[TRTLLM-14810][chore] Catch-up merge: main into feat/kimi_k3

.claude/skills/exec-local-compile/SKILL.md

@@ -45,7 +45,7 @@ git checkout main && git pull
 Run the build command (**incremental by default** — omit `-c`/`--clean` unless explicitly requested or the incremental build fails):
 
 ```bash
-./scripts/build_wheel.py --trt_root /usr/local/tensorrt --benchmarks --use_ccache -a "<arch>" -f --nvtx
+./scripts/build_wheel.py --use_ccache -a "<arch>" -f --nvtx
 ```
 
 Replace `<arch>` with the target GPU architecture (see Architecture Reference below). If not specified by the user, auto-detect from `nvidia-smi`.
@@ -66,8 +66,6 @@ python3 -c "import tensorrt_llm; print(tensorrt_llm.__version__)"
 
 | Flag | Description |
 |------|-------------|
-| `--trt_root /usr/local/tensorrt` | TensorRT installation path (standard in NVIDIA containers) |
-| `--benchmarks` | Build the C++ benchmarks |
 | `-a "<arch>"` | Target GPU architecture(s) |
 | `--nvtx` | Enable NVTX markers for profiling |
 | `--use_ccache` | Use ccache for faster recompilation |

.claude/skills/exec-slurm-compile/SKILL.md

@@ -204,8 +204,6 @@ A successful build ends with a message like `Successfully built tensorrt_llm` or
 
 | Flag | Description |
 |------|-------------|
-| `--trt_root /usr/local/tensorrt` | TensorRT installation path (standard in NVIDIA containers) |
-| `--benchmarks` | Build the C++ benchmarks |
 | `-a "100-real"` | Target architecture — `100` for Blackwell, `90` for Hopper, etc. |
 | `--nvtx` | Enable NVTX markers for profiling |
 | `--no-venv` | Skip virtual environment creation |
@@ -228,7 +226,6 @@ Common architecture values:
 | `sbatch: error: invalid partition` | Verify partition name with `sinfo -s` |
 | `sbatch: error: invalid account` | Check available accounts with `sacctmgr show assoc user=$USER` |
 | Container image not found | Verify the `.sqsh` path exists and is readable |
-| Build fails with missing TensorRT | Ensure `--trt_root` points to the correct path inside the container |
 | Build OOM (out of memory) | Reduce parallelism with `-j <N>` flag to `build_wheel.py` |
 | `srun: error: Unable to create step` | The node may lack enroot/pyxis — check with cluster admin |
 | Job stuck in `PD` state | Check `squeue -j <id> -o %R` for the reason (e.g., resource limits, priority) |

.claude/skills/exec-slurm-compile/scripts/compile.sh

@@ -19,7 +19,7 @@
 # Usage: compile.sh <repo_dir> [build_wheel_args...]
 #
 # Default build_wheel.py flags:
-#   --trt_root /usr/local/tensorrt --benchmarks -a "100-real" --nvtx --no-venv
+#   -a "100-real" --nvtx --no-venv
 # Any extra arguments after repo_dir are forwarded to build_wheel.py,
 # overriding the defaults above.
 
@@ -36,8 +36,6 @@ if [[ $# -gt 0 ]]; then
 else
     echo "[compile.sh] Running default build command"
     python3 ./scripts/build_wheel.py \
-        --trt_root /usr/local/tensorrt \
-        --benchmarks \
         -a "100-real" \
         --nvtx
 fi

