Skip to content
Draft
Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
28 commits
Select commit Hold shift + click to select a range
08a1435
[TRTLLM-17264][test] Characterize the VisualGen validation-error enve…
JunyiXu-nv Oct 28, 2026
2ca8b27
[TRTLLM-17264][refactor] Merge the two VisualGen validation-error ren…
JunyiXu-nv Oct 28, 2026
b291cca
[TRTLLM-17261][test] Add route, middleware-order and shim-identity sn…
JunyiXu-nv Oct 28, 2026
c9311ab
[TRTLLM-17261][test] Record the trtllm-serve route reference and test…
JunyiXu-nv Nov 1, 2026
81b0350
[TRTLLM-17261][test] Cover the attached-frontend route registration i…
JunyiXu-nv Nov 1, 2026
2eb184a
[None][test] Address review of the serve snapshot and layering tests
JunyiXu-nv Nov 2, 2026
5138e15
[TRTLLM-17261][test] Refresh the serve route reference and baselines …
JunyiXu-nv Nov 4, 2026
045a69e
[None][test] Make the route snapshot and the object.__new__ ratchet f…
JunyiXu-nv Nov 4, 2026
68344ed
[TRTLLM-17261][refactor] Move the msgpack route classes to tensorrt_l…
JunyiXu-nv Nov 2, 2026
9f4a48e
[TRTLLM-17261][refactor] Add the trtllm-serve router skeleton and ser…
JunyiXu-nv Nov 2, 2026
caf3a29
[TRTLLM-17261][infra] Multi-GPU triggers and type checking for the ne…
JunyiXu-nv Nov 2, 2026
4a6325d
[TRTLLM-17261][test] Register the system router in the multi-frontend…
JunyiXu-nv Nov 2, 2026
7e29509
[TRTLLM-17261][test] Match a request instead of reading route paths i…
JunyiXu-nv Nov 2, 2026
d3aacdd
[TRTLLM-17261][test] Classify endpoint-package imports by the importe…
JunyiXu-nv Nov 2, 2026
5bcd0a1
[TRTLLM-17264][refactor] Move the VisualGen protocol models into serv…
JunyiXu-nv Nov 3, 2026
9b8661f
[TRTLLM-17264][infra] Give tensorrt_llm/serve/api/visual_gen to the V…
JunyiXu-nv Nov 3, 2026
e122006
[TRTLLM-17264][test] Gate the VisualGen protocol models and their laz…
JunyiXu-nv Nov 3, 2026
18a8b2f
[TRTLLM-17264][test] Harden the VisualGen api-stability gate and pin …
JunyiXu-nv Nov 3, 2026
51fa901
Merge the VisualGen validation-renderer branch (#21408) into the Visu…
JunyiXu-nv Nov 4, 2026
1842630
[TRTLLM-17264][refactor] Move the VisualGen handlers and modules into…
JunyiXu-nv Nov 3, 2026
0de3561
[TRTLLM-17264][refactor] Wire the VisualGen serving classes and route…
JunyiXu-nv Nov 3, 2026
4401fd1
[TRTLLM-17264][test] Pin the VisualGen serving classes, router and me…
JunyiXu-nv Nov 3, 2026
5c7b7ed
[TRTLLM-17264][chore] Name the VisualGen serving package in serve/AGE…
JunyiXu-nv Nov 3, 2026
ece037b
[TRTLLM-17264][refactor] Keep the moved video module faithful and rew…
JunyiXu-nv Nov 3, 2026
7ef5426
[TRTLLM-17264][test] Move the VisualGen startup and shutdown test und…
JunyiXu-nv Nov 3, 2026
d326f00
[TRTLLM-17264][test] Guard the import isolation of the VisualGen serv…
JunyiXu-nv Nov 3, 2026
f937aa9
[TRTLLM-17264][infra] Let the CI test selector claim tensorrt_llm/ser…
JunyiXu-nv Nov 3, 2026
2fd70e2
[TRTLLM-17264][test] Check that the VisualGen shim table holds interf…
JunyiXu-nv Nov 4, 2026
File filter

