[TRTLLM-6994][feat] FP8 Context MLA integration (Cherry-pick https://github.com/NVIDIA/TensorRT-LLM/pull/6059 from release/1.1.0rc2) #7610
Conversation
|
/bot run --disable-fail-fast |
yuxianq
changed the title
|
PR_Github #18020 [ run ] triggered by Bot |
📝 WalkthroughWalkthroughEnables FP8-context MLA by relaxing checks and removing forced output dtype; adds mFP8ContextMLA and KV cache quant mode plumbing; switches Runner allocations to shared_ptr; expands FMHA kernel hash info; centralizes quant scale/out_scale handling in PyTorch Attention; increases initialization traceback depth; adds a max prompt length assertion; updates tests with Hopper gating and MOE backend selection. Changes
Sequence Diagram(s)sequenceDiagram
autonumber
participant User
participant TorchAttention as Attention (PyTorch)
participant OProj as o_proj
participant Backend as TRT-LLM AttentionOp
Note over TorchAttention,OProj: Weight creation and quant flag discovery
User->>TorchAttention: create_weights()
TorchAttention->>OProj: create_weights()
OProj-->>TorchAttention: returns (quant flags, out_scale)
TorchAttention->>TorchAttention: has_quant_scale = (FP8/NVFP4 flags)\nout_scale = o_proj.out_scale
Note over TorchAttention,Backend: Forward path with centralized out_scale
User->>TorchAttention: forward(...)
TorchAttention->>Backend: attention(..., out_scale=self.out_scale)
Backend-->>TorchAttention: outputs
TorchAttention-->>User: result
sequenceDiagram
autonumber
participant Host as Host Init
participant AttnOp as AttentionOp
participant Runner as FMHA/MLA Runner
Note over AttnOp: MLA enablement with FP8-context
Host->>AttnOp: initialize(...)
AttnOp->>AttnOp: mFP8ContextMLA = (SM in {100,120} && KvCache supports FP8)
AttnOp->>AttnOp: if MLA enabled ensure !DenseContextFMHA
AttnOp->>Runner: construct (shared_ptr)
Runner-->>AttnOp: ready
Estimated code review effort🎯 4 (Complex) | ⏱️ ~60 minutes Possibly related PRs
Suggested labels
Suggested reviewers
✨ Finishing Touches
🧪 Generate unit tests
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 4
Caution
Some comments are outside the diff and can’t be posted inline due to platform limitations.
⚠️ Outside diff range comments (1)
tensorrt_llm/_torch/modules/attention.py (1)
317-327: Fix potential dtype mismatch under torch.compile path (custom op disables NVFP4 output).create_output may allocate an FP8 output tensor when has_quant_scale and FP8/FP4 KV cache are enabled. However, attn_custom_op_inplace invokes _attn_impl(..., enable_attn_nvfp4_output=False) under torch.compile, which can yield BF16 output into an FP8 buffer. Make dtype selection conditional on the same enable flag.
Apply:
- def create_output(self, q: torch.Tensor): + def create_output(self, q: torch.Tensor, enable_attn_nvfp4_output: bool = True): num_tokens = q.shape[0] hidden_size = self.o_proj.in_features out_dtype = q.dtype - if self.attn_backend == "TRTLLM": + if self.attn_backend == "TRTLLM" and enable_attn_nvfp4_output: if self.has_quant_scale and (self.attn.has_fp8_kv_cache or self.attn.has_fp4_kv_cache): out_dtype = torch.float8_e4m3fn output = q.new_empty([num_tokens, hidden_size], dtype=out_dtype) return outputAnd update the compile path call site:
# In forward_impl(), inside if use_custom_inplace_op: - output = self.create_output(q) + output = self.create_output(q, enable_attn_nvfp4_output=False)This keeps buffer dtype aligned with the execution path.
🧹 Nitpick comments (13)
tensorrt_llm/_torch/pyexecutor/model_engine.py (2)
1-1: Missing NVIDIA Apache-2.0 header (2025).Per guidelines, prepend the NVIDIA Apache-2.0 header with the current year.
