[TRTLLM-13782][doc] Remove legacy TensorRT docs and add migration guide#15767
Merged
Wanli-Jiang merged 2 commits intoJul 2, 2026
Merged
Conversation
Contributor
📝 WalkthroughWalkthroughRemoves all legacy TensorRT backend documentation pages, adds a new migration guide ( ChangesTensorRT Backend Removal Documentation
Estimated code review effort🎯 2 (Simple) | ⏱️ ~10 minutes 🚥 Pre-merge checks | ✅ 4 | ❌ 1❌ Failed checks (1 warning)
✅ Passed checks (4 passed)
✨ Finishing Touches🧪 Generate unit tests (beta)
Comment |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
Contributor
There was a problem hiding this comment.
Actionable comments posted: 3
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In `@docs/source/blogs/Best_perf_practice_on_DeepSeek-R1_in_TensorRT-LLM.md`:
- Around line 417-423: The performance tuning guidance mentions for both
max_batch_size and max_num_tokens are plain text and do not point readers to the
actual tuning guide. Update the relevant references in this blog section so the
“performance tuning guidance” text links directly to the existing tuning-guide
documentation, using the same link target wherever the guidance is referenced in
this passage.
In `@docs/source/legacy/tensorrt-backend-removal.md`:
- Around line 52-54: The link in the legacy TensorRT backend removal docs points
to a drifting GitHub tree URL instead of the local LLM API examples page. Update
the reference in the documentation snippet near the PyTorch model usage note to
use the in-repo LLM API docs link, keeping the mention of the LLM Python API
examples and the example names like quickstart_advanced.py and
quickstart_multimodal.py intact.
In `@docs/source/torch/arch_overview.md`:
- Around line 20-21: Update the architecture overview wording for the PyExecutor
path so it describes the PyTorch backend implementation directly rather than
calling it an alternative backend; revise the text near PyExecutor/Executor to
say the PyTorch backend uses PyExecutor as its implementation and remove any
“alternative backend” phrasing, keeping the description aligned with the current
LLM backend flow.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: Path: .coderabbit.yaml
Review profile: CHILL
Plan: Enterprise
Run ID: 8191e2d0-93a9-40da-88db-7e96e00a4f6c
⛔ Files ignored due to path filters (2)
docs/source/features/images/8x_l20_L40S_node_architecture.pngis excluded by!**/*.pngdocs/source/legacy/advanced/images/disaggregated-service_usage.pngis excluded by!**/*.png
📒 Files selected for processing (52)
docs/source/blogs/Best_perf_practice_on_DeepSeek-R1_in_TensorRT-LLM.mddocs/source/blogs/XQA-kernel.mddocs/source/commands/trtllm-build.rstdocs/source/features/lowprecision-allreduce.mddocs/source/index.rstdocs/source/legacy/advanced/disaggregated-service.mddocs/source/legacy/advanced/executor.mddocs/source/legacy/advanced/expert-parallelism.mddocs/source/legacy/advanced/gpt-attention.mddocs/source/legacy/advanced/gpt-runtime.mddocs/source/legacy/advanced/graph-rewriting.mddocs/source/legacy/advanced/kv-cache-management.mddocs/source/legacy/advanced/kv-cache-reuse.mddocs/source/legacy/advanced/lora.mddocs/source/legacy/advanced/open-sourced-cutlass-kernels.mddocs/source/legacy/advanced/speculative-decoding.mddocs/source/legacy/advanced/weight-streaming.mddocs/source/legacy/architecture/add-model.mddocs/source/legacy/architecture/checkpoint.mddocs/source/legacy/architecture/core-concepts.mddocs/source/legacy/architecture/model-weights-loader.mddocs/source/legacy/architecture/workflow.mddocs/source/legacy/dev-on-cloud/build-image-to-dockerhub.mddocs/source/legacy/dev-on-cloud/dev-on-runpod.mddocs/source/legacy/key-features.mddocs/source/legacy/performance/perf-analysis.mddocs/source/legacy/performance/perf-benchmarking.mddocs/source/legacy/performance/performance-tuning-guide/benchmarking-default-performance.mddocs/source/legacy/performance/performance-tuning-guide/deciding-model-sharding-strategy.mddocs/source/legacy/performance/performance-tuning-guide/fp8-quantization.mddocs/source/legacy/performance/performance-tuning-guide/index.rstdocs/source/legacy/performance/performance-tuning-guide/introduction.mddocs/source/legacy/performance/performance-tuning-guide/tuning