[None][chore] BREAKING: Test KV cache manager v2 as the default by yizhang-nv · Pull Request #17495 · NVIDIA/TensorRT-LLM

yizhang-nv · 2026-08-11T09:05:08Z

Description

This draft PR changes KvCacheConfig.use_kv_cache_manager_v2 from the "auto" default to True and updates the corresponding declaration test.

The purpose is to run the full pre-merge CI suite and identify failures or compatibility gaps that are currently avoided by the automatic fallback logic. This is an experimental PR and is not intended for merge until the CI fallout has been reviewed.

Test Coverage

python3 scripts/generate_llm_args_golden_manifest.py passed; the manifest did not change because it does not record field defaults.
pytest -s tests/unittest/llmapi/test_llm_args.py: 278 passed, 8 failed. Seven failures are direct fallout from bypassing the "auto" fallback: the no-preference fallback, three incompatible disaggregated-serving cases, two DEFAULT-backend environment routes, and the no-checkpoint-loader path. The remaining test_runtime_sizes failure came from the container's unrelated Triton API mismatch (triton_kernels.tensor.bitwidth).
Two additional targeted tests failed as expected: Kimi without a V2 preference now selects V2 instead of the mixed manager, and two-model GPT-OSS Eagle3 now raises the explicit-V2 incompatibility error instead of falling back to V1.
Full pre-merge CI will be triggered with fail-fast disabled so failures across all stages can be observed.

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.

yizhang-nv · 2026-08-11T09:05:19Z

/bot run --disable-fail-fast

tensorrt-cicd · 2026-08-11T09:11:38Z

PR_Github #65256 [ run ] triggered by Bot. Commit: 2595749 Link to invocation

tensorrt-cicd · 2026-08-11T15:53:49Z

PR_Github #65256 [ run ] completed with state SUCCESS. Commit: 2595749
/LLM/main/L0_MergeRequest_PR pipeline #53032 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

yizhang-nv · 2026-08-13T07:00:36Z

/bot run --disable-fail-fast

yizhang-nv · 2026-08-14T07:17:16Z

/bot run --disable-fail-fast

tensorrt-cicd · 2026-08-14T07:25:14Z

PR_Github #66215 [ run ] triggered by Bot. Commit: e289dd8 Link to invocation

tensorrt-cicd · 2026-08-14T14:38:23Z

PR_Github #66215 [ run ] completed with state SUCCESS. Commit: e289dd8
/LLM/main/L0_MergeRequest_PR pipeline #53884 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

yizhang-nv · 2026-08-17T06:33:01Z

/bot run --disable-fail-fast --disable-reuse-test --stage-list "A10-PyTorch-1, A10-PyTorch-2, A10-PyTorch-3, A30-AutoDeploy-1, CPU-Generic-arm-1, CPU-Generic-x86-1, DGX_B200-AutoDeploy-1, DGX_B200-PyTorch-1, DGX_B200-PyTorch-2, DGX_B200-PyTorch-3, DGX_B200-PyTorch-4, DGX_B200-PyTorch-5, DGX_B200-PyTorch-6, DGX_B200-PyTorch-7, DGX_B200-PyTorch-8, DGX_B200-PyTorch-9, DGX_H100-PyTorch-1, DGX_H100-PyTorch-2, DGX_H100-PyTorch-3, DGX_H100-PyTorch-4, DGX_H100-PyTorch-5, DGX_H100-PyTorch-6, H100_PCIe-AutoDeploy-1, H100_PCIe-PyTorch-Ray-1, DGX_B200-PyTorch-Post-Merge-1, DGX_B200-PyTorch-Post-Merge-2, DGX_H100-PyTorch-Post-Merge-1, DGX_H100-PyTorch-Post-Merge-2"

tensorrt-cicd · 2026-08-17T06:39:26Z

PR_Github #66680 [ run ] triggered by Bot. Commit: 1db4b89 Link to invocation

tensorrt-cicd · 2026-08-17T08:35:06Z

PR_Github #66680 [ run ] completed with state SUCCESS. Commit: 1db4b89
/LLM/main/L0_MergeRequest_PR pipeline #54296 (Partly Tested) 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