Apply:
+# Copyright (c) 2025, NVIDIA CORPORATION. All rights reserved. +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# http://www.apache.org/licenses/LICENSE-2.0 +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License.
1009-1013: Traceback limit increased to 10: consider log level.Good for debugging, but verbose for INFO. Suggest logging the full traceback at DEBUG, or gate by an env flag.
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/fmhaKernels.h (2)
1-15: Update copyright year.Header shows 2020–2023; update to include 2025.
- * Copyright (c) 2020-2023, NVIDIA CORPORATION. All rights reserved. + * Copyright (c) 2020-2025, NVIDIA CORPORATION. All rights reserved.
544-547: Log readability: print dtype names instead of ints.The info string logs dtypeQ/Kv/Out as integers. Prefer symbolic names for quicker debugging.
Example:
- std::string info = "dtypeQ=" + std::to_string(static_cast<int>(mDtypeQ)) + ", dtypeKv=" - + std::to_string(static_cast<int>(mDtypeKv)) + ", dtypeOut=" + std::to_string(static_cast<int>(mDtypeOut)) + std::string info = "dtypeQ=" + toString(mDtypeQ) + ", dtypeKv=" + + toString(mDtypeKv) + ", dtypeOut=" + toString(mDtypeOut) + ", sm=" + std::to_string(mSM) + ", qkvLayout=" + std::to_string(static_cast<int>(params.mQkvLayout))(Add a small toString(Data_type) helper if not present.)
tensorrt_llm/executor/worker.py (1)
519-525: Minor: variable name typo.Consider renaming splited_prompt_len → split_prompt_len for clarity (optional).
- splited_prompt_len = int(len(prompt_token_ids) / cp_size) - default_max_tokens = max_seq_len - splited_prompt_len - query_token_len + split_prompt_len = int(len(prompt_token_ids) / cp_size) + default_max_tokens = max_seq_len - split_prompt_len - query_token_lentests/integration/defs/accuracy/test_llm_api_pytorch.py (4)
1239-1241: Avoid repetition: factor MOE backend selection into a helper.The "DEEPGEMM if SM>=100 else CUTLASS" logic is duplicated across tests. Suggest a small helper to keep tests DRY.
Example (place near the top of this file):
def _moe_backend_for_ci(): return "DEEPGEMM" if get_sm_version() >= 100 else "CUTLASS"Then here:
moe_config=MoeConfig(backend=_moe_backend_for_ci())
1329-1331: Same refactor applies here.Use the shared helper to choose the MOE backend.
1353-1355: Same refactor applies here.Use the shared helper to choose the MOE backend.
1397-1399: Same refactor applies here.Use the shared helper to choose the MOE backend.
tensorrt_llm/_torch/modules/attention.py (1)
298-299: Guard against re-initialization of o_proj weights.Attention.create_weights() now unconditionally calls self.o_proj.create_weights(). If init already created weights (default path), this may reinitialize or conflict unless Linear.create_weights is idempotent.
Please confirm Linear.create_weights is idempotent (e.g., via an internal _weights_created guard). If not, guard:
- self.o_proj.create_weights() + if not getattr(self.o_proj, "_weights_created", False): + self.o_proj.create_weights()cpp/tensorrt_llm/common/attentionOp.cpp (1)
2573-2574: Fix wording in user-visible error messageChange “currently not support dense fmha” to “does not currently support dense FMHA” for clarity.
- TLLM_CHECK_WITH_INFO(!mDenseContextFMHA, "MLA(Deepseek v2) currently not support dense fmha"); + TLLM_CHECK_WITH_INFO(!mDenseContextFMHA, "MLA (Deepseek v2) does not currently support dense FMHA");cpp/tensorrt_llm/thop/attentionOp.cpp (2)
709-723: Validate workspace dtype and size in bytesThe check uses numel() (elements) against workspace_size (bytes). If a caller passes a non-Byte tensor, the comparison and resize logic become inconsistent. Guard for dtype Byte or compute in bytes.
- if (workspace_.has_value()) + if (workspace_.has_value()) { - if (workspace_.value().numel() < workspace_size) + auto ws = workspace_.value(); + TORCH_CHECK(ws.dtype() == torch::kByte, "workspace must be a torch.uint8 (Byte) tensor"); + if (ws.numel() < workspace_size) // numel == bytes for Byte tensors { TLLM_LOG_WARNING("Attention workspace size is not enough, increase the size from %ld bytes to %ld bytes", - workspace_.value().numel(), workspace_size); - workspace_.value().resize_({workspace_size}); + ws.numel(), workspace_size); + ws.resize_({workspace_size}); } - workspace = workspace_.value(); + workspace = ws; }
1-16: Header year nitGuidelines ask to prepend the NVIDIA Apache-2.0 header with the current year; file shows 1993-2024. Consider updating to include 2025 where applicable.
📜 Review details
Configuration used: Path: .coderabbit.yaml
Review profile: CHILL
Plan: Pro
📒 Files selected for processing (8)
cpp/tensorrt_llm/common/attentionOp.cpp(1 hunks)cpp/tensorrt_llm/common/attentionOp.h(1 hunks)cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/fmhaKernels.h(1 hunks)cpp/tensorrt_llm/thop/attentionOp.cpp(3 hunks)tensorrt_llm/_torch/modules/attention.py(9 hunks)tensorrt_llm/_torch/pyexecutor/model_engine.py(1 hunks)tensorrt_llm/executor/worker.py(1 hunks)tests/integration/defs/accuracy/test_llm_api_pytorch.py(9 hunks)
🧰 Additional context used
📓 Path-based instructions (8)
**/*.{h,hpp,hh,hxx,cpp,cxx,cc,cu,cuh,py}
📄 CodeRabbit inference engine (CODING_GUIDELINES.md)
Use only spaces, no tabs; indent with 4 spaces.
Files:
tensorrt_llm/_torch/pyexecutor/model_engine.pycpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/fmhaKernels.hcpp/tensorrt_llm/common/attentionOp.cpptensorrt_llm/_torch/modules/attention.pycpp/tensorrt_llm/thop/attentionOp.cpptensorrt_llm/executor/worker.pytests/integration/defs/accuracy/test_llm_api_pytorch.pycpp/tensorrt_llm/common/attentionOp.h
**/*.py
📄 CodeRabbit inference engine (CODING_GUIDELINES.md)
**/*.py: Python code must target Python 3.8+.
Indent Python code with 4 spaces; do not use tabs.
Maintain module namespace when importing; prefer 'from package.subpackage import foo' then 'foo.SomeClass()' instead of importing the class directly.
Python filenames should be snake_case (e.g., some_file.py).
Python classes use PascalCase names.
Functions and methods use snake_case names.
Local variables use snake_case; prefix 'k' for variables that start with a number (e.g., k_99th_percentile).
Global variables use upper SNAKE_CASE prefixed with 'G' (e.g., G_MY_GLOBAL).
Constants use upper SNAKE_CASE (e.g., MY_CONSTANT).
Avoid shadowing variables from an outer scope.
Initialize all externally visible members of a class in the constructor.
Prefer docstrings for interfaces that may be used outside a file; comments for in-function or file-local interfaces.
Use Google-style docstrings for classes and functions (Sphinx-parsable).
Document attributes and variables inline so they render under the class/function docstring.
Avoid reflection when a simpler, explicit approach suffices (e.g., avoid dict(**locals()) patterns).
In try/except, catch the most specific exceptions possible.
For duck-typing try/except, keep the try body minimal and use else for the main logic.
Files:
tensorrt_llm/_torch/pyexecutor/model_engine.pytensorrt_llm/_torch/modules/attention.pytensorrt_llm/executor/worker.pytests/integration/defs/accuracy/test_llm_api_pytorch.py
**/*.{cpp,cxx,cc,h,hpp,hh,hxx,cu,cuh,py}
📄 CodeRabbit inference engine (CODING_GUIDELINES.md)
Prepend the NVIDIA Apache-2.0 copyright header with current year to the top of all source files (e.g., .cpp, .h, .cu, .py).