Filter by extension

Filter by extension


Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
1 change: 1 addition & 0 deletions .github/CODEOWNERS
Original file line number Diff line number Diff line change
Expand Up @@ -371,6 +371,7 @@
/cpp/tensorrt_llm/thop/*lysses* @NVIDIA/trt-llm-torch-visual-gen-devs
/cpp/tensorrt_llm/kernels/ulysses* @NVIDIA/trt-llm-torch-visual-gen-devs
/tensorrt_llm/media @NVIDIA/trt-llm-torch-visual-gen-devs
/tensorrt_llm/serve/api/visual_gen @NVIDIA/trt-llm-torch-visual-gen-devs
/tensorrt_llm/visual_gen @NVIDIA/trt-llm-torch-visual-gen-devs @NVIDIA/trt-llm-runtime-devs
/tests/integration/defs/examples/visual_gen @NVIDIA/trt-llm-torch-visual-gen-devs
/tests/integration/defs/visual_gen @NVIDIA/trt-llm-torch-visual-gen-devs
Expand Down
17 changes: 15 additions & 2 deletions .pre-commit-config.yaml
Original file line number Diff line number Diff line change
Expand Up @@ -417,6 +417,12 @@ common-files: &common_files |
tensorrt_llm/scaffolding/worker.py |
tensorrt_llm/scheduling_params.py |
tensorrt_llm/serve/__init__.py |
tensorrt_llm/serve/_base/msgspec_route.py |
tensorrt_llm/serve/_base/serving.py |
tensorrt_llm/serve/api/visual_gen/media.py |
tensorrt_llm/serve/api/visual_gen/protocol.py |
tensorrt_llm/serve/api/visual_gen/serving_base.py |
tensorrt_llm/serve/api/visual_gen/serving_images.py |
tensorrt_llm/serve/chat_utils.py |
tensorrt_llm/serve/cluster_storage.py |
tensorrt_llm/serve/disagg_auto_scaling.py |
Expand Down Expand Up @@ -1144,6 +1150,12 @@ legacy-files: &legacy_files |
tensorrt_llm/scaffolding/worker.py |
tensorrt_llm/scheduling_params.py |
tensorrt_llm/serve/__init__.py |
tensorrt_llm/serve/_base/msgspec_route.py |
tensorrt_llm/serve/_base/serving.py |
tensorrt_llm/serve/api/visual_gen/media.py |
tensorrt_llm/serve/api/visual_gen/protocol.py |
tensorrt_llm/serve/api/visual_gen/serving_base.py |
tensorrt_llm/serve/api/visual_gen/serving_images.py |
tensorrt_llm/serve/chat_utils.py |
tensorrt_llm/serve/cluster_storage.py |
tensorrt_llm/serve/disagg_auto_scaling.py |
Expand Down Expand Up @@ -1620,6 +1632,7 @@ repos:
# Run mypy with no explicit paths so it type-checks the set declared in
# pyproject.toml [tool.mypy] `files` (single source of truth). This
# `files:` pattern only decides when the hook triggers — any change under
# the sampler source/test dirs.
# the sampler source/test dirs or the serving frontends. A PR that adds
# a typed module elsewhere extends this pattern in the same PR.
pass_filenames: false
files: ^(tensorrt_llm/_torch/pyexecutor/sampler/|tests/unittest/_torch/sampler/)
files: ^(tensorrt_llm/_torch/pyexecutor/sampler/|tests/unittest/_torch/sampler/|tensorrt_llm/serve/|tensorrt_llm/commands/serve_cli/)
4 changes: 4 additions & 0 deletions jenkins/L0_MergeRequest.groovy
Original file line number Diff line number Diff line change
Expand Up @@ -1948,6 +1948,10 @@ def getMultiGpuFileChanged(pipeline, testFilter, globalVars)
"tensorrt_llm/models/llama/",
"tensorrt_llm/parameter.py",
"tensorrt_llm/serve/__init__.py",
// _MsgspecRoute decodes the disaggregated frontend's msgpack bodies.
"tensorrt_llm/serve/_base/",
// /health is the disaggregated readiness probe.
"tensorrt_llm/serve/api/ops/",
"tensorrt_llm/serve/cluster_storage.py",
"tensorrt_llm/serve/disagg_auto_scaling.py",
"tensorrt_llm/serve/extensions/",
Expand Down
7 changes: 4 additions & 3 deletions jenkins/scripts/cbts/README.md
Original file line number Diff line number Diff line change
Expand Up @@ -41,14 +41,15 @@ filter chain.

