, but
only for the half an operator types. This is the other half.

## The parallel quotients are declared, not written out

`attn_tp_size` and its five siblings were sixty lines of near-identical
properties in the runtime context, a file away from the leaves they are
quotients of, so reading `parallel.py` told you what you could set and nothing
about what that decides.

They are declared in `Parallel` now, in the same class as those leaves. They
carry no annotation, so they are not dataclass fields and
`collect_input_fields` never puts them on the record -- the same mechanism that
already keeps `_NS_PATH` off it. That is the right exclusion: a quotient has no
operator input to preserve, and the record is what crosses a process boundary,
where a stamped width is one an elastic scale-up will not refresh.

## A quotient is a value in the bag, like every other derived one

`_derived_width` answered from a stamp or, failing that, a live process group.
The group read could never disagree with the stamp:

- `initialize_model_parallel` stamps all six as its last statement,
  unconditionally;
- an elastic scale-up restamps `attn_dp_size` through
  `update_dp_attention_post_scale` -- the comment claiming it does *not* was
  wrong;
- no hardware backend builds groups of its own;
- `multimodal_gen`, which has its own `initialize_model_parallel` and does not
  stamp, never reads a quotient.

So a built group was always already stamped, and the group read goes -- and with
it the last reason for a quotient to be resolved on every read.

Every input to `derive_parallel_widths` is a record field. `dcp_enabled` is
`decode_context_parallel_size > 1`, not a fact about a built group; it was
spelled `_DCP is not None`, which is a longer way to say the same thing. So the
six are fixed once the configuration is fixed -- the same test every other
`Derived(fn=...)` in this PR passes. They are declared the same way and computed
the same way: once, at publish, into ordinary bag leaves.

What remains is override -> stamp -> published leaf. The stamp stays above the
leaf because an elastic scale-up restamps `attn_dp_size`; the override stays on
top because that is how a test names a width.

## One answer for the config-derived predicates

`enable_mamba_extra_buffer` and its lazy variant, `is_ep_joiner`,
`is_ep_scale_joiner`, `is_startup_weight_load_overlap`: each existed as a
`ServerArgs` member for the resolution pipeline and, for most of them, again as
a `runtime_context` function for readers after publish. Three places to keep
saying the same thing.

A `Derived(fn=...)` is a pure function of the published configuration, so
`publish` computes it once and stores it as an ordinary bag leaf -- a plain
attribute load, which is what a read inside compiled model code needs. The
function is handed the whole resolved config rather than the bag it lands in,
because a derivation is free to span namespaces and the mamba one does: it
reads `memory.disable_radix_cache` alongside its own `exec.mamba` strategy,
which is why it could never have been a method on either bag.

The pre-publish helpers stay -- resolution needs the predicate before there is
a bag to read -- and three readers keep them, because they run before their own
process publishes: `initialize_dp_attention`, which the weight-cache daemon
calls while building its groups thirty lines before its `publish`, and
`PortArgs.init_new`, a factory handed the record that already reads eighteen
other fields off it.

## Notes for a reviewer

**Overriding a leaf does not move its quotient.** `override(tp_size=2)` leaves
`attn_tp_size` where the published config put it, because nothing is recomputed
on read. A test states a topology by publishing a config -- which is what a
real process does -- or by naming the width it wants, `override(attn_tp_size=2)`.
Six tests say it that way now. This is the price of having one answer computed
once, and it is the same price every other derived value in the config already
carries.

A caller that reads a quotient without publishing or overriding now gets an
explicit error naming the field, instead of a default that an uninitialised
group happened to supply. One fixture was in that state --
`TestMlaWriteDoorsUnderDcp` built a bare pool and asked whether DCP was on --
and it publishes a config now, which is what the process it stands in for
does.

Eighteen sites read these predicates without calling them. That is correct --
they are properties -- but it is worth saying they were checked, because a
census that assumes otherwise reports eighteen always-true conditions.

## The skill that documents this subsystem is updated with it

`.claude/rules/modify-component-must-read.md` points at
`.claude/skills/sglang-runtime-context/SKILL.md` before anyone touches these
files, so a stale sentence there is a wrong instruction rather than a stale
note. Four of its load-bearing statements stopped being true across this series
and are corrected here: `NS(...)` is no longer how a field states its namespace
(the declaring class is); the DCP degrade rule is gone, because the quotients
are not live reads; `mamba_extra_buffer_enabled()` and the other predicate
functions it named as the shape to copy no longer exist; and the
namespace-coverage ratchet is described in terms of the marker. The docstring of
`test_server_args_namespaces.py` said the same thing and is fixed too.

The consequence a test author actually trips over is stated there as well:
overriding a leaf no longer moves its quotient, so a topology is stated by
publishing a config or by naming the width.

## Verification

A full registered-unit sweep (648 files) against this stack's merge-base:
19 failures on both sides, the same 19 -- AMD `gfx950`, `modelopt`,
`cuda_vmm`, `weight_checker` and friends, none of them config. The narrower 139-file config sweep used earlier in this series
does not contain the files this change reaches -- `test_kv_index_translator`
never names `get_parallel()`, it constructs an object that does -- which is why
the baseline differential over everything is what is quoted here.
2026-09-06 21:44:24 -07:00
Alex NailsandAlison Shao 28262c20df [CI][RFC] Replace black-jupyter with ruff-format (#37210)
Co-authored-by: Alison Shao <a.shao@wustl.edu>
2026-09-02 19:46:08 -07:00
Oguz Ulgen f15748d965 [bench] Support real-traffic replay with early-stop-aware steady-state metrics in bench_one_batch_server (#37469) 2026-09-02 17:20:25 -07:00
Wes 2d9c64394f Fix reasoning metrics and add TPOT to bench_multiturn (#35443) 2026-09-02 11:28:47 +08:00
Liangsheng Yin 2530204502 [mem_cache] Make release, row-reuse asserts, and presence checks read the KV record (#37167) 2026-08-31 12:46:15 -07:00
Shuwen Wang 9e9d26a4af Fix Mooncake serving benchmark trace rows (#37201) 2026-08-31 11:43:59 -07:00
Cheng WanandClaude Opus 5 fd40a331bf config: a parallel size has one spelling; a patched scope declares its own (#36621)
Co-authored-by: Claude Opus 5 <noreply@anthropic.com>
2026-08-27 12:56:42 -07:00
Cheng WanandClaude Opus 5 ca1d7ed8e6 config: a parallel leaf with no live counterpart is read bare (#36620)
Co-authored-by: Claude Opus 5 <noreply@anthropic.com>
2026-08-27 12:56:11 -07:00
Cheng Wan 413df1f8db config: ServerArgs holds the raw input (#36255) 2026-08-26 05:14:05 -07:00
Cheng Wan 937af8538b config: the runtime readers take the published bags (#36254) 2026-08-26 05:08:25 -07:00
Cheng Wan ae5feb4b9c config: stop handing the record to code that does not read it (#36252) 2026-08-26 05:02:17 -07:00
Cheng Wan d7b144f64e config: publishing is the process entry's job (#36251) 2026-08-26 04:58:41 -07:00
Cheng Wan 8005df61d3 config: spell the parallel config tier at the call site (#36250) 2026-08-26 03:00:28 -07:00
BingjiaWang 3c481b9421 [Benchmark] Add optional steady-state window for serving metrics (#30918) 2026-08-24 14:49:20 -07:00
Cheng WanandClaude Opus 5 362c2ee849 config: borrowed-record reads follow the config bags (#35908)
Co-authored-by: Claude Opus 5 <noreply@anthropic.com>
2026-08-23 01:19:20 -07:00
Cheng WanandClaude Opus 5 64aa859da2 config: constructing a config no longer resolves it (#35907)
Co-authored-by: Claude Opus 5 <noreply@anthropic.com>
2026-08-23 01:18:53 -07:00
R0CKSTARandAlex Nails d90318b3e2 [MLX] Upgrade to Torch 2.13/MLX 0.32+ and redesign the Torch-MLX tensor bridge (#32984)
Co-authored-by: Alex Nails <alex.nails@radixark.ai>
2026-08-21 18:51:42 -07:00
6b94d39f13 [Model Loading] Overlap checkpoint staging with CUDA graph capture during startup (#32017)
Co-authored-by: Wenhui Zhu <wzhu59@asu.edu>
Co-authored-by: Alex Nails <alex.nails@radixark.ai>
2026-08-13 12:26:25 -07:00
fde9ad2531 [Feature] Add Muse Glimmer model support (#34262)
Co-authored-by: sglang-bot <232288953+sglang-bot@users.noreply.github.com>
Co-authored-by: Brayden Zhong <brayden.zhong@radixark.ai>
Co-authored-by: Jimmy Shong <69131491+Jiminator@users.noreply.github.com>
Co-authored-by: hnyls2002 <lsyincs@gmail.com>
Co-authored-by: Alex Nails <alex.nails@radixark.ai>
Co-authored-by: Liangsheng Yin <hnyls2002@gmail.com>
2026-08-11 15:41:52 -07:00
Khoa PhamandClaude Opus 5 ce1b9f88b6 config: derive the runner's DCP topology from its ParallelState (#34133)
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
2026-08-09 01:18:24 -07:00
Mohammad Miadh Angkad 8a1637a479 Fix serving benchmark post-warmup cache flush race (#33663) 2026-08-06 15:27:10 +00:00