-max-batch-size-and-max-num-tokens.mddocs/source/legacy/performance/performance-tuning-guide/useful-build-time-flags.mddocs/source/legacy/performance/performance-tuning-guide/useful-runtime-flags.mddocs/source/legacy/python-api/tensorrt_llm.functional.rstdocs/source/legacy/python-api/tensorrt_llm.layers.rstdocs/source/legacy/python-api/tensorrt_llm.models.rstdocs/source/legacy/python-api/tensorrt_llm.plugin.rstdocs/source/legacy/python-api/tensorrt_llm.quantization.rstdocs/source/legacy/python-api/tensorrt_llm.runtime.rstdocs/source/legacy/reference/memory.mddocs/source/legacy/reference/multimodal-feature-support-matrix.mddocs/source/legacy/reference/precision.mddocs/source/legacy/reference/support-matrix.mddocs/source/legacy/reference/troubleshooting.mddocs/source/legacy/tensorrt-backend-removal.mddocs/source/legacy/tensorrt_quickstart.mddocs/source/legacy/torch.mddocs/source/release-notes.mddocs/source/torch/arch_overview.mddocs/source/torch/kv_cache_manager.md
💤 Files with no reviewable changes (44)
- docs/source/legacy/reference/precision.md
- docs/source/legacy/key-features.md
- docs/source/legacy/performance/performance-tuning-guide/index.rst
- docs/source/legacy/tensorrt_quickstart.md
- docs/source/legacy/advanced/expert-parallelism.md
- docs/source/legacy/python-api/tensorrt_llm.functional.rst
- docs/source/legacy/python-api/tensorrt_llm.runtime.rst
- docs/source/legacy/python-api/tensorrt_llm.plugin.rst
- docs/source/legacy/advanced/disaggregated-service.md
- docs/source/legacy/reference/support-matrix.md
- docs/source/legacy/performance/performance-tuning-guide/introduction.md
- docs/source/legacy/performance/perf-benchmarking.md
- docs/source/legacy/advanced/weight-streaming.md
- docs/source/legacy/performance/performance-tuning-guide/tuning-max-batch-size-and-max-num-tokens.md
- docs/source/legacy/python-api/tensorrt_llm.models.rst
- docs/source/legacy/performance/performance-tuning-guide/useful-runtime-flags.md
- docs/source/legacy/python-api/tensorrt_llm.layers.rst
- docs/source/legacy/advanced/graph-rewriting.md
- docs/source/legacy/performance/performance-tuning-guide/benchmarking-default-performance.md
- docs/source/legacy/python-api/tensorrt_llm.quantization.rst
- docs/source/legacy/reference/multimodal-feature-support-matrix.md
- docs/source/legacy/performance/perf-analysis.md
- docs/source/commands/trtllm-build.rst
- docs/source/legacy/architecture/model-weights-loader.md
- docs/source/legacy/advanced/open-sourced-cutlass-kernels.md
- docs/source/legacy/architecture/checkpoint.md
- docs/source/legacy/advanced/lora.md
- docs/source/legacy/advanced/gpt-runtime.md
- docs/source/legacy/reference/troubleshooting.md
- docs/source/legacy/advanced/executor.md
- docs/source/legacy/architecture/workflow.md
- docs/source/legacy/performance/performance-tuning-guide/useful-build-time-flags.md
- docs/source/legacy/advanced/kv-cache-management.md
- docs/source/legacy/advanced/speculative-decoding.md
- docs/source/legacy/torch.md
- docs/source/legacy/advanced/gpt-attention.md
- docs/source/legacy/reference/memory.md
- docs/source/legacy/architecture/core-concepts.md
- docs/source/legacy/dev-on-cloud/build-image-to-dockerhub.md
- docs/source/legacy/advanced/kv-cache-reuse.md
- docs/source/legacy/architecture/add-model.md
- docs/source/legacy/performance/performance-tuning-guide/deciding-model-sharding-strategy.md
- docs/source/legacy/performance/performance-tuning-guide/fp8-quantization.md
- docs/source/legacy/dev-on-cloud/dev-on-runpod.md
Wanli-Jiang
force-pushed
the
user/williamj/deprecated-trt-backend-step1p2
branch
2 times, most recently
from
66887fe to
9f9523b
Compare
Collaborator
Author
|
/bot run --disable-fail-fast |
Collaborator
|
PR_Github #56578 [ run ] triggered by Bot. Commit: |
nv-guomingz
reviewed
Jun 30, 2026
nv-guomingz
reviewed
Jun 30, 2026
Collaborator
|
PR_Github #56578 [ run ] completed with state |
Wanli-Jiang
force-pushed
the
user/williamj/deprecated-trt-backend-step1p2
branch
2 times, most recently
from
46c9fb7 to
a73a87f
Compare
nv-guomingz
reviewed
Jul 1, 2026
nv-guomingz
approved these changes
Jul 1, 2026
nv-guomingz
left a comment
nv-guomingz
left a comment
Collaborator
There was a problem hiding this comment.