## Rules

Nine rules, registered in `cbts/command/main.py::RULE_CLASSES`:
Eleven rules, registered in `cbts/command/main.py::RULE_CLASSES`:

| Rule | Scope | Files |
|---|---|---|
| `PythonFormattingRule` | `noop` | Python files whose patch is a verified pure reformat: identical complete ASTs (docstrings and type comments included) and unchanged comments |
| `WaivesRule` | `waiveonly` | `tests/integration/test_lists/waives.txt` |
| `TestsDefRule` | `testdefonly` | `tests/**/*` (.py via AST; data files via dir walk-up) |
| `TestListRule` | `testlistonly` | `tests/integration/test_lists/test-db/*.yml` |
| `VisualGenRule` | `visualgenonly` | VG source and the exact VG serve schema; public VG/media changes also keep full CPU and shared serving tests, package sanity, and performance sanity; missing required validation forces fallback |
| `VisualGenRule` | `visualgenonly` | VG source (`examples/visual_gen/**`, `scripts/visualgen_eval/**`, `tensorrt_llm/_torch/visual_gen/**`, `tensorrt_llm/media/**`, `tensorrt_llm/serve/api/visual_gen/**`, `tensorrt_llm/visual_gen/**`; excl. docs) and the exact VG serve schema; public VG/media changes also keep full CPU and shared serving tests, package sanity, and performance sanity; missing required validation forces fallback |
| `SpecDecRule` | `specdeconly` | `tensorrt_llm/_torch/speculative/**`, `tensorrt_llm/models/{eagle,medusa,redrafter}/**`, `examples/{eagle,medusa,redrafter,draft_target_model,ngram}/**`, `examples/llm-api/llm_speculative_decoding.py` (excl. docs; other suffixes incl. images kept as potential test fixtures) |
| `ModelingV2Rule` | `modelingv2only` | `tensorrt_llm/_torch/_experimental/modeling_v2/**`, `cpp/tensorrt_llm/modeling_v2/**`, `cpp/cmake/modules/modeling_v2.cmake` (excl. docs); per model node for changes under `models/` |
| `AgentFlowRule` | `agentflowonly` | `agent-flow/**` (excl. docs and out-of-scope metadata) |
Expand All @@ -65,7 +66,7 @@ See `cbts/rules/README.md` for per-rule logic.
| `waiveonly` | `WaivesRule` fired solo: PR only edits `waives.txt`. |
| `testdefonly` | `TestsDefRule` fired solo: PR only edits files under `tests/**/*`. |
| `testlistonly` | `TestListRule` fired solo: PR only adds entries under `tests/integration/test_lists/test-db/*.yml`. |
| `visualgenonly` | `VisualGenRule` fired solo. Internal VG changes select VG test entries; schema changes also retain schema validation. Public VG/media changes additionally retain the complete CPU suite, shared serving integration tests, package sanity, and performance sanity. Missing required validation forces `null` fallback. |
| `visualgenonly` | `VisualGenRule` fired solo. Internal VG changes (`examples/visual_gen/**`, `scripts/visualgen_eval/**`, `tensorrt_llm/_torch/visual_gen/**`, `tensorrt_llm/serve/api/visual_gen/**`) select VG test entries; schema changes also retain schema validation. Public VG/media changes additionally retain the complete CPU suite, shared serving integration tests, package sanity, and performance sanity. Missing required validation forces `null` fallback. The serving package counts as internal because nothing outside it imports it (`tests/unittest/llmapi/apps/visual_gen/test_serve_visual_gen_isolation.py`). |
