[None][chore] Simplify MoE lora

[brb-nv]

Description

This MR collapses the routed-expert MoE LoRA code from three execution variants down to a single grouped-GEMM core fed by two input front-ends (eager per-request and CUDA-graph slot-indexed) - with no change to runtime behavior.

Currently, there are 3 execution methods for routed-expert MoE LoRA:

• host_path - eager, legacy cuBLAS/LoraImpl path (per-request D2H pointer expand)
• device_path_eager - eager, capture-safe grouped-GEMM path (per-request input), formerly forced on via TLLM_MOE_LORA_USE_DEVICE_PATH.
• device_path_cudagraph - CUDA-graph decode, slot-indexed grouped-GEMM path.

This MR removed the first one. Goal is to minimize codepaths for maintainability and share as much code as possible for eager and cudagraph paths.

This is just a refactor guarded by existing tests.

Test Coverage

$ pytest tests/unittest/_torch/lora/test_moe_lora_op.py -s -v
$ pytest tests/unittest/_torch/lora/test_moe_lora_grouped_gemm.py -s -v
$ pytest tests/unittest/_torch/lora/test_moe_lora_extract.py -s -v
$ pytest tests/unittest/_torch/lora/test_moe_lora_cuda_graph_params.py -s -v
$ pytest tests/unittest/llmapi/test_llm_pytorch.py::test_qwen_moe_routed_expert_multi_lora_varying_ranks -s -v
$ ./cpp/tests/unit_tests/kernels/moeLoraProblemBuilderTest --gtest_filter='MoeLoraProblemBuilderTest.*'

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-compatible or api-breaking. For api-breaking, include BREAKING in 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.

[brb-nv]

/bot run --disable-fail-fast

[coderabbitai]

📝 Walkthrough

Walkthrough

The PR replaces the legacy MoE LoRA "device-path / cuBLAS" execution with a unified capture-safe "grouped-GEMM" core. C++ header types are renamed (MoeLoraDevicePath → MoeLoraGroupedGemm), kernel execution branches are rewritten to dispatch through grouped-GEMM, cuBLAS infrastructure is removed from the TorchBind op, Python LoRA module constants are centralized, and tests are updated to drop TLLM_MOE_LORA_USE_DEVICE_PATH environment-variable toggling.