Overall LGTM, awesome work.
Collaborator
Author
|
/bot run --disable-fail-fast |
Collaborator
|
PR_Github #56888 [ run ] triggered by Bot. Commit: |
Wanli-Jiang
force-pushed
the
user/williamj/deprecated-trt-backend-step1p2
branch
from
a73a87f to
97bfcb0
Compare
…tion guide Prune the docs/source/legacy/ TensorRT-era documentation now that the TensorRT backend is removed. Delete the docs that are purely about the TensorRT engine mechanism: commands/trtllm-build.rst, the six python-api autodoc stubs, advanced/gpt-runtime, advanced/graph-rewriting, advanced/lora, advanced/open-sourced-cutlass-kernels, advanced/weight-streaming, architecture/core-concepts, reference/support-matrix, tensorrt_quickstart, and performance/.../useful-build-time-flags. Keep every other legacy doc at its original content for cross-reference. Add a caution banner to the six kept docs still specifically about the obsolete TensorRT engine mechanism (checkpoint, workflow, precision, memory, fp8-quantization, troubleshooting); leave the rest unbannered. A few link/toctree fixes are forced by the deletions. Add docs/source/legacy/tensorrt-backend-removal.md migration guide (removed-surface table, how-to-migrate, retained-doc index), add a release-note entry, and drop the stale PyExecutor "alternative backend" sentence in torch/arch_overview.md. No Python/C++ source touched. Signed-off-by: Wanli Jiang <35160485+Wanli-Jiang@users.noreply.github.com>
Enable the MyST `html_inline` extension in docs/source/conf.py and wrap the "[BREAKING CHANGE] TensorRT backend removed" label in an inline `<span style="color: red">` so it renders red in the HTML documentation. Signed-off-by: Wanli Jiang <35160485+Wanli-Jiang@users.noreply.github.com>
Wanli-Jiang
force-pushed
the
user/williamj/deprecated-trt-backend-step1p2
branch
from
97bfcb0 to
0a23609
Compare
Collaborator
Author
|
/bot run --disable-fail-fast |
nv-guomingz
approved these changes
Jul 1, 2026
Collaborator
|
PR_Github #56892 [ run ] triggered by Bot. Commit: |
Collaborator
|
PR_Github #56888 [ run ] completed with state |
Collaborator
|
PR_Github #56892 [ run ] completed with state
|
Collaborator
Author
|
/bot run --disable-fail-fast |
Collaborator
|
PR_Github #56897 [ run ] triggered by Bot. Commit: |
Collaborator
|
PR_Github #56897 [ run ] completed with state |
evezhier
pushed a commit
to evezhier/TensorRT-LLM
that referenced
this pull request
Jul 2, 2026
…de (NVIDIA#15767) Signed-off-by: Wanli Jiang <35160485+Wanli-Jiang@users.noreply.github.com>
BrianLi23
pushed a commit
to BrianLi23/TensorRT-LLM
that referenced
this pull request
Jul 9, 2026
…de (NVIDIA#15767) Signed-off-by: Wanli Jiang <35160485+Wanli-Jiang@users.noreply.github.com>
This was referenced Jul 9, 2026
Open
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Step 1.2 of removing the legacy TensorRT backend. Low-touch, docs-only:
maincontent — no rewriting.TensorRT engine mechanism; leave the rest untouched.
docs/only. Zero Python/C++ source touched.30 files changed, +114 / −1522. Commit46c9fb74f4.Breakdown
main(no banner, no edit)1. Deleted (16) — pure TensorRT-backend files
docs/source/…)commands/trtllm-build.rsttrtllm-buildCLIlegacy/python-api/tensorrt_llm.{functional,layers,models,plugin,quantization,runtime}.rst(6)legacy/advanced/gpt-runtime.mdlegacy/advanced/graph-rewriting.mdINetworkDefinition/ILayergraph rewritinglegacy/advanced/lora.mdlegacy/advanced/open-sourced-cutlass-kernels.mdlegacy/advanced/weight-streaming.mdtrtllm-buildweight streaming (no PyTorch support)legacy/architecture/core-concepts.mdBuilder/INetworkDefinition/pluginslegacy/reference/support-matrix.mdlegacy/tensorrt_quickstart.mdliteralinclude)legacy/performance/…/useful-build-time-flags.md2. Bannered (6) — kept, but obsolete TRT engine mechanism
Each =
maincontent + a caution admonition (below), placed after the H1.architecture/checkpoint.mdtrtllm-buildarchitecture/workflow.mdtrtllm-build→ engine workflowreference/precision.mdQuantModeflags + TRT quant recipes/support matrixreference/memory.mddevice_memory_size_v2/TrtGptModelperformance/…/fp8-quantization.mdBuildConfig.plugin_config(reduce-fusion / gemm / low-latency-gemm)reference/troubleshooting.mdregister_network_output/trtllm-buildengine debugging3. Other modified
Small fixes (no banner):
key-features.md../features/lora.md(advanced/lora.mddeleted)architecture/add-model.mdworkflow[./workflow.md]linkperformance/…/tuning-max-batch-size-and-max-num-tokens.mduseful-build-time-flags.md(
fp8-quantization.md, bannered above, similarly de-linked 3 such refs.)Infra:
index.rsttensorrt_quickstartwith the migration guide in the toctreeperformance/…/index.rstuseful-build-time-flagstoctree entryrelease-notes.md[BREAKING CHANGE] TensorRT backend removednotetorch/arch_overview.md4. Added
legacy/tensorrt-backend-removal.md— concise migration guide: what changed(removed-surface table), how to migrate (Python + CLI), and an index of the
retained reference docs.
5. Kept unchanged (19) — no banner
The remaining legacy docs are byte-identical to
main. Their capability is stilllive in the PyTorch backend or they are general knowledge / infrastructure:
gpt-attention,executor,expert-parallelism,kv-cache-management,kv-cache-reuse,speculative-decoding,disaggregated-service,lowprecision-pcie-allreduce,model-weights-loader,perf-analysis,perf-benchmarking,benchmarking-default-performance,deciding-model-sharding-strategy,introduction,useful-runtime-flags,multimodal-feature-support-matrix,torch, and bothdev-on-cloud/*.6. Integrity
(label)=anchors still resolve.pre-commitpassed (docs-only). No source touched.Summary by CodeRabbit
New Features
Breaking Changes
Documentation
Description
Test Coverage
PR Checklist
Please review the following before submitting your PR:
PR description clearly explains what and why. If using CodeRabbit's summary, please make sure it makes sense.
PR Follows TRT-LLM CODING GUIDELINES to the best of your knowledge.
Test cases are provided for new code paths (see test instructions)
If PR introduces API changes, an appropriate PR label is added - either
api-compatibleorapi-breaking. Forapi-breaking, includeBREAKINGin the PR title.Any new dependencies have been scanned for license and vulnerabilities
CODEOWNERS updated if ownership changes
Documentation updated as needed
Update tava architecture diagram if there is a significant design change in PR.
The reviewers assigned automatically/manually are appropriate for the PR.
Please check this after reviewing the above items as appropriate for this PR.
GitHub Bot Help
To see a list of available CI bot commands, please comment
/bot help.