| `specdeconly` | `SpecDecRule` fired solo: PR only touches speculative-decoding source paths (`tensorrt_llm/_torch/speculative/**`, `tensorrt_llm/models/{eagle,medusa,redrafter}/**`, `examples/{eagle,medusa,redrafter,draft_target_model,ngram}/**`, `examples/llm-api/llm_speculative_decoding.py`; excl. docs). Narrows to blocks containing spec-dec test entries (eagle / medusa / redrafter / ngram / draft-target-model / MTP). |
| `modelingv2only` | `ModelingV2Rule` fired solo: PR only touches modeling_v2 code -- the subtree (`tensorrt_llm/_torch/_experimental/modeling_v2/**`, each family's native code in its `csrc/` included: one torch-op library per model family) or what the family libraries share (`cpp/tensorrt_llm/modeling_v2/**`, `cpp/cmake/modules/modeling_v2.cmake`); excl. docs, which are a fifth of the Python subtree — every catalog entry carries a contract document. A change under `models/` selects its model node's unit tests and target accuracy gates, plus whole-tree checks; shared code or an unreadable routing tree selects every modeling_v2 entry. No outward-facing fallback is needed: nothing beyond the resolver module is imported, and no family library is loaded, unless `LLM(modeling_v2=...)` is on; the CPU checks that keep this true run in every scope. The callers outside the subtree (`_torch/pyexecutor/model_loader.py`, `_torch/models/modeling_auto.py`) are left unclaimed, so touching them falls back to baseline. |
| `agentflowonly` | `AgentFlowRule` fired solo: PR only touches `agent-flow/**` source or test files (excl. docs). Runs `CPU-AgentFlow-UnitTest`. |
Expand Down
19 changes: 16 additions & 3 deletions jenkins/scripts/cbts/cbts/rules/README.md
Original file line number Diff line number Diff line change
Expand Up @@ -15,7 +15,7 @@ for the overall CBTS architecture.
| `waives.py` | `WaivesRule` | `waiveonly` | `tests/integration/test_lists/waives.txt` |
| `tests_def.py` | `TestsDefRule` | `testdefonly` | `tests/**/*` (any file under tests/) |
| `test_list.py` | `TestListRule` | `testlistonly` | `tests/integration/test_lists/test-db/*.yml` |
| `visual_gen.py` | `VisualGenRule` | `visualgenonly` | `examples/visual_gen/**`, `scripts/visualgen_eval/**`, `tensorrt_llm/_torch/visual_gen/**`, `tensorrt_llm/media/**`, `tensorrt_llm/visual_gen/**` (each excl. docs), and the exact VisualGen serve schema |
| `visual_gen.py` | `VisualGenRule` | `visualgenonly` | `examples/visual_gen/**`, `scripts/visualgen_eval/**`, `tensorrt_llm/_torch/visual_gen/**`, `tensorrt_llm/media/**`, `tensorrt_llm/serve/api/visual_gen/**`, `tensorrt_llm/visual_gen/**` (each excl. docs), and the exact VisualGen serve schema |
| `spec_dec.py` | `SpecDecRule` | `specdeconly` | `tensorrt_llm/_torch/speculative/**`, `tensorrt_llm/models/{eagle,medusa,redrafter}/**`, `examples/{eagle,medusa,redrafter,draft_target_model,ngram}/**`, `examples/llm-api/llm_speculative_decoding.py` (each excl. docs) |
| `modeling_v2.py` | `ModelingV2Rule` | `modelingv2only` | `tensorrt_llm/_torch/_experimental/modeling_v2/**`, `cpp/tensorrt_llm/modeling_v2/**`, `cpp/cmake/modules/modeling_v2.cmake` (excl. docs); per model node for changes under `models/` |
| `agent_flow.py` | `AgentFlowRule` | `agentflowonly` | `agent-flow/**` (excl. docs and out-of-scope metadata) → the single `CPU-AgentFlow-UnitTest` stage; not test-db-driven |
Expand Down Expand Up @@ -217,7 +217,8 @@ Outcomes:

