[TRTLLM-13948][feat] DeepSeek V3.2 TranceiverV2 migration by asfiyab-nvidia · Pull Request #16255 · NVIDIA/TensorRT-LLM

asfiyab-nvidia · 2026-07-10T22:49:58Z

Summary by CodeRabbit

Bug Fixes
- Standardized disaggregated serving and performance configurations to explicitly use the Python transceiver runtime.
- Ensured both generation and context worker configurations consistently use the intended NIXL backend settings.
Tests
- Updated accuracy and performance-sanity coverage to validate the corrected transceiver configuration across supported hardware and workload scenarios.

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-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.

Signed-off-by: Asfiya Baig <asfiyab@nvidia.com>

asfiyab-nvidia · 2026-07-10T22:52:08Z

/bot run --stage "GB200-Multi-GPUs-L0-Perf-Sanity-1" --test "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb200_d
eepseek-v32-fp4_1k1k_con1_ctx1_dep4_gen1_tep8_eplb0_mtp3_ccb-NIXL]"

coderabbitai · 2026-07-10T22:54:32Z

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: 247eb10e-4e4d-4ebe-9e3b-5b057cf4835b

📥 Commits

Reviewing files that changed from the base of the PR and between 2cbdaa0 and 20a1299.

📒 Files selected for processing (47)

tests/integration/defs/accuracy/test_disaggregated_serving.py
tests/scripts/perf-sanity/disaggregated/gb200_deepseek-v32-fp4_1k1k_con1024_ctx1_dep4_gen1_dep32_eplb256_mtp3_ccb-NIXL.yaml
tests/scripts/perf-sanity/disaggregated/gb200_deepseek-v32-fp4_1k1k_con1_ctx1_dep4_gen1_tep8_eplb0_mtp3_ccb-NIXL.yaml
tests/scripts/perf-sanity/disaggregated/gb200_deepseek-v32-fp4_1k1k_con2048_ctx1_dep4_gen1_dep4_eplb0_mtp1_ccb-NIXL.yaml
tests/scripts/perf-sanity/disaggregated/gb200_deepseek-v32-fp4_32k4k_con1_ctx1_dep4_gen1_tep8_eplb0_mtp3_ccb-NIXL.yaml
tests/scripts/perf-sanity/disaggregated/gb200_deepseek-v32-fp4_32k4k_con2048_ctx1_dep4_gen1_dep32_eplb288_mtp1_ccb-NIXL.yaml
tests/scripts/perf-sanity/disaggregated/gb200_deepseek-v32-fp4_32k4k_con256_ctx1_dep4_gen1_dep32_eplb0_mtp3_ccb-NIXL.yaml
tests/scripts/perf-sanity/disaggregated/gb200_deepseek-v32-fp4_32k4k_con256_ctx1_dep8_gen1_dep8_eplb0_mtp0_ccb-NIXL.yaml
tests/scripts/perf-sanity/disaggregated/gb200_deepseek-v32-fp4_8k1k_con1024_ctx1_dep4_gen1_dep32_eplb256_mtp3_ccb-NIXL.yaml
tests/scripts/perf-sanity/disaggregated/gb200_deepseek-v32-fp4_8k1k_con1_ctx1_dep4_gen1_tep8_eplb0_mtp3_ccb-NIXL.yaml
tests/scripts/perf-sanity/disaggregated/gb200_deepseek-v32-fp4_8k1k_con4096_ctx1_dep4_gen1_dep32_eplb256_mtp0_ccb-NIXL.yaml
tests/scripts/perf-sanity/disaggregated/gb300_deepseek-v32-fp4_1k1k_con1024_ctx1_dep4_gen1_dep32_eplb256_mtp3_ccb-NIXL.yaml
tests/scripts/perf-sanity/disaggregated/gb300_deepseek-v32-fp4_1k1k_con1_ctx1_dep4_gen1_tep8_eplb0_mtp3_ccb-NIXL.yaml
tests/scripts/perf-sanity/disaggregated/gb300_deepseek-v32-fp4_1k1k_con2048_ctx1_dep4_gen1_dep4_eplb0_mtp1_ccb-NIXL.yaml
tests/scripts/perf-sanity/disaggregated/gb300_deepseek-v32-fp4_32k4k_con1_ctx1_dep4_gen1_tep8_eplb0_mtp3_ccb-NIXL.yaml
tests/scripts/perf-sanity/disaggregated/gb300_deepseek-v32-fp4_32k4k_con2048_ctx1_dep4_gen1_dep32_eplb288_mtp1_ccb-NIXL.yaml
tests/scripts/perf-sanity/disaggregated/gb300_deepseek-v32-fp4_32k4k_con256_ctx1_dep4_gen1_dep32_eplb0_mtp3_ccb-NIXL.yaml
tests/scripts/perf-sanity/disaggregated/gb300_deepseek-v32-fp4_32k4k_con256_ctx1_dep8_gen1_dep8_eplb0_mtp0_ccb-NIXL.yaml
tests/scripts/perf-sanity/disaggregated/gb300_deepseek-v32-fp4_8k1k_con1024_ctx1_dep4_gen1_dep32_eplb256_mtp3_ccb-NIXL.yaml
tests/scripts/perf-sanity/disaggregated/gb300_deepseek-v32-fp4_8k1k_con1_ctx1_dep4_gen1_tep8_eplb0_mtp3_ccb-NIXL.yaml
tests/scripts/perf-sanity/disaggregated/gb300_deepseek-v32-fp4_8k1k_con4096_ctx1_dep4_gen1_dep32_eplb256_mtp0_ccb-NIXL.yaml
tests/scripts/perf/disaggregated/gb200_deepseek-v32-fp4_1k1k_con1024_ctx1_dep4_gen1_dep32_eplb256_mtp3_ccb-NIXL.yaml
tests/scripts/perf/disaggregated/gb200_deepseek-v32-fp4_1k1k_con1_ctx1_dep4_gen1_tep8_eplb0_mtp3_ccb-NIXL.yaml
tests/scripts/perf/disaggregated/gb200_deepseek-v32-fp4_1k1k_con2048_ctx1_dep4_gen1_dep4_eplb0_mtp1_ccb-NIXL.yaml
tests/scripts/perf/disaggregated/gb200_deepseek-v32-fp4_32k4k_con1_ctx1_dep4_gen1_tep8_eplb0_mtp3_ccb-NIXL.yaml
tests/scripts/perf/disaggregated/gb200_deepseek-v32-fp4_32k4k_con2048_ctx1_dep4_gen1_dep32_eplb288_mtp1_ccb-NIXL.yaml
tests/scripts/perf/disaggregated/gb200_deepseek-v32-fp4_32k4k_con256_ctx1_dep4_gen1_dep32_eplb0_mtp3_ccb-NIXL.yaml
tests/scripts/perf/disaggregated/gb200_deepseek-v32-fp4_32k4k_con256_ctx1_dep8_gen1_dep8_eplb0_mtp0_ccb-NIXL.yaml
tests/scripts/perf/disaggregated/gb200_deepseek-v32-fp4_8k1k_con1024_ctx1_dep4_gen1_dep32_eplb256_mtp3_ccb-NIXL.yaml
tests/scripts/perf/disaggregated/gb200_deepseek-v32-fp4_8k1k_con1_ctx1_dep4_gen1_tep8_eplb0_mtp3_ccb-NIXL.yaml
tests/scripts/perf/disaggregated/gb200_deepseek-v32-fp4_8k1k_con4096_ctx1_dep4_gen1_dep32_eplb256_mtp0_ccb-NIXL.yaml
tests/scripts/perf/disaggregated/gb200_wideep_deepseek-v32-fp4_1k1k_ctx1_gen1_dep32_bs32_eplb288_mtp0_con1024_ccb-NIXL.yaml
tests/scripts/perf/disaggregated/gb200_wideep_deepseek-v32-fp4_1k1k_ctx2_gen1_dep16_bs128_eplb288_mtp3_con2048_ccb-NIXL.yaml
tests/scripts/perf/disaggregated/gb200_wideep_deepseek-v32-fp4_1k1k_ctx2_gen1_dep48_bs16_eplb288_mtp3_con12288_ccb-NIXL.yaml
tests/scripts/perf/disaggregated/gb200_wideep_deepseek-v32-fp4_8k1k_ctx2_gen1_dep32_bs128_eplb288_mtp3_con1024_ccb-NIXL.yaml
tests/scripts/perf/disaggregated/gb300_deepseek-v32-fp4_1k1k_con1024_ctx1_dep4_gen1_dep32_eplb256_mtp3_ccb-NIXL.yaml
tests/scripts/perf/disaggregated/gb300_deepseek-v32-fp4_1k1k_con1_ctx1_dep4_gen1_tep8_eplb0_mtp3_ccb-NIXL.yaml
tests/scripts/perf/disaggregated/gb300_deepseek-v32-fp4_1k1k_con2048_ctx1_dep4_gen1_dep4_eplb0_mtp1_ccb-NIXL.yaml
tests/scripts/perf/disaggregated/gb300_deepseek-v32-fp4_32k4k_con1_ctx1_dep4_gen1_tep8_eplb0_mtp3_ccb-NIXL.yaml
tests/scripts/perf/disaggregated/gb300_deepseek-v32-fp4_32k4k_con2048_ctx1_dep4_gen1_dep32_eplb288_mtp1_ccb-NIXL.yaml
tests/scripts/perf/disaggregated/gb300_deepseek-v32-fp4_32k4k_con256_ctx1_dep4_gen1_dep32_eplb0_mtp3_ccb-NIXL.yaml
tests/scripts/perf/disaggregated/gb300_deepseek-v32-fp4_32k4k_con256_ctx1_dep8_gen1_dep8_eplb0_mtp0_ccb-NIXL.yaml
tests/scripts/perf/disaggregated/gb300_deepseek-v32-fp4_8k1k_con1024_ctx1_dep4_gen1_dep32_eplb256_mtp3_ccb-NIXL.yaml
tests/scripts/perf/disaggregated/gb300_deepseek-v32-fp4_8k1k_con1_ctx1_dep4_gen1_tep8_eplb0_mtp3_ccb-NIXL.yaml
tests/scripts/perf/disaggregated/gb300_deepseek-v32-fp4_8k1k_con4096_ctx1_dep4_gen1_dep32_eplb256_mtp0_ccb-NIXL.yaml
tests/scripts/perf/disaggregated/gb300_wideep_deepseek-v32-fp4_1k1k_ctx2_gen1_dep16_bs128_eplb288_mtp3_con2048_ccb-NIXL.yaml
tests/scripts/perf/disaggregated/gb300_wideep_deepseek-v32-fp4_8k1k_ctx2_gen1_dep32_bs128_eplb288_mtp3_con1024_ccb-NIXL.yaml