yizhang-nv · 2026-08-17T15:28:15Z

/bot run --disable-fail-fast --disable-reuse-test --extra-stage "Post-Merge"

tensorrt-cicd · 2026-08-17T15:35:09Z

PR_Github #66784 [ ] completed with state FAILURE. Commit: ``

Link to invocation

yizhang-nv · 2026-08-18T02:57:19Z

/bot help

github-actions · 2026-08-18T02:57:28Z

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.

Details

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) --high-priority]

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. 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.

--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. Requires the ci: full pre-merge approved label on the PR (ask a member of NVIDIA/trt-llm-ci-approvers). 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. Requires the ci: full pre-merge approved label on the PR (ask a member of NVIDIA/trt-llm-ci-approvers).

--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.

--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.

--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 the stage-list parameter to access the appropriate container environment. Note: Does NOT update GitHub check status.

--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.

kill

kill

Kill all running builds associated with pull request.

skip

skip --comment COMMENT

Skip 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-pipeline

Reuse 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.

yizhang-nv · 2026-08-18T02:59:06Z

/bot run --post-merge --disable-fail-fast --disable-reuse-test

tensorrt-cicd · 2026-08-18T03:04:41Z

PR_Github #66943 [ run ] triggered by Bot. Commit: 39ed067 Link to invocation

tensorrt-cicd · 2026-08-18T08:31:04Z

PR_Github #66943 [ run ] completed with state FAILURE. Commit: 39ed067
/LLM/main/L0_MergeRequest_PR pipeline #54495 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

yizhang-nv · 2026-08-18T13:43:24Z

/bot run --disable-fail-fast

tensorrt-cicd · 2026-08-18T13:49:09Z

PR_Github #67082 [ run ] triggered by Bot. Commit: 910065d Link to invocation

tensorrt-cicd · 2026-08-18T19:04:19Z

PR_Github #67082 [ run ] completed with state FAILURE. Commit: 910065d
/LLM/main/L0_MergeRequest_PR pipeline #54620 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

yizhang-nv · 2026-08-19T02:51:23Z

/bot run --disable-fail-fast

nv-xtf · 2026-09-17T06:10:31Z

            echo "  --tp_size N          Tensor parallel size (default: 1)"
            echo "  --transceiver_backend UCX|NIXL  Cache-transceiver backend (default: NIXL)"
-            echo "  --transceiver_runtime CPP|PYTHON  Cache transceiver runtime (default: CPP)"
+            echo "  --transceiver_runtime CPP|PYTHON  Cache transceiver runtime (default: PYTHON)"


When --transceiver_runtime CPP is selected, should the generated configuration also set kv_cache_config.use_kv_cache_manager_v2: false? The script currently omits this field, and this PR changes its default from auto to True, bypassing the previous V1 fallback for CPP and pairing the C++ transceiver with a V2 manager. Is there another compatibility path that handles this?

tensorrt-cicd · 2026-09-17T11:07:54Z

PR_Github #74021 [ run ] completed with state SUCCESS. Commit: 3be5469
/LLM/main/L0_MergeRequest_PR pipeline #60868 (Partly Tested) 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

Signed-off-by: Yi Zhang <187001205+yizhang-nv@users.noreply.github.com>

yizhang-nv · 2026-09-18T07:44:57Z

/bot run --post-merge --disable-multi-gpu-test --disable-fail-fast

tensorrt-cicd · 2026-09-18T07:52:51Z

PR_Github #74372 [ run ] triggered by Bot. Commit: f518b8b Link to invocation

tensorrt-cicd · 2026-09-18T20:10:21Z

PR_Github #74372 [ run ] completed with state SUCCESS. Commit: f518b8b
/LLM/main/L0_MergeRequest_PR pipeline #61184 (Partly Tested) 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