Path-only rule. Claims source files under `examples/visual_gen/`,
`scripts/visualgen_eval/`, `tensorrt_llm/_torch/visual_gen/`,
`tensorrt_llm/media/`, and `tensorrt_llm/visual_gen/` (excluding documentation,
`tensorrt_llm/media/`, `tensorrt_llm/serve/api/visual_gen/`, and
`tensorrt_llm/visual_gen/` (excluding documentation,
which `DocsRule` handles). It also claims the exact generated schema
`tensorrt_llm/schemas/trtllm-serve-visual-gen-config.schema.json`.
Fused-DiT paths stay unclaimed:
Expand Down Expand Up @@ -258,11 +259,23 @@ pre-merge multi-GPU stages cannot satisfy this requirement. Missing or filtered
required validation keeps `scope=None`. Public changes therefore get a wider
test set than internal VG changes without requiring all unrelated GPU tests.

Internal serving package: `tensorrt_llm/serve/api/visual_gen/`, the VisualGen
HTTP serving package, is internal rather than shared. No module outside it
imports it except the trtllm-serve composition root for the VISUAL_GEN role,
the routing table's module-path string, the lazy re-export in
`tensorrt_llm.serve.openai_protocol` and the VisualGen benchmark client, so a
change under it narrows like the other internal paths.
`tests/unittest/llmapi/apps/visual_gen/test_serve_visual_gen_isolation.py`
asserts this (the static import graph, a runtime check per serving role and
CLI entry point, and the placement of VisualGen tests under `visual_gen/`
paths); if it fails, the prefix can no longer be treated as internal.

Outcomes:

- No VG source files in the diff → rule returns `None`.
- VG source touched, all internal (`examples/visual_gen/**`,
`scripts/visualgen_eval/**`, or `tensorrt_llm/_torch/visual_gen/**`)
`scripts/visualgen_eval/**`, `tensorrt_llm/_torch/visual_gen/**`, or
`tensorrt_llm/serve/api/visual_gen/**`)
→ `scope=visualgenonly`;
sanity off (VG changes don't affect wheel sanity); perfsanity on iff
a matched block lives in `l0_perf` or `*perf_sanity*`.
Expand Down
17 changes: 15 additions & 2 deletions jenkins/scripts/cbts/cbts/rules/visual_gen.py
Original file line number Diff line number Diff line change
Expand Up @@ -15,8 +15,9 @@

VisualGen is a self-contained subsystem under
`tensorrt_llm/_torch/visual_gen/` (Flux / LTX-2 / Wan diffusion models)
plus the public API surface at `tensorrt_llm/visual_gen/` and the media
helpers at `tensorrt_llm/media/`.
plus the public API surface at `tensorrt_llm/visual_gen/`, the media
helpers at `tensorrt_llm/media/` and the HTTP serving package at
`tensorrt_llm/serve/api/visual_gen/`.

Block selection — entry-pattern based only:
VisualGen does NOT have its own `condition.terms.backend`; VG test
Expand All @@ -29,6 +30,15 @@
Their changes keep the complete CPU suite, shared serving integration tests,
and package/performance sanity checks in addition to VG entries. Missing API,
schema, lazy-import, or serving validation keeps the baseline fallback.

Internal serving package:
`tensorrt_llm/serve/api/visual_gen/` is internal, not shared: no module
outside it imports it except the trtllm-serve composition root for the
VISUAL_GEN role, the routing table's module-path string, the lazy
re-export in `tensorrt_llm.serve.openai_protocol` and the VisualGen
benchmark client, so a change under it narrows like the other internal
VG paths. `tests/unittest/llmapi/apps/visual_gen/test_serve_visual_gen_isolation.py`
asserts this; if it fails, claiming this prefix is no longer safe.
"""

from __future__ import annotations
Expand Down Expand Up @@ -57,13 +67,16 @@
"scripts/visualgen_eval/",
"tensorrt_llm/_torch/visual_gen/",
"tensorrt_llm/media/",
"tensorrt_llm/serve/api/visual_gen/",
"tensorrt_llm/visual_gen/",
)

_VG_SCHEMA_PATH = "tensorrt_llm/schemas/trtllm-serve-visual-gen-config.schema.json"
_SCHEMA_TEST_TARGET = "unittest/llmapi/test_trtllm_serve_config_schemas.py"