📝 Walkthrough

Walkthrough

The changes configure disaggregated serving accuracy and performance scenarios to use the NIXL backend with the Python transceiver runtime for both generation and context workers.

Changes

NIXL Python transceiver configuration

Layer / File(s)	Summary
Accuracy test backend selection `tests/integration/defs/accuracy/test_disaggregated_serving.py`	The DeepSeek V32 auto-dtype test changes its cache transceiver backend from `DEFAULT` to `NIXL` while retaining the Python runtime.
Performance-sanity runtime settings `tests/scripts/perf-sanity/disaggregated/*-NIXL.yaml`	GB200 and GB300 performance-sanity configurations add `transceiver_runtime: PYTHON` to generation and context cache transceiver blocks.
Performance runtime settings `tests/scripts/perf/disaggregated/*-NIXL.yaml`	GB200, GB300, and WideEP performance configurations add `transceiver_runtime: PYTHON` to generation and context cache transceiver blocks.

Estimated code review effort: 2 (Simple) | ~10 minutes

Suggested reviewers: jieli-matrix, brb-nv, zhaoyangwang-nvidia, HuiGao-NV, pcastonguay, Shixiaowei02, chuangz0, Tabrizian

🚥 Pre-merge checks | ✅ 3 | ❌ 2

❌ Failed checks (2 warnings)

Check name	Status	Explanation	Resolution
Description check	⚠️ Warning	The description is only the template skeleton and lacks real Description, Test Coverage, and checklist content.	Replace the placeholders with a short problem/solution summary, test coverage, and completed PR checklist items.
Docstring Coverage	⚠️ Warning	Docstring coverage is 0.00% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (3 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title matches the PR's main change and follows the required [ticket][type] summary format.
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.

✨ Finishing Touches

🧪 Generate unit tests (beta)

Create PR with unit tests

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

tensorrt-cicd · 2026-07-10T23:01:01Z

PR_Github #58714 [ run ] triggered by Bot. Commit: 20a1299 Link to invocation

tensorrt-cicd · 2026-07-10T23:32:59Z

PR_Github #58714 [ run ] completed with state FAILURE. Commit: 20a1299
/LLM/main/L0_MergeRequest_PR pipeline #47297 (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

asfiyab-nvidia · 2026-07-14T18:38:24Z

/bot run --disable-fail-fast --stage-list "DGX_B200-8_GPUs-PyTorch-1, DGX_B200-8_GPUs-PyTorch-2, DGX_B200-8_GPUs-PyTorch-3, DGX_B200-8_GPUs-PyTorch-4, GB200-4_GPUs-PyTorch-PerfSanity-1, GB200-4_GPUs-PyTorch-PerfSanity-2, GB200-8_GPUs-2_Nodes-PyTorch-Disagg-PerfSanity-CTX1-NODE1-GPU4-GEN1-NODE1-GPU4-Post-Merge, GB200-12_GPUs-3_Nodes-PyTorch-Disagg-PerfSanity-CTX1-NODE1-GPU4-GEN1-NODE2-GPU8-Post-Merge, GB200-16_GPUs-4_Nodes-PyTorch-Disagg-PerfSanity-CTX1-NODE2-GPU8-GEN1-NODE2-GPU8-Post-Merge, GB200-36_GPUs-9_Nodes-PyTorch-Disagg-PerfSanity-CTX1-NODE1-GPU4-GEN1-NODE8-GPU32-Post-Merge, GB200-8_GPUs-2_Nodes-PyTorch-PerfSanity-Node2-GPU8-Post-Merge"

tensorrt-cicd · 2026-07-14T18:44:45Z

PR_Github #59256 [ run ] triggered by Bot. Commit: 20a1299 Link to invocation

tensorrt-cicd · 2026-07-14T19:27:06Z

PR_Github #59256 [ run ] completed with state FAILURE. Commit: 20a1299
/LLM/main/L0_MergeRequest_PR pipeline #47745 (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

asfiyab-nvidia · 2026-07-14T20:54:41Z

/bot run --disable-fail-fast --stage-list "DGX_B200-8_GPUs-PyTorch-1, DGX_B200-8_GPUs-PyTorch-2, DGX_B200-8_GPUs-PyTorch-3, DGX_B200-8_GPUs-PyTorch-4, GB200-4_GPUs-PyTorch-PerfSanity-1, GB200-4_GPUs-PyTorch-PerfSanity-2, GB200-8_GPUs-2_Nodes-PyTorch-PerfSanity-Node2-GPU8-Post-Merge-1, GB200-8_GPUs-2_Nodes-PyTorch-Disagg-PerfSanity-CTX1-NODE1-GPU4-GEN1-NODE1-GPU4-Post-Merge-1, GB200-12_GPUs-3_Nodes-PyTorch-Disagg-PerfSanity-CTX1-NODE1-GPU4-GEN1-NODE2-GPU8-Post-Merge-1, GB200-16_GPUs-4_Nodes-PyTorch-Disagg-PerfSanity-CTX1-NODE2-GPU8-GEN1-NODE2-GPU8-Post-Merge-1, GB200-36_GPUs-9_Nodes-PyTorch-Disagg-PerfSanity-CTX1-NODE1-GPU4-GEN1-NODE8-GPU32-Post-Merge-1"

tensorrt-cicd · 2026-07-14T21:00:54Z

PR_Github #59278 [ run ] triggered by Bot. Commit: 20a1299 Link to invocation

tensorrt-cicd · 2026-07-14T23:52:42Z

PR_Github #59278 [ run ] completed with state FAILURE. Commit: 20a1299
/LLM/main/L0_MergeRequest_PR pipeline #47765 (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

asfiyab-nvidia · 2026-07-15T19:32:43Z

/bot run --disable-fail-fast --stage-list "DGX_B200-8_GPUs-PyTorch-1, DGX_B200-8_GPUs-PyTorch-2, DGX_B200-8_GPUs-PyTorch-3,DGX_B200-8_GPUs-PyTorch-4, GB200Disagg-PerfSanity-CTX1-NODE1-GPU4-GEN1-NODE1-GPU4-Post-Merge*,GB200Disagg-PerfSanity-CTX1-NODE1-GPU4-GEN1-NODE2-GPU8-Post-Merge*,GB200Disagg-PerfSanity-CTX1-NODE2-GPU8-GEN1-NODE2-GPU8-Post-Merge*,GB200Disagg-PerfSanity-CTX1-NODE1-GPU4-GEN1-NODE8-GPU32-Post-Merge*"

tensorrt-cicd · 2026-07-15T19:39:22Z

PR_Github #59515 [ run ] triggered by Bot. Commit: 20a1299 Link to invocation

tensorrt-cicd · 2026-07-15T21:07:41Z

PR_Github #59515 [ run ] completed with state FAILURE. Commit: 20a1299
/LLM/main/L0_MergeRequest_PR pipeline #47970 (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

asfiyab-nvidia · 2026-07-15T22:54:15Z

/bot run --disable-reuse-test --disable-fail-fast --stage-list "DGX_B200-8_GPUs-PyTorch-1, DGX_B200-8_GPUs-PyTorch-2, DGX_B200-8_GPUs-PyTorch-3,DGX_B200-8_GPUs-PyTorch-4, GB200Disagg-PerfSanity-CTX1-NODE1-GPU4-GEN1-NODE1-GPU4-Post-Merge*,GB200Disagg-PerfSanity-CTX1-NODE1-GPU4-GEN1-NODE2-GPU8-Post-Merge*,GB200Disagg-PerfSanity-CTX1-NODE2-GPU8-GEN1-NODE2-GPU8-Post-Merge*,GB200Disagg-PerfSanity-CTX1-NODE1-GPU4-GEN1-NODE8-GPU32-Post-Merge*"

tensorrt-cicd · 2026-07-15T23:00:37Z

PR_Github #59548 [ run ] triggered by Bot. Commit: 20a1299 Link to invocation

tensorrt-cicd · 2026-07-15T23:42:50Z

PR_Github #59548 [ run ] completed with state FAILURE. Commit: 20a1299
/LLM/main/L0_MergeRequest_PR pipeline #47996 (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

[TRTLLM-13948][feat] DeepSeek V3.2 TranceiverV2 migration

20a1299

Signed-off-by: Asfiya Baig <asfiyab@nvidia.com>

asfiyab-nvidia requested a review from a team as a code owner July 10, 2026 22:49

asfiyab-nvidia marked this pull request as draft July 10, 2026 22:50

github-actions Bot assigned asfiyab-nvidia Jul 10, 2026

nv-xtf mentioned this pull request Jul 21, 2026

[TRTLLM-13948][test] Migrate DeepSeek R1/V3.2 disagg perf cases to transceiver V2 #16669

Merged

1 task

Uh oh!

Conversation

asfiyab-nvidia commented Jul 10, 2026 • edited by coderabbitai Bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary by CodeRabbit

Description

Test Coverage

PR Checklist

GitHub Bot Help

Uh oh!

asfiyab-nvidia commented Jul 10, 2026

Uh oh!

coderabbitai Bot commented Jul 10, 2026

Walkthrough

Changes

❌ Failed checks (2 warnings)

Uh oh!

tensorrt-cicd commented Jul 10, 2026

Uh oh!

tensorrt-cicd commented Jul 10, 2026

Uh oh!

asfiyab-nvidia commented Jul 14, 2026

Uh oh!

tensorrt-cicd commented Jul 14, 2026

Uh oh!

tensorrt-cicd commented Jul 14, 2026

Uh oh!

asfiyab-nvidia commented Jul 14, 2026

Uh oh!

tensorrt-cicd commented Jul 14, 2026

Uh oh!

tensorrt-cicd commented Jul 14, 2026

Uh oh!

asfiyab-nvidia commented Jul 15, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

tensorrt-cicd commented Jul 15, 2