Changes

MoE LoRA Device-Path → Grouped-GEMM Migration

Layer / File(s)	Summary
C++ grouped-GEMM type contracts (headers) cpp/tensorrt_llm/kernels/cutlass_kernels/include/moe_lora_grouped_gemm.h, cpp/tensorrt_llm/kernels/cutlass_kernels/include/moe_kernels.h	Renames MoeLoraDevicePath/MoeLoraDevicePathModule to MoeLoraGroupedGemm/MoeLoraGroupedGemmModule, introduces MoeLoraGroupedGemmRunFn dispatch type, and replaces LoraParams.device_path with LoraParams.grouped_gemm.
C++ kernel grouped-GEMM dispatch branches cpp/tensorrt_llm/kernels/cutlass_kernels/moe_gemm/moe_kernels.cu, cpp/tensorrt_llm/kernels/cutlass_kernels/moe_gemm/moe_lora_pointer_expand.cu, cpp/tests/unit_tests/kernels/moeLoraProblemBuilderTest.cu	Adds runMoeLoraGroupedGemmModule wrapper with null-checks; adds grouped_gemm.enabled early-return branches in setupLoraWorkspace, loraFC1, and loraFC2 (zero + dispatch via grouped-GEMM); updates D2H copy skip condition; updates pointer-expand comment.
TorchBind op grouped-GEMM wiring and cuBLAS removal cpp/tensorrt_llm/thop/moeOp.cpp	Introduces moeLoraGroupedGemmRunImpl; removes cuBLAS LoraImpl cache, mLoraMemcpyEvent, and TLLM_MOE_LORA_USE_DEVICE_PATH env-var; replaces LoraDevicePathBuffers with LoraGroupedGemmBuffers; rewrites ensureLoraDeviceScratch and buildMoeLoraParams for grouped-GEMM scratch allocation; removes computeLoraWorkspaceSize.
Python canonical MoE LoRA module constants and LoRA param utilities tensorrt_llm/_torch/peft/lora/layer.py, tensorrt_llm/_torch/peft/lora/moe_layout.py, tensorrt_llm/_torch/peft/lora/validation.py, tensorrt_llm/_torch/peft/lora/cuda_graph_lora_params.py, tensorrt_llm/_torch/peft/lora/cuda_graph_lora_manager.py	Adds MOE_LORA_MODULE_NAMES and MOE_LORA_MODULE_TO_KERNEL_SLOT in layer.py; updates moe_layout.py and validation.py to import from layer.py; eagerly initializes _moe_slot_ptrs_cache in CudaGraphLoraParams.
Python MoE frontend: slot-extraction helpers, CUDA-graph validation, run_moe call refactor tensorrt_llm/_torch/modules/fused_moe/fused_moe_cutlass.py, tensorrt_llm/_torch/modules/fused_moe/moe_scheduler.py, tensorrt_llm/_torch/modules/fused_moe/ops/moe_op_cutlass.py, tensorrt_llm/_torch/custom_ops/torch_custom_ops.py	Adds _empty_kernel_slot_dict, _gather_moe_lora_slots, _require_fc1_fc2 helpers and raise_moe_lora_multichunk_unsupported; refactors _extract_moe_lora_tensors paths; strengthens reserve_moe_lora_cuda_graph_workspace dtype assertions; refactors compute_moe and fused_moe to build shared run_moe_args lists.
Tests and docs tests/unittest/_torch/lora/test_moe_lora_grouped_gemm.py, tests/unittest/_torch/lora/test_moe_lora_op.py, tests/unittest/_torch/lora/test_moe_lora_extract.py, tests/unittest/_torch/lora/test_moe_lora_cuda_graph_params.py, tests/unittest/llmapi/test_llm_pytorch.py, docs/source/features/lora.md	Drops TLLM_MOE_LORA_USE_DEVICE_PATH monkeypatching from all tests; renames test functions to remove "device path" wording; updates _ExtractStub helpers; refactors llmapi test to run eager and CUDA-graph paths without per-mode parameterization; updates LoRA docs execution-mode table and rejection rules.

Sequence Diagram(s)

sequenceDiagram
  participant PY as fused_moe (Python)
  participant OP as FusedMoeRunner (C++ TorchBind)
  participant BUILD as buildMoeLoraParams
  participant SCRATCH as ensureLoraDeviceScratch
  participant KERNEL as moe_kernels.cu

  PY->>OP: run_moe(..., lora_ranks, lora_weight_ptrs, token_to_slot)
  OP->>OP: capturability check (reject per-request under CUDA graph)
  OP->>BUILD: lora tensors + slot-indexed data
  BUILD->>SCRATCH: allocate LoraGroupedGemmBuffers (kGroupedGemmSplitKSlices)
  SCRATCH-->>BUILD: fc1/fc2/gated device buffers
  BUILD->>BUILD: populateLoraGroupedGemmModule(fc1, fc2, gated)
  BUILD->>BUILD: grouped_gemm.enabled=true, run=moeLoraGroupedGemmRunImpl
  BUILD-->>OP: LoraParams{grouped_gemm}
  OP->>KERNEL: launchMoeKernel(LoraParams)
  KERNEL->>KERNEL: setupLoraWorkspace → grouped-GEMM early-return branch
  KERNEL->>KERNEL: loraFC1 → zero + runMoeLoraGroupedGemmModule(fc1[, gated])
  KERNEL->>KERNEL: loraFC2 → zero + runMoeLoraGroupedGemmModule(fc2)
  KERNEL-->>OP: output
  OP-->>PY: output tensor