# These packages expose shared protocol and media dependencies to serving.
# `tensorrt_llm/serve/api/visual_gen/` is not listed: it is internal (see the
# module docstring).
_VG_OUTWARD_PREFIXES: tuple[str, ...] = (
"tensorrt_llm/media/",
"tensorrt_llm/visual_gen/",
Expand Down
6 changes: 6 additions & 0 deletions legacy-files.txt
Original file line number Diff line number Diff line change
Expand Up @@ -409,6 +409,12 @@ tensorrt_llm/scaffolding/task_collection.py
tensorrt_llm/scaffolding/worker.py
tensorrt_llm/scheduling_params.py
tensorrt_llm/serve/__init__.py
tensorrt_llm/serve/_base/msgspec_route.py
tensorrt_llm/serve/_base/serving.py
tensorrt_llm/serve/api/visual_gen/media.py
tensorrt_llm/serve/api/visual_gen/protocol.py
tensorrt_llm/serve/api/visual_gen/serving_base.py
tensorrt_llm/serve/api/visual_gen/serving_images.py
tensorrt_llm/serve/chat_utils.py
tensorrt_llm/serve/cluster_storage.py
tensorrt_llm/serve/disagg_auto_scaling.py
Expand Down
11 changes: 11 additions & 0 deletions pyproject.toml
Original file line number Diff line number Diff line change
Expand Up @@ -474,6 +474,12 @@ exclude = [
"tensorrt_llm/scaffolding/worker.py",
"tensorrt_llm/scheduling_params.py",
"tensorrt_llm/serve/__init__.py",
"tensorrt_llm/serve/_base/msgspec_route.py",
"tensorrt_llm/serve/_base/serving.py",
"tensorrt_llm/serve/api/visual_gen/media.py",
"tensorrt_llm/serve/api/visual_gen/protocol.py",
"tensorrt_llm/serve/api/visual_gen/serving_base.py",
"tensorrt_llm/serve/api/visual_gen/serving_images.py",
"tensorrt_llm/serve/chat_utils.py",
"tensorrt_llm/serve/cluster_storage.py",
"tensorrt_llm/serve/disagg_auto_scaling.py",
Expand Down Expand Up @@ -862,6 +868,11 @@ files = [
"tensorrt_llm/_torch/pyexecutor/sampler/token_ban.py",
"tensorrt_llm/_torch/pyexecutor/sampler/ops/flashinfer.py",
"tensorrt_llm/_torch/pyexecutor/sampler/ops/vanilla.py",
"tensorrt_llm/serve/_base/routing.py",
"tensorrt_llm/serve/api/ops/system/api_router.py",
"tensorrt_llm/serve/api/ops/system/serving.py",
"tensorrt_llm/serve/api/visual_gen/api_router.py",
"tensorrt_llm/serve/api/visual_gen/errors.py",
"tests/unittest/_torch/sampler/test_torch_sampler.py",
"tests/unittest/_torch/sampler/test_token_ban.py",
"tests/unittest/_torch/sampler/test_beam_search.py",
Expand Down
6 changes: 6 additions & 0 deletions ruff-legacy.toml
Original file line number Diff line number Diff line change
Expand Up @@ -426,6 +426,12 @@ include = [
"tensorrt_llm/scaffolding/worker.py",
"tensorrt_llm/scheduling_params.py",
"tensorrt_llm/serve/__init__.py",
"tensorrt_llm/serve/_base/msgspec_route.py",
"tensorrt_llm/serve/_base/serving.py",
"tensorrt_llm/serve/api/visual_gen/media.py",
"tensorrt_llm/serve/api/visual_gen/protocol.py",
"tensorrt_llm/serve/api/visual_gen/serving_base.py",
"tensorrt_llm/serve/api/visual_gen/serving_images.py",
"tensorrt_llm/serve/chat_utils.py",
"tensorrt_llm/serve/cluster_storage.py",
"tensorrt_llm/serve/disagg_auto_scaling.py",
Expand Down
7 changes: 4 additions & 3 deletions tensorrt_llm/bench/benchmark/visual_gen.py
Original file line number Diff line number Diff line change
Expand Up @@ -336,9 +336,10 @@ def _run_benchmark(

Each per-request ``latency`` window covers ``generate()`` plus a
``save()`` to a per-run temp directory. The save mirrors what the
serving path (``openai_video_routes.py`` / ``openai_server.py``) does
inside its own ``latency`` window: encode and write to a persisted
location. The directory is cleaned up at the end of the run.
serving path (``serve/api/visual_gen/serving_videos.py`` /
``serving_images.py``) does inside its own ``latency`` window: encode
and write to a persisted location. The directory is cleaned up at the
end of the run.
"""
import asyncio

Expand Down
6 changes: 6 additions & 0 deletions tensorrt_llm/commands/AGENTS.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,6 @@
<!--
SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
SPDX-License-Identifier: Apache-2.0
-->

Rules for the trtllm-serve CLI (`commands/serve.py`) are in `tensorrt_llm/serve/AGENTS.md`.
2 changes: 2 additions & 0 deletions tensorrt_llm/commands/CLAUDE.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,2 @@
# In ./CLAUDE.md
@AGENTS.md
Loading