Files:
tensorrt_llm/_torch/pyexecutor/model_engine.pycpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/fmhaKernels.hcpp/tensorrt_llm/common/attentionOp.cpptensorrt_llm/_torch/modules/attention.pycpp/tensorrt_llm/thop/attentionOp.cpptensorrt_llm/executor/worker.pytests/integration/defs/accuracy/test_llm_api_pytorch.pycpp/tensorrt_llm/common/attentionOp.h
**/*.{h,hpp,hh,hxx,cpp,cxx,cc,cu,cuh}
📄 CodeRabbit inference engine (CODING_GUIDELINES.md)
**/*.{h,hpp,hh,hxx,cpp,cxx,cc,cu,cuh}: Namespace closing braces must include a trailing comment with the namespace name (e.g., '} // namespace foo').
Prefer const or constexpr variables over #define for constants.
Declare variables that are not modified after initialization as const.
Avoid magic literals in code; except for 0, nullptr, true, false. Use named constants for comparisons and logic.
Use Allman brace style for formatting.
Place the semicolon of an empty for/while loop on a new line.
Bodies of switch/while/do-while/for must be compound statements (brace-delimited), and if/else must always be followed by brace-delimited statements.
Type names (e.g., classes) must be CamelCase starting with an uppercase letter (e.g., FooBar).
Local variables, methods, and namespaces use lowerCamelCase (e.g., localFooBar).
Non-magic-number global variables that are non-static and not in an anonymous namespace must be lowerCamelCase prefixed with 'g' (e.g., gDontUseGlobalFoos).
Non-magic-number globals that are static or in an anonymous namespace use lowerCamelCase prefixed with 's' (e.g., sMutableStaticGlobal).
Locally visible static variables use lowerCamelCase with 's' prefix (e.g., static std::once_flag sFlag).
Private/protected member variables use 'm' prefix with CamelCase (e.g., mNbFooValues). Public members may omit, but 'm' is encouraged for clarity.
Constants (enums, global constants, static constants, and function-scope magic/literal constants) use uppercase SNAKE_CASE with 'k' prefix (e.g., kDIGIT_NUM).
Function-scope constants that are not magic numbers or literals are named like non-constant variables (e.g., bool const pass = a && b).
If macros are necessary, name them in UPPER_SNAKE_CASE (e.g., FOO_VERSION) and prefer constants over #define.
Use LLVM clang-format; wrap lines at a maximum of 120 columns; use '// clang-format off/on' sparingly with justification.
Use smart pointers for heap allocations; prefer unique_ptr for sole ownership, shared_ptr for shared...
Files:
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/fmhaKernels.hcpp/tensorrt_llm/common/attentionOp.cppcpp/tensorrt_llm/thop/attentionOp.cppcpp/tensorrt_llm/common/attentionOp.h
**/*.{cpp,cxx,cc,cu,h,hpp,hh,hxx,cuh}
📄 CodeRabbit inference engine (CODING_GUIDELINES.md)
C++ filenames should be lowerCamelCase (first letter lowercase) and must be case-insensitive unique within a compilation target.
Files:
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/fmhaKernels.hcpp/tensorrt_llm/common/attentionOp.cppcpp/tensorrt_llm/thop/attentionOp.cppcpp/tensorrt_llm/common/attentionOp.h
**/*.{h,hpp,hh,hxx}
📄 CodeRabbit inference engine (CODING_GUIDELINES.md)
Document new class interfaces and function prototypes with Doxygen; use //! for single-line and //!< for members.
Files:
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/fmhaKernels.hcpp/tensorrt_llm/common/attentionOp.h
**/*.{h,hpp,hh,hxx,cpp,cxx,cc}
📄 CodeRabbit inference engine (CODING_GUIDELINES.md)
**/*.{h,hpp,hh,hxx,cpp,cxx,cc}: Prefer anonymous namespaces over 'static' for internal linkage of functions.
All templates (class/function/member/static) must be instantiated at least once; non-POD classes should have private data members.
Files:
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/fmhaKernels.hcpp/tensorrt_llm/common/attentionOp.cppcpp/tensorrt_llm/thop/attentionOp.cppcpp/tensorrt_llm/common/attentionOp.h
**/*.{h,hpp,hh,hxx,cuh}
📄 CodeRabbit inference engine (CODING_GUIDELINES.md)
Use include guards named 'TRTLLM_<FILE_NAME_IN_CAPS_WITH_UNDERSCORES>_H' (no leading or trailing underscore; directory names excluded).
Files:
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/fmhaKernels.hcpp/tensorrt_llm/common/attentionOp.h
🧠 Learnings (3)
📚 Learning: 2025-08-14T21:04:50.248Z
Learnt from: thorjohnsen
PR: NVIDIA/TensorRT-LLM#6910
File: cpp/tensorrt_llm/batch_manager/kvCacheManager.cpp:0-0
Timestamp: 2025-08-14T21:04:50.248Z
Learning: In KV cache onboarding logic during prefill in cpp/tensorrt_llm/batch_manager/kvCacheManager.cpp, when calculating which blocks fall within the attention window, use getTokensPerBlock() to advance token indices rather than block->getUniqueTokens().size(), because the calculation needs to consider the post-prefill state where blocks will be filled to capacity, not their current token count.
Applied to files:
cpp/tensorrt_llm/thop/attentionOp.cpp
📚 Learning: 2025-08-26T09:37:10.463Z
Learnt from: jiaganc
PR: NVIDIA/TensorRT-LLM#7031
File: tensorrt_llm/bench/dataclasses/configuration.py:90-104
Timestamp: 2025-08-26T09:37:10.463Z
Learning: In TensorRT-LLM's bench configuration, the `get_pytorch_perf_config()` method returns `self.pytorch_config` which is a Dict[str, Any] that can contain default values including `cuda_graph_config`, making the fallback `llm_args["cuda_graph_config"]` safe to use.
Applied to files:
tests/integration/defs/accuracy/test_llm_api_pytorch.py
📚 Learning: 2025-08-26T09:37:10.463Z
Learnt from: jiaganc
PR: NVIDIA/TensorRT-LLM#7031
File: tensorrt_llm/bench/dataclasses/configuration.py:90-104
Timestamp: 2025-08-26T09:37:10.463Z
Learning: In TensorRT-LLM, the `get_pytorch_perf_config()` method returns `self.pytorch_config` which can contain default `cuda_graph_config` values, so `llm_args` may already have this config before the extra options processing.
Applied to files:
tests/integration/defs/accuracy/test_llm_api_pytorch.py
🧬 Code graph analysis (4)
tensorrt_llm/_torch/modules/attention.py (3)
tensorrt_llm/_torch/modules/fused_moe/fused_moe_trtllm_gen.py (1)
create_weights(157-179)tensorrt_llm/_torch/modules/linear.py (18)
create_weights(219-222)create_weights(274-284)create_weights(317-340)create_weights(484-504)create_weights(585-608)create_weights(717-757)create_weights(904-944)create_weights(1096-1119)create_weights(1208-1227)create_weights(1324-1351)create_weights(1467-1507)create_weights(1723-1726)create_weights(1851-1860)has_fp8_qdq(1869-1872)has_nvfp4(1887-1890)has_fp8_block_scales(1881-1884)has_fp8_rowwise(1875-1878)has_w4a8_nvfp4_fp8(1911-1914)tensorrt_llm/quantization/mode.py (2)
has_fp8_kv_cache(166-167)has_fp4_kv_cache(169-170)
cpp/tensorrt_llm/thop/attentionOp.cpp (3)
cpp/tensorrt_llm/common/attentionOp.cpp (1)
if(1363-1366)cpp/tensorrt_llm/kernels/unfusedAttentionKernels.h (3)
if(334-337)if(340-354)if(374-377)tensorrt_llm/models/modeling_utils.py (3)
quant_mode(156-163)quant_mode(301-303)quant_mode(543-544)
tensorrt_llm/executor/worker.py (1)
tensorrt_llm/executor/result.py (1)
prompt_token_ids(525-526)
tests/integration/defs/accuracy/test_llm_api_pytorch.py (3)
tensorrt_llm/llmapi/llm_args.py (1)
MoeConfig(168-196)tensorrt_llm/layers/moe.py (1)
MoeConfig(104-140)tensorrt_llm/_utils.py (1)
get_sm_version(689-691)
🪛 Ruff (0.12.2)
tensorrt_llm/executor/worker.py
515-515: Use of assert detected
(S101)
⏰ Context from checks skipped due to timeout of 90000ms. You can increase the timeout in your CodeRabbit configuration to a maximum of 15 minutes (900000ms). (1)
- GitHub Check: Pre-commit Check
🔇 Additional comments (8)
cpp/tensorrt_llm/common/attentionOp.h (1)
469-476: Drop the tuple‐layout break warning
AttentionOp::data()is only used wholesale as a key in anunordered_map—no structured bindings orstd::tieunpack its elements, nor is it serialized for engine/plugins—so insertingmFP8ContextMLAin the middle has no positional‐unpack or ABI impact.tests/integration/defs/accuracy/test_llm_api_pytorch.py (2)
1215-1215: Gating change looks good.The Hopper gating via @skip_pre_hopper aligns with the FP8-block-scales coverage.
1485-1486: Gating change looks good.The Hopper gating for static_eplb covers kernel availability appropriately.
tensorrt_llm/_torch/modules/attention.py (2)
848-850: LGTM: MLA keeps out_scale None (BF16 output), consistent with comment.No issues.
1062-1062: LGTM: Pass-through of out_scale to MHA/MQA is consistent.With out_scale=None in MLA, these calls retain BF16 outputs while enabling scale plumbing when needed later.
Also applies to: 1126-1126, 1226-1226, 1274-1274, 1376-1376
cpp/tensorrt_llm/thop/attentionOp.cpp (3)
532-533: Good switch to make_sharedReplacing raw new/reset with std::make_shared reduces verbosity and improves exception safety.
Also applies to: 536-537, 541-542, 547-548, 554-555, 558-559, 563-564
582-582: QuantMode initialization looks correctSetting mKVCacheQuantMode early from the torch int matches the new plumbing; no issues spotted here.
630-633: Clarify SM gating for mFP8ContextMLA
mFP8ContextMLA is currently enabled only on SM100 and SM120; if Hopper (SM90) should support FP8 KV-cache MLA, broaden the check to includesm == 90(e.g.sm == 90 || sm == 100 || sm == 120), otherwise add a brief comment explaining why SM90 is excluded. [attentionOp.cpp:630-633]
|
PR_Github #18020 [ run ] completed with state |
|
/bot run --disable-fail-fast |
|
PR_Github #18135 [ run ] triggered by Bot |
|
PR_Github #18135 [ run ] completed with state |
Signed-off-by: Yuxian Qiu <[email protected]>
yuxianq
force-pushed
the
fp8-context-mla-main
branch
from
b550a05 to
70af66a
Compare
|
/bot run --disable-fail-fast |
|
PR_Github #18770 [ run ] triggered by Bot |
|
PR_Github #18770 [ run ] completed with state |
|
/bot run --disable-fail-fast |
|
PR_Github #18892 [ run ] triggered by Bot |
|
PR_Github #18892 [ run ] completed with state |
|
/bot run |
|
PR_Github #18993 [ run ] triggered by Bot |
|
PR_Github #18993 [ run ] completed with state |
Signed-off-by: Yuxian Qiu <[email protected]>
|
/bot run --disable-fail-fast |
|
PR_Github #19106 [ run ] triggered by Bot |
|
PR_Github #19106 [ run ] completed with state |
|
/bot run --disable-fail-fast |
|
PR_Github #19185 [ run ] triggered by Bot |
|
PR_Github #19185 [ run ] completed with state |
from release/1.1.0rc2) (NVIDIA#7610) Signed-off-by: Yuxian Qiu <[email protected]>
from release/1.1.0rc2) (NVIDIA#7610) Signed-off-by: Yuxian Qiu <[email protected]>
from release/1.1.0rc2) (NVIDIA#7610) Signed-off-by: Yuxian Qiu <[email protected]>
Summary by CodeRabbit
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)
Any new dependencies have been scanned for license and vulnerabilities
CODEOWNERS updated if ownership changes
Documentation updated as needed
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
/bot [-h] ['run', 'kill', 'skip', 'reuse-pipeline'] ...Provide a user friendly way for developers to interact with a Jenkins server.
Run
/bot [-h|--help]to print this help message.See details below for each supported subcommand.
run [--reuse-test (optional)pipeline-id --disable-fail-fast --skip-test --stage-list "A10-PyTorch-1, xxx" --gpu-type "A30, H100_PCIe" --test-backend "pytorch, cpp" --add-multi-gpu-test --only-multi-gpu-test --disable-multi-gpu-test --post-merge --extra-stage "H100_PCIe-TensorRT-Post-Merge-1, xxx" --detailed-log --debug(experimental)]Launch build/test pipelines. All previously running jobs will be killed.
--reuse-test (optional)pipeline-id(OPTIONAL) : Allow the new pipeline to reuse build artifacts and skip successful test stages from a specified pipeline or the last pipeline if no pipeline-id is indicated. If the Git commit ID has changed, this option will be always ignored. The DEFAULT behavior of the bot is to reuse build artifacts and successful test results from the last pipeline.--disable-reuse-test(OPTIONAL) : Explicitly prevent the pipeline from reusing build artifacts and skipping successful test stages from a previous pipeline. Ensure that all builds and tests are run regardless of previous successes.--disable-fail-fast(OPTIONAL) : Disable fail fast on build/tests/infra failures.--skip-test(OPTIONAL) : Skip all test stages, but still run build stages, package stages and sanity check stages. Note: Does NOT update GitHub check status.--stage-list "A10-PyTorch-1, xxx"(OPTIONAL) : Only run the specified test stages. Examples: "A10-PyTorch-1, xxx". Note: Does NOT update GitHub check status.--gpu-type "A30, H100_PCIe"(OPTIONAL) : Only run the test stages on the specified GPU types. Examples: "A30, H100_PCIe". Note: Does NOT update GitHub check status.--test-backend "pytorch, cpp"(OPTIONAL) : Skip test stages which don't match the specified backends. Only support [pytorch, cpp, tensorrt, triton]. Examples: "pytorch, cpp" (does not run test stages with tensorrt or triton backend). Note: Does NOT update GitHub pipeline status.--only-multi-gpu-test(OPTIONAL) : Only run the multi-GPU tests. Note: Does NOT update GitHub check status.--disable-multi-gpu-test(OPTIONAL) : Disable the multi-GPU tests. Note: Does NOT update GitHub check status.--add-multi-gpu-test(OPTIONAL) : Force run the multi-GPU tests in addition to running L0 pre-merge pipeline.--post-merge(OPTIONAL) : Run the L0 post-merge pipeline instead of the ordinary L0 pre-merge pipeline.--extra-stage "H100_PCIe-TensorRT-Post-Merge-1, xxx"(OPTIONAL) : Run the ordinary L0 pre-merge pipeline and specified test stages. Examples: --extra-stage "H100_PCIe-TensorRT-Post-Merge-1, xxx".--detailed-log(OPTIONAL) : Enable flushing out all logs to the Jenkins console. This will significantly increase the log volume and may slow down the job.--debug(OPTIONAL) : Experimental feature. Enable access to the CI container for debugging purpose. Note: Specify exactly one stage in thestage-listparameter to access the appropriate container environment. Note: Does NOT update GitHub check status.For guidance on mapping tests to stage names, see
docs/source/reference/ci-overview.mdand the
scripts/test_to_stage_mapping.pyhelper.kill
killKill all running builds associated with pull request.
skip
skip --comment COMMENTSkip testing for latest commit on pull request.
--comment "Reason for skipping build/test"is required. IMPORTANT NOTE: This is dangerous since lack of user care and validation can cause top of tree to break.reuse-pipeline
reuse-pipelineReuse a previous pipeline to validate current commit. This action will also kill all currently running builds associated with the pull request. IMPORTANT NOTE: This is dangerous since lack of user care and validation can cause top of tree to break.