Loading

Estimated code review effort

🎯 4 (Complex) | ⏱️ ~60 minutes

Possibly related PRs

• NVIDIA/TensorRT-LLM#14881: Directly related — touches the same routed-expert MoE LoRA CUDA-graph wiring in moeOp.cpp and the slot/token-to-slot expansion path that this PR replaces with grouped-GEMM dispatch.
• NVIDIA/TensorRT-LLM#14923: Directly related — introduced the device_path member in LoraParams/moe_kernels.h and the device-path dispatch in moe_kernels.cu that this PR replaces with grouped_gemm.
• NVIDIA/TensorRT-LLM#15271: Directly related — covers MoE LoRA grouped-GEMM rank-0 "skip out-GEMM to avoid null adapter pointer" behavior and CUDA-graph/LoRA state handling that this PR's kernel branches and test comments build on.

Suggested reviewers

• schetlur-nv
• leslie-fang25
• chang-l
• kaiyux

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 54.93% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title '[None][chore] Simplify MoE lora' clearly summarizes the main change: simplifying the MoE LoRA implementation by consolidating multiple execution paths into a single grouped-GEMM core.
Description check	✅ Passed	The PR description comprehensively explains the refactoring: removing the legacy host_path variant and consolidating device_path_eager and device_path_cudagraph into a single grouped-GEMM core with two input front-ends. It includes test coverage details, a supporting diagram, and completion of the PR checklist.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[coderabbitai]

Actionable comments posted: 3

Caution

Some comments are outside the diff and can’t be posted inline due to platform limitations.

⚠️ Outside diff range comments (2)

cpp/tensorrt_llm/kernels/cutlass_kernels/moe_gemm/moe_kernels.cu (1)

3735-3767: ⚠️ Potential issue | 🟠 Major | ⚡ Quick win

Validate grouped-GEMM metadata before pointer expansion.

This branch now trusts dp.dtype_bytes and each module’s dimensions/buffers to compute byte offsets in launchMoeLoraPointerExpand. If the caller passes stale dimensions, a mismatched dtype size, or a partially populated module, the result is wrong expert weight addresses or a device illegal access before the later GEMM wrapper guard can help.

🛡️ Proposed validation before constructing expand modules

         auto const& dp = lora_params.grouped_gemm;
+        int64_t const expectedDtypeBytes = static_cast<int64_t>(sizeof(ScaleBiasType));
+        TLLM_CHECK_WITH_INFO(dp.dtype_bytes == expectedDtypeBytes,
+            "Grouped-GEMM LoRA dtype_bytes does not match the runner scalar type.");
+
+        auto validateLoraModule = [](MoeLoraGroupedGemmModule const& mod, char const* name,
+                                      int64_t expectedDimA, int64_t expectedDimB)
+        {
+            TLLM_CHECK_WITH_INFO(mod.ranks_src_dev != nullptr && mod.ptrs_src_dev != nullptr
+                    && mod.permuted_ranks_dev != nullptr && mod.permuted_ptrs_dev != nullptr,
+                "Grouped-GEMM LoRA %s module is missing pointer-expand buffers.", name);
+            TLLM_CHECK_WITH_INFO(mod.dim_a == expectedDimA && mod.dim_b == expectedDimB,
+                "Grouped-GEMM LoRA %s module dimensions do not match the MoE runner dimensions.", name);
+        };
+        validateLoraModule(dp.fc1, "fc1", hidden_size, inter_size);
+        validateLoraModule(dp.fc2, "fc2", inter_size, hidden_size);
+        if (is_gated_activation)
+        {
+            validateLoraModule(dp.gated, "gated", hidden_size, inter_size);
+        }
+
         // Translate per-module grouped-GEMM metadata into the MoeLoraExpandModule

🤖 Prompt for 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.