Enable KV cache manager V2 and the Python cache transceiver by default, and migrate supported examples and tests to the V2 path. Keep explicit V1 and C++ settings only for legacy transport, compatibility, and dedicated coverage. Signed-off-by: Yi Zhang <187001205+yizhang-nv@users.noreply.github.com>

Signed-off-by: Yi Zhang <187001205+yizhang-nv@users.noreply.github.com>

yizhang-nv · 2026-09-27T06:35:04Z

/bot run --post-merge --add-multi-gpu-test --disable-fail-fast

tensorrt-cicd · 2026-09-27T06:43:11Z

PR_Github #75481 [ run ] triggered by Bot. Commit: a27e923 Link to invocation

tensorrt-cicd · 2026-09-27T10:26:03Z

PR_Github #75481 [ run ] completed with state SUCCESS. Commit: a27e923
/LLM/main/L0_MergeRequest_PR pipeline #62215 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

github-actions Bot assigned yizhang-nv Aug 11, 2026

yizhang-nv changed the title ~~[None][chore] Test KV cache manager v2 as the default~~ [None][chore] BREAKING: Test KV cache manager v2 as the default Aug 11, 2026

yizhang-nv added the api-breaking Accepted LLM API contract change that is backwards-incompatible label Aug 11, 2026

yizhang-nv force-pushed the codex/kvcm-v2-default-true-ci branch from 8a8cd95 to e289dd8 Compare August 14, 2026 07:16

yiqingy0 added the ci: post-merge approved Approved by TRT-LLM CI approvers for broad post-merge CI requests label Aug 17, 2026

yizhang-nv force-pushed the codex/kvcm-v2-default-true-ci branch 2 times, most recently from 6e62635 to 39ed067 Compare August 17, 2026 15:27

yiqingy0 added the ci: full pre-merge approved label Aug 18, 2026

yizhang-nv force-pushed the codex/kvcm-v2-default-true-ci branch from 39ed067 to 9c5c909 Compare August 18, 2026 09:38

nv-xtf reviewed Sep 17, 2026

View reviewed changes

yizhang-nv added a commit that referenced this pull request Sep 17, 2026

[#17495][fix] Share draft length updates with warmup (#19204)

33193c2

Signed-off-by: Yi Zhang <187001205+yizhang-nv@users.noreply.github.com>

yizhang-nv force-pushed the codex/kvcm-v2-default-true-ci branch from 3be5469 to b0f1682 Compare September 17, 2026 12:50

yizhang-nv mentioned this pull request Sep 18, 2026

[None][fix] Enable KVCM V2 by default for KimiLinear #19387

Open

7 tasks

yizhang-nv force-pushed the codex/kvcm-v2-default-true-ci branch from b0f1682 to f518b8b Compare September 18, 2026 07:44

yizhang-nv force-pushed the codex/kvcm-v2-default-true-ci branch from f518b8b to afbe888 Compare September 20, 2026 15:20