.coderabbit.yaml

@@ -22,7 +22,26 @@ reviews:
   auto_title_placeholder: '@coderabbitai title'
   auto_title_instructions: 'Format: "[<category>] <title>". Category must be one of: fix, feat, doc, infra, style, refactor, perf, test, chore, revert. Enclose the category in square brackets. Title should be concise (<= 60 chars). Example: "[feat] Add logit_bias support".'
   commit_status: false
-  collapse_walkthrough: true
+  high_level_summary_instructions: |
+    Always produce two review sections in the summary:
+
+    **Dev Engineer Review**
+    Review all changes for correctness and consistency, including:
+    - Code changes: correctness, performance, API consistency (CODING_GUIDELINES.md), error handling, regressions.
+    - Config files: valid values, no typos, consistency with related configs, no unintended scope changes.
+    - Test list files (test-db/, qa/, waives.txt): correct format, valid test paths, appropriate bug references, no duplicates.
+
+    **QA Engineer Review**
+    Always include this section when any files under tests/ are touched.
+    For test-list-only changes (only tests/integration/test_lists/ files):
+      - List which test-db/ or qa/ files were modified and what entries were added or removed.
+      - Verdict: "needs follow-up" if CBTS coverage data is unavailable, otherwise "sufficient" or "insufficient".
+    For test-code changes (files outside tests/integration/test_lists/):
+      - List test functions added, modified, or removed.
+      - State whether each is covered in tests/integration/test_lists/ (test-db/ for CI, qa/ for manual QA).
+      - Verdict: sufficient, insufficient, or needs follow-up.
+    If no test files are touched, write "No test changes."
+  collapse_walkthrough: false
   assess_linked_issues: true
   related_issues: true
   related_prs: true
@@ -31,15 +50,27 @@ reviews:
   poem: false
   review_status: false
   auto_review:
-    auto_incremental_review: false
+    auto_incremental_review: true
     drafts: false
     base_branches: ["main", "release/.+"]
   path_instructions:
     - path: "tests/**"
       instructions: |
         Act as a QA engineer reviewing test changes and coverage for TensorRT-LLM.
-        Keep feedback actionable: suggest concrete list file names and whether
-        coverage is sufficient, insufficient, or needs follow-up outside the PR.
+        Always produce a test coverage summary, even if no issues are found.
+
+        If the change touches ONLY files under tests/integration/test_lists/ (no test-code changes):
+        - Report which test-db/ or qa/ list files were modified and what entries were added or removed.
+        - Do NOT require changed test functions for this path.
+        - Use verdict "needs follow-up" when cbts_touchmap.sqlite or a CBTS coverage report is unavailable
+          to confirm the impacted test scope; otherwise use "sufficient" or "insufficient".
+
+        If the change includes test-code files (outside tests/integration/test_lists/), the summary must include:
+        1. Which test functions were added, modified, or removed.
+        2. Whether each changed test is listed in the appropriate test list files under
+           tests/integration/test_lists/ (test-db/ for CI, qa/ for manual QA).
+        3. A coverage verdict: sufficient, insufficient, or needs follow-up.
+        Keep feedback actionable: reference concrete list file names when suggesting additions.
     - path: "tests/integration/test_lists/qa/**"
       instructions: |
         Files here are manually-triggered QA perf/regression lists, maintained

.gitattributes

@@ -16,4 +16,5 @@ docs/source/blogs/media/tech_blog10_full_strategy_performance.png filter=lfs dif
 docs/source/blogs/media/tech_blog10_context_wait_performance.png  filter=lfs diff=lfs merge=lfs -text
 cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/kernelMetaInfo_cubin.cpp filter=lfs diff=lfs merge=lfs -text
 cpp/tensorrt_llm/kernels/decoderMaskedMultiheadAttention/cubin/xqa_kernel_cubin.cpp filter=lfs diff=lfs merge=lfs -text
-tensorrt_llm/_torch/visual_gen/cute_dsl_kernels/blackwell/attention/cubins/*/*/*.so filter=lfs diff=lfs merge=lfs -text
+docs/source/blogs/media/tech_blog26_deepseek_v4_hybrid_attention.png filter=lfs diff=lfs merge=lfs -text
+docs/source/blogs/media/tech_blog26_deepseek_v4_mhc_moe.png filter=lfs diff=lfs merge=lfs -text

.github/CODEOWNERS

@@ -7,7 +7,7 @@
 #   Infra                    			  @NVIDIA/trt-llm-infra-devs
 #   Agent config             			  @NVIDIA/trt-llm-agent-devs
 #   Docs / Examples          			  @NVIDIA/trt-llm-doc-owners
-#   QA                       			  @NVIDIA/trt-llm-qa
+#   QA                       			  @NVIDIA/trt-llm-qa / qa-perf / qa-function
 #   Runtime                  			  @NVIDIA/trt-llm-runtime-devs
 #   Kernels - Misc           			  @NVIDIA/trt-llm-kernels-devs
 #   Models                   			  @NVIDIA/trt-llm-models-devs
@@ -31,14 +31,12 @@
 /tensorrt_llm/commands/eval.py @NVIDIA/trt-llm-devs
 /tensorrt_llm/evaluate @NVIDIA/trt-llm-devs
 /tensorrt_llm/tools @NVIDIA/trt-llm-devs
-/tests/integration/test_lists/test-db @NVIDIA/trt-llm-devs @NVIDIA/trt-llm-qa @NVIDIA/trt-llm-infra-devs
-/tests/integration/test_lists/waives.txt @NVIDIA/trt-llm-devs @NVIDIA/trt-llm-qa @NVIDIA/trt-llm-infra-devs
+/tests/integration/test_lists/test-db @NVIDIA/trt-llm-devs @NVIDIA/trt-llm-qa-function @NVIDIA/trt-llm-infra-devs
+/tests/integration/test_lists/waives.txt @NVIDIA/trt-llm-devs @NVIDIA/trt-llm-qa-function @NVIDIA/trt-llm-infra-devs
 /tests/test_common @NVIDIA/trt-llm-devs
 /tests/unittest @NVIDIA/trt-llm-devs
 
 # ===== TensorRT backend (will be deprecated soon) — also on the trt-llm-devs fallback =====
-/cpp/include/tensorrt_llm/plugins @NVIDIA/trt-llm-devs
-/cpp/tensorrt_llm/plugins @NVIDIA/trt-llm-devs
 /tensorrt_llm/builder.py @NVIDIA/trt-llm-devs
 /tensorrt_llm/commands/build.py @NVIDIA/trt-llm-devs
 /tensorrt_llm/commands/prune.py @NVIDIA/trt-llm-devs
@@ -95,6 +93,8 @@
 # ===== QA =====
 /tests/integration/defs @NVIDIA/trt-llm-devs @NVIDIA/trt-llm-qa @NVIDIA/trt-llm-infra-devs
 /tests/integration/test_lists/qa @NVIDIA/trt-llm-qa
+/tests/integration/test_lists/qa/llm_perf_* @NVIDIA/trt-llm-qa-perf
+/tests/integration/test_lists/qa/llm_function_* @NVIDIA/trt-llm-qa-function
 
 # ===== RUNTIME =====
 /cpp/include/tensorrt_llm/batch_manager @NVIDIA/trt-llm-runtime-devs
@@ -105,7 +105,6 @@
 /cpp/tensorrt_llm/batch_manager @NVIDIA/trt-llm-runtime-devs
 /cpp/tensorrt_llm/common @NVIDIA/trt-llm-runtime-devs
 /cpp/tensorrt_llm/executor @NVIDIA/trt-llm-runtime-devs
-/cpp/tensorrt_llm/executor_worker @NVIDIA/trt-llm-runtime-devs
 /cpp/tensorrt_llm/layers @NVIDIA/trt-llm-runtime-devs
 /cpp/tensorrt_llm/nanobind @NVIDIA/trt-llm-runtime-devs
 /cpp/tensorrt_llm/runtime @NVIDIA/trt-llm-runtime-devs
@@ -169,13 +168,12 @@
 /examples/llm-api/quickstart_multimodal.py @NVIDIA/trt-llm-models-devs @NVIDIA/trt-llm-doc-owners
 /examples/models @NVIDIA/trt-llm-models-devs @NVIDIA/trt-llm-doc-owners
 /examples/serve/*multimodal* @NVIDIA/trt-llm-models-devs @NVIDIA/trt-llm-doc-owners
-/scripts/build_cpp_examples.py @NVIDIA/trt-llm-models-devs
 /scripts/generate_config_database_tests.py @NVIDIA/trt-llm-models-devs @NVIDIA/trt-llm-doc-owners
 /scripts/generate_config_table.py @NVIDIA/trt-llm-models-devs @NVIDIA/trt-llm-doc-owners
 /tensorrt_llm/_torch/models @NVIDIA/trt-llm-models-devs
 /tensorrt_llm/_torch/modules/mamba @NVIDIA/trt-llm-models-devs
 /tensorrt_llm/quantization @NVIDIA/trt-llm-models-devs
-/tests/integration/defs/accuracy/test_llm_api_pytorch_multimodal.py @NVIDIA/trt-llm-models-devs @NVIDIA/trt-llm-qa
+/tests/integration/defs/accuracy/test_llm_api_pytorch_multimodal.py @NVIDIA/trt-llm-models-devs @NVIDIA/trt-llm-qa-function
 /tests/unittest/_torch/modeling @NVIDIA/trt-llm-models-devs
 /tests/unittest/_torch/models @NVIDIA/trt-llm-models-devs
 /tests/unittest/_torch/modules/mamba @NVIDIA/trt-llm-models-devs
@@ -212,6 +210,7 @@
 /cpp/tensorrt_llm/batch_manager/blockKey* @NVIDIA/trt-llm-kv-cache-manager-devs
 /cpp/tensorrt_llm/batch_manager/evictionPolicy* @NVIDIA/trt-llm-kv-cache-manager-devs
 /cpp/tensorrt_llm/batch_manager/kvCache* @NVIDIA/trt-llm-kv-cache-manager-devs
+/cpp/tensorrt_llm/batch_manager/kv_cache_manager_v2 @NVIDIA/trt-llm-kv-cache-manager-devs
 /cpp/tensorrt_llm/nanobind/batch_manager/kvCacheManager* @NVIDIA/trt-llm-kv-cache-manager-devs
 /cpp/tests/unit_tests/batch_manager/blockKey* @NVIDIA/trt-llm-kv-cache-manager-devs
 /cpp/tests/unit_tests/batch_manager/evictionPolicy* @NVIDIA/trt-llm-kv-cache-manager-devs
@@ -246,9 +245,9 @@
 /tensorrt_llm/disaggregated_params.py @NVIDIA/trt-llm-disagg-devs
 /tensorrt_llm/serve/openai_disagg_server.py @NVIDIA/trt-llm-disagg-devs
 # Disagg tests: co-own with the owning team so disagg-devs review disagg-test changes.
-/tests/integration/defs/accuracy/*disagg* @NVIDIA/trt-llm-disagg-devs @NVIDIA/trt-llm-qa
-/tests/integration/defs/disaggregated @NVIDIA/trt-llm-disagg-devs @NVIDIA/trt-llm-qa
-/tests/integration/defs/stress_test/disagg_cancel @NVIDIA/trt-llm-disagg-devs @NVIDIA/trt-llm-qa
+/tests/integration/defs/accuracy/*disagg* @NVIDIA/trt-llm-disagg-devs @NVIDIA/trt-llm-qa-serving
+/tests/integration/defs/disaggregated @NVIDIA/trt-llm-disagg-devs @NVIDIA/trt-llm-qa-serving
+/tests/integration/defs/stress_test/disagg_cancel @NVIDIA/trt-llm-disagg-devs @NVIDIA/trt-llm-qa-serving
 /tests/scripts/perf-sanity/disaggregated @NVIDIA/trt-llm-perf-devs @NVIDIA/trt-llm-disagg-devs
 /tests/scripts/perf/disaggregated @NVIDIA/trt-llm-perf-devs @NVIDIA/trt-llm-disagg-devs
 /tests/unittest/_torch/executor/*disagg* @NVIDIA/trt-llm-runtime-devs @NVIDIA/trt-llm-disagg-devs
@@ -404,7 +403,7 @@
 /scripts/check_auto_deploy_imports.py @NVIDIA/trt-llm-torch-autodeploy-devs
 /scripts/check_model_registry.py @NVIDIA/trt-llm-torch-autodeploy-devs
 /tensorrt_llm/_torch/auto_deploy @NVIDIA/trt-llm-torch-autodeploy-devs
-/tests/integration/defs/accuracy/test_llm_api_autodeploy.py @NVIDIA/trt-llm-torch-autodeploy-devs @NVIDIA/trt-llm-qa
+/tests/integration/defs/accuracy/test_llm_api_autodeploy.py @NVIDIA/trt-llm-torch-autodeploy-devs @NVIDIA/trt-llm-qa-function
 /tests/unittest/_torch/auto_deploy @NVIDIA/trt-llm-torch-autodeploy-devs
 /tests/unittest/auto_deploy @NVIDIA/trt-llm-torch-autodeploy-devs
 /tests/integration/defs/accuracy/test_llm_api_autodeploy.py @NVIDIA/trt-llm-torch-autodeploy-devs @NVIDIA/trt-llm-qa-function
@@ -441,6 +440,7 @@
 /cpp/tensorrt_llm/batch_manager/allocateKvCache.cpp @NVIDIA/trt-llm-kv-cache-manager-devs
 /cpp/tests/unit_tests/batch_manager/kvCacheManagerTest.cpp @NVIDIA/trt-llm-kv-cache-manager-devs
 /cpp/tests/unit_tests/batch_manager/kvCacheUtilsTest.cpp @NVIDIA/trt-llm-kv-cache-manager-devs
+/tensorrt_llm/_torch/attention_backend/sparse/*/cache_manager.py @NVIDIA/trt-llm-kv-cache-manager-devs
 /tensorrt_llm/_torch/pyexecutor/kv_cache_manager_v2.py @NVIDIA/trt-llm-kv-cache-manager-devs
 /tensorrt_llm/_torch/pyexecutor/resource_manager.py @NVIDIA/trt-llm-kv-cache-manager-devs
 /cpp/tensorrt_llm/nanobind/batch_manager/kvCacheManager.h @NVIDIA/trt-llm-kv-cache-manager-devs

.github/workflows/blossom-ci.yml

@@ -48,7 +48,9 @@ jobs:
         "achartier",
         "ajrasane",
         "alec-flowers",
+        "AlessioNetti",
         "alexmsettle",
+        "allisonlim-nv",
         "ameynaik-hub",
         "amirkl94",
         "amitz-nv",
@@ -94,6 +96,7 @@ jobs:
         "chzblych",
         "cjluo-nv",
         "crazydemo",
+        "daichu-nv",
         "DanBlanaru",
         "danielafrimi",
         "davidclark-nv",
@@ -187,6 +190,7 @@ jobs:
         "JunyiXu-nv",
         "JyChang012",
         "kaiyux",
+        "Kambili",
         "kanghui0204",
         "karljang",
         "karthikvetrivel",
@@ -228,6 +232,7 @@ jobs:
         "MatthiasKohl",
         "mayani-nv",
         "meenchen",
+        "MengmSun",
         "mgluhovskoi",
         "mikeiovine",
         "milesial",
@@ -276,6 +281,7 @@ jobs:
         "PerkzZheng",
         "poweiw",
         "pranav-nvidia",
+        "pst2154",
         "qiangxu1996",
         "qiaoxj07",
         "QiJune",
@@ -309,6 +315,7 @@ jobs:
         "shuyixiong",
         "shyeh25",
         "SimengLiu-nv",
+        "siyidNV",
         "sklevtsov-nvidia",
         "StanleySun639",
         "stnie",
@@ -410,6 +417,7 @@ jobs:
         "zhangcl",
         "ZhanruiSunCh",
         "zhaoyangwang-nvidia",
+        "zhaoyuanh-nvidia",
         "zhengd-nv",
         "zhenhuaw-me",
         "zheyuf",

.github/workflows/bot-command.yml

@@ -1,4 +1,4 @@
-# SPDX-FileCopyrightText: Copyright (c) 2024 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-FileCopyrightText: Copyright (c) 2024-2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 # SPDX-License-Identifier: Apache-2.0
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
@@ -52,14 +52,14 @@ jobs:
             "`--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.\n\n" +
             "`--disable-fail-fast ` *(OPTIONAL)* : Disable fail fast on build/tests/infra failures.\n\n" +
             "`--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.\n\n" +
-            "`--stage-list \"A10-PyTorch-1, xxx\"` *(OPTIONAL)* : Only run the specified test stages. Supports wildcard `*` for pattern matching (e.g., `\"*PerfSanity*\"` matches all stages containing PerfSanity). Examples: \"A10-PyTorch-1, xxx\", \"*PerfSanity*\". Note: Does **NOT** update GitHub check status.\n\n" +
+            "`--stage-list \"A10-PyTorch-1, xxx\"` *(OPTIONAL)* : Only run the specified test stages. Supports wildcard `*` for pattern matching (e.g., `\"*PerfSanity*\"` matches all stages containing PerfSanity). Examples: \"A10-PyTorch-1, xxx\", \"*PerfSanity*\". The patterns `\"*\"`, `\"*Post-Merge*\"`, and `\"*PerfSanity*\"`, including equivalent escaped or repeated-star forms and their use in comma-separated lists, require the `ci: post-merge approved` PR label. Note: Does **NOT** update GitHub check status.\n\n" +
             "`--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.\n\n" +
             "`--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.\n\n" +
             "`--only-multi-gpu-test ` *(OPTIONAL)* : Only run the multi-GPU tests. Note: Does **NOT** update GitHub check status.\n\n" +
             "`--disable-multi-gpu-test ` *(OPTIONAL)* : Disable the multi-GPU tests. Note: Does **NOT** update GitHub check status.\n\n" +
             "`--add-multi-gpu-test ` *(OPTIONAL)* : Force run the multi-GPU tests in addition to running L0 pre-merge pipeline.\n\n" +
-            "`--post-merge ` *(OPTIONAL)* : Run the L0 post-merge pipeline instead of the ordinary L0 pre-merge pipeline.\n\n" +
-            "`--extra-stage \"H100_PCIe-TensorRT-Post-Merge-1, xxx\"` *(OPTIONAL)* : Run the ordinary L0 pre-merge pipeline and specified test stages. Supports wildcard `*` for pattern matching. Examples: --extra-stage \"H100_PCIe-TensorRT-Post-Merge-1, xxx\", --extra-stage \"*Post-Merge*\".\n\n" +
+            "`--post-merge ` *(OPTIONAL)* : Run the L0 post-merge pipeline instead of the ordinary L0 pre-merge pipeline. Requires the `ci: post-merge approved` PR label applied by an active member of `NVIDIA/trt-llm-ci-approvers`. The approval label remains in place when new commits are pushed.\n\n" +
+            "`--extra-stage \"H100_PCIe-TensorRT-Post-Merge-1, xxx\"` *(OPTIONAL)* : Run the ordinary L0 pre-merge pipeline and specified test stages. Supports wildcard `*` for pattern matching. Examples: --extra-stage \"H100_PCIe-TensorRT-Post-Merge-1, xxx\", --extra-stage \"*Post-Merge*\". The patterns `\"*\"`, `\"*Post-Merge*\"`, and `\"*PerfSanity*\"`, including equivalent escaped or repeated-star forms and their use in comma-separated lists, require the `ci: post-merge approved` PR label.\n\n" +
             "`--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.\n\n" +
             "`--debug ` *(OPTIONAL)* : **Experimental feature**. Enable access to the CI container for debugging purpose. Note: Specify exactly one stage in the `stage-list` parameter to access the appropriate container environment. Note: Does **NOT** update GitHub check status.\n\n" +
             "`--high-priority ` *(OPTIONAL)* : Run the pipeline with high priority. This option is restricted to authorized users only and will route the job to a high-priority queue.\n\n" +