In `@cpp/tensorrt_llm/kernels/cutlass_kernels/moe_gemm/moe_kernels.cu` around
lines 3735 - 3767, Add validation of the grouped-GEMM metadata (dp) before
constructing the MoeLoraExpandModule objects fc1_mod, fc2_mod, and gated_mod.
Validate that dp.dtype_bytes is valid and that each module's dimensions (dim_a,
dim_b) and device buffers (ranks_src_dev, ptrs_src_dev, permuted_ranks_dev,
permuted_ptrs_dev) are properly populated and non-null for fc1 and fc2. For
gated_mod, perform the same validation only when is_gated_activation is true.
This validation should occur before constructing any of the expand modules and
before calling launchMoeLoraPointerExpand to prevent incorrect expert weight
address computation or device illegal memory access.

tensorrt_llm/_torch/modules/fused_moe/fused_moe_cutlass.py (1)

516-523: ⚠️ Potential issue | 🟠 Major | ⚡ Quick win

Detect CUDA-graph MoE LoRA params in _moe_lora_active.

_extract_moe_lora_tensors handles use_cuda_graph_mode without a per-layer eager dict, but _moe_lora_active returns False before checking cuda_graph_params. That lets slot-indexed MoE LoRA bypass the multi-chunk rejection and stray-param guard.

Proposed fix

     def _moe_lora_active(self, lora_params: Optional[Dict]) -> bool:
         """Return True when lora_params carries routed-expert MoE LoRA tensors
         for this layer, meaning run_moe would fuse a LoRA delta.
         """
         if not lora_params or self.layer_idx is None:
             return False
+        if lora_params.get("use_cuda_graph_mode", False):
+            cuda_graph_params = lora_params.get("cuda_graph_params")
+            if cuda_graph_params is None:
+                return False
+            layer_module2key = getattr(cuda_graph_params, "layer_module2key", {})
+            return any(
+                (self.layer_idx, int(LoraModuleType.from_string(name))) in layer_module2key
+                for name in MOE_LORA_MODULE_NAMES
+            )
         layer_params = lora_params.get(self.layer_idx, {})
         if not layer_params:
             return False

🤖 Prompt for 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.

In `@tensorrt_llm/_torch/modules/fused_moe/fused_moe_cutlass.py` around lines 516
- 523, The `_moe_lora_active` method currently only checks `lora_params` to
determine if MoE LoRA is active, but it should also check `cuda_graph_params`
since `_extract_moe_lora_tensors` handles use_cuda_graph_mode without a
per-layer eager dict. Update the method to check both `lora_params` and
`cuda_graph_params` dictionaries, returning True if any MoE LoRA module types
are found in either parameter dictionary for the current layer index.

🤖 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 `@tensorrt_llm/_torch/modules/fused_moe/fused_moe_cutlass.py`:
- Around line 413-416: The targets set constructed from lora_config's
lora_target_modules (at line 413) uses raw strings without normalization, but
the has_moe_lora_targets check at line 398 normalizes strings to lowercase for
comparison. This inconsistency can cause mixed-case configs to behave
incorrectly. Normalize the targets by converting each string in
lora_target_modules to lowercase when constructing the targets set, ensuring
consistency with the normalization performed in has_moe_lora_targets.

In `@tensorrt_llm/_torch/modules/fused_moe/ops/moe_op_cutlass.py`:
- Around line 207-226: The non-min-latency code path for run_moe is only
appending 10 arguments but the C++ binding expects 17 trailing arguments in
total. Update the run_moe_args list extension in the if not min_latency_mode
block to include all 17 required positional arguments by adding 7 more None
values and 0 values to match the full schema expected by the TorchBind binding,
even though all LoRA-related arguments should be None or 0 since LoRA is not
applied in compute_moe.

In `@tests/unittest/llmapi/test_llm_pytorch.py`:
- Around line 1200-1202: Update the docstring comment that describes the test's
expectation about eager and CUDA-graph mode outputs to clarify that they are not
expected to be bit-identical, aligning it with the actual test assertion logic
that only verifies general correctness rather than exact token-for-token
matching. The comment currently contradicts what the test actually validates, so
revise the description to accurately reflect that the outputs may differ
slightly due to floating-point precision while still being functionally correct.

---

Outside diff comments:
In `@cpp/tensorrt_llm/kernels/cutlass_kernels/moe_gemm/moe_kernels.cu`:
- Around line 3735-3767: Add validation of the grouped-GEMM metadata (dp) before
constructing the MoeLoraExpandModule objects fc1_mod, fc2_mod, and gated_mod.
Validate that dp.dtype_bytes is valid and that each module's dimensions (dim_a,
dim_b) and device buffers (ranks_src_dev, ptrs_src_dev, permuted_ranks_dev,
permuted_ptrs_dev) are properly populated and non-null for fc1 and fc2. For
gated_mod, perform the same validation only when is_gated_activation is true.
This validation should occur before constructing any of the expand modules and
before calling launchMoeLoraPointerExpand to prevent incorrect expert weight
address computation or device illegal memory access.

In `@tensorrt_llm/_torch/modules/fused_moe/fused_moe_cutlass.py`:
- Around line 516-523: The `_moe_lora_active` method currently only checks
`lora_params` to determine if MoE LoRA is active, but it should also check
`cuda_graph_params` since `_extract_moe_lora_tensors` handles
use_cuda_graph_mode without a per-layer eager dict. Update the method to check
both `lora_params` and `cuda_graph_params` dictionaries, returning True if any
MoE LoRA module types are found in either parameter dictionary for the current
layer index.

🪄 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: a89651ca-07af-482e-8ac9-9fbeaeb18325

📥 Commits

Reviewing files that changed from the base of the PR and between eddaa3a and 653d04f.

📒 Files selected for processing (21)

• cpp/tensorrt_llm/kernels/cutlass_kernels/include/moe_kernels.h
• cpp/tensorrt_llm/kernels/cutlass_kernels/include/moe_lora_grouped_gemm.h
• cpp/tensorrt_llm/kernels/cutlass_kernels/moe_gemm/moe_kernels.cu
• cpp/tensorrt_llm/kernels/cutlass_kernels/moe_gemm/moe_lora_pointer_expand.cu
• cpp/tensorrt_llm/thop/moeOp.cpp
• cpp/tests/unit_tests/kernels/moeLoraProblemBuilderTest.cu
• docs/source/features/lora.md
• tensorrt_llm/_torch/custom_ops/torch_custom_ops.py
• tensorrt_llm/_torch/modules/fused_moe/fused_moe_cutlass.py
• tensorrt_llm/_torch/modules/fused_moe/moe_scheduler.py
• tensorrt_llm/_torch/modules/fused_moe/ops/moe_op_cutlass.py
• tensorrt_llm/_torch/peft/lora/cuda_graph_lora_manager.py
• tensorrt_llm/_torch/peft/lora/cuda_graph_lora_params.py
• tensorrt_llm/_torch/peft/lora/layer.py
• tensorrt_llm/_torch/peft/lora/moe_layout.py
• tensorrt_llm/_torch/peft/lora/validation.py
• tests/unittest/_torch/lora/test_moe_lora_cuda_graph_params.py
• tests/unittest/_torch/lora/test_moe_lora_extract.py
• tests/unittest/_torch/lora/test_moe_lora_grouped_gemm.py
• tests/unittest/_torch/lora/test_moe_lora_op.py
• tests/unittest/llmapi/test_llm_pytorch.py

[tensorrt-cicd]

PR_Github #55077 [ run ] triggered by Bot. Commit: 653d04f Link to invocation

[tensorrt-cicd]

PR_Github #55077 [ run ] completed with state FAILURE. Commit: 653d04f
/LLM/main/L0_MergeRequest_PR pipeline #44064 completed with status: 'FAILURE'

CI Report

⚠️ Action Required:

• Please check the failed tests and fix your PR
• If you cannot view the failures, ask the CI triggerer to share details
• Once fixed, request an NVIDIA team member to trigger CI again

CI Agent Failure Analysis

Link to invocation

[byshiue]

LGTM

[Shixiaowei02]

LGTM for the document changes.

[brb-nv]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #56241 [ run ] triggered by Bot. Commit: fb7c4bc Link to invocation

[tensorrt-cicd]

PR_Github #56241 [ run ] completed with state FAILURE. Commit: fb7c4bc
/LLM/main/L0_MergeRequest_PR pipeline #45101 completed with status: 'FAILURE'

CI Report

⚠️ Action Required:

• Please check the failed tests and fix your PR
• If you cannot view the failures, ask the CI triggerer to share details
• Once fixed, request an NVIDIA team member to trigger CI again

CI Agent Failure Analysis

Link to invocation

[tensorrt-cicd]

PR_Github #56416 [ run ] triggered by Bot. Commit: 0733e42 Link to invocation

[tensorrt-cicd]

PR_Github #56416 [ run ] completed with state FAILURE. Commit: 0733e42
/LLM/main/L0_MergeRequest_PR pipeline #45259 completed with status: 'FAILURE'

CI Report

⚠️ Action Required:

• Please check the failed tests and fix your PR
• If you cannot view the failures, ask the CI triggerer to share details
• Once fixed, request an NVIDIA team member to trigger CI again

CI Agent Failure Analysis

Link to invocation

[brb-nv]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #56681 [ run ] triggered by Bot. Commit: 0733e42 Link to invocation

[tensorrt-cicd]

PR_Github #56681 [ run ] completed with state SUCCESS. Commit: 0733e42
/LLM/main/L0_MergeRequest_PR pipeline #45503 completed with status: 'FAILURE'

CI Report

⚠️ Action Required:

• Please check the failed tests and fix your PR
• If you cannot view the failures, ask the CI triggerer to share details
• Once fixed, request an NVIDIA team member to trigger CI again

CI Agent Failure Analysis

Link to invocation

[brb-nv]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #56731 [ run ] triggered by Bot. Commit: 71860d2 Link to invocation

[brb-nv]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #56758 [ run ] triggered by Bot. Commit: 94b1a10 Link to invocation

[tensorrt-cicd]

PR_Github #56731 [ run ] completed with state ABORTED. Commit: 71860d2

Link to invocation

[tensorrt-cicd]

PR_Github #56758 [ run ] completed with state SUCCESS. Commit: 94b1a10
/LLM/main/L0_MergeRequest_PR pipeline #45576 completed with status: 'FAILURE'

CI Report

⚠️ Action Required:

• Please check the failed tests and fix your PR
• If you cannot view the failures, ask the CI triggerer to share details
• Once fixed, request an NVIDIA team member to trigger CI again

CI Agent Failure Analysis

Link to invocation

[brb-nv]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #56982 [ run ] triggered by Bot. Commit: 94b1a10 Link to invocation

[tensorrt-cicd]

PR_Github #56982 [ run ] completed with state SUCCESS. Commit: 94b1a10
/LLM/main/L0_MergeRequest_PR pipeline #45781 completed with status: 'FAILURE'

CI Report

⚠️ Action Required:

• Please check the failed tests and fix your PR
• If you cannot view the failures, ask the CI triggerer to share details
• Once fixed, request an NVIDIA team member to trigger CI again

CI Agent Failure Analysis

Link to invocation

[brb-nv]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #57025 [ run ] triggered by Bot. Commit: 94b1a10 Link to invocation

[brb-nv]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #57027 [ run ] triggered by Bot. Commit: 94b1a10 Link to invocation

[tensorrt-cicd]

PR_Github #57025 [ run ] completed with state ABORTED. Commit: 94b1a10

Link to invocation

[tensorrt-cicd]

PR_Github #57027 [ run ] completed with state SUCCESS. Commit: 94b1a10
/LLM/main/L0_MergeRequest_PR pipeline #45822 completed with status: 'SUCCESS'

CI Report

Link to invocation