yizhang-nv added 16 commits September 26, 2026 23:30

[None][test] Skip known KVCM V2 failures

2da184a

Signed-off-by: Yi Zhang <187001205+yizhang-nv@users.noreply.github.com>

[None][test] Skip latest KVCM V2 failures

64bb58c

Signed-off-by: Yi Zhang <187001205+yizhang-nv@users.noreply.github.com>

[None][test] Skip concurrent KVCM V2 OOM

a5fb9b2

Signed-off-by: Yi Zhang <187001205+yizhang-nv@users.noreply.github.com>

[None][test] Re-enable fixed and rerunnable KVCM V2 tests

3db41cb

Signed-off-by: Yi Zhang <187001205+yizhang-nv@users.noreply.github.com>

test: remove redundant V1 disaggregation coverage

ca1d4fc

Signed-off-by: Yi Zhang <187001205+yizhang-nv@users.noreply.github.com>

[None][fix] Use default KV cache manager for aggregated DeepSeek V3.2

d3929e4

Signed-off-by: Yi Zhang <187001205+yizhang-nv@users.noreply.github.com>

[None][test] Pin v1 disagg metrics and update disagg coverage notes

7caf188

Signed-off-by: Yi Zhang <187001205+yizhang-nv@users.noreply.github.com>

[None][chore] Remove local disagg review documents

fef8957

Signed-off-by: Yi Zhang <187001205+yizhang-nv@users.noreply.github.com>

[None][test] Waive single-GPU failures pending KVCM V2 fixes

3abf01a

Signed-off-by: Yi Zhang <187001205+yizhang-nv@users.noreply.github.com>

[None][test] Remove stale V1 perf overrides and redundant test

74da14b

Signed-off-by: Yi Zhang <187001205+yizhang-nv@users.noreply.github.com>

[None][test] Re-enable Qwen3 LoRA after merged V2 test fix

172bc6b

Signed-off-by: Yi Zhang <187001205+yizhang-nv@users.noreply.github.com>

[None][test] Re-enable dynamic draft after merged warmup fix

3a18425

Signed-off-by: Yi Zhang <187001205+yizhang-nv@users.noreply.github.com>

[None][test] Update KVCM V2 CI coverage and waivers

5905384

Signed-off-by: Yi Zhang <187001205+yizhang-nv@users.noreply.github.com>

[None][test] Unwaive KVCM V2 cases except initialization OOM

11d06ea

Signed-off-by: Yi Zhang <187001205+yizhang-nv@users.noreply.github.com>

[None][test] Initialize max beam width in KV cache budget fixture

a27e923

Signed-off-by: Yi Zhang <187001205+yizhang-nv@users.noreply.github.com>

yizhang-nv force-pushed the codex/kvcm-v2-default-true-ci branch from afbe888 to a27e923 Compare September 27, 2026 06:34

Conversation

yizhang-nv commented Aug 11, 2026

Description

Test Coverage

PR Checklist

GitHub Bot Help

Uh oh!

yizhang-nv commented Aug 11, 2026

Uh oh!

tensorrt-cicd commented Aug 11, 2026

Uh oh!

tensorrt-cicd commented Aug 11, 2026

Uh oh!

yizhang-nv commented Aug 13, 2026

Uh oh!

yizhang-nv commented Aug 14, 2026

Uh oh!

tensorrt-cicd commented Aug 14, 2026

Uh oh!

tensorrt-cicd commented Aug 14, 2026

Uh oh!

yizhang-nv commented Aug 17, 2026

Uh oh!

tensorrt-cicd commented Aug 17, 2026

Uh oh!

tensorrt-cicd commented Aug 17, 2026

Uh oh!

yizhang-nv commented Aug 17, 2026

Uh oh!

tensorrt-cicd commented Aug 17, 2026

Uh oh!

yizhang-nv commented Aug 18, 2026

Uh oh!

github-actions Bot commented Aug 18, 2026

GitHub Bot Help

kill

skip

reuse-pipeline

Uh oh!

yizhang-nv commented Aug 18, 2026

Uh oh!

tensorrt-cicd commented Aug 18, 2026

Uh oh!

tensorrt-cicd commented Aug 18, 2026

Uh oh!

yizhang-nv commented Aug 18, 2026

Uh oh!

tensorrt-cicd commented Aug 18, 2026

Uh oh!

tensorrt-cicd commented Aug 18, 2026

Uh oh!

yizhang-nv commented Aug 19, 2026

Uh oh!

nv-xtf Sep 17, 2026

Choose a reason for hiding this comment

Uh oh!

tensorrt-cicd commented Sep 17, 2026

Uh oh!

yizhang-nv commented Sep 18, 2026

Uh oh!

tensorrt-cicd commented Sep 18, 2026

Uh oh!

tensorrt-cicd commented Sep 18, 2026

Uh oh!

yizhang-nv commented Sep 27, 2026

Uh oh!

tensorrt-cicd commented Sep 27, 2026

Uh oh!

tensorrt-cicd commented Sep 27, 2026

Uh oh!

This branch has not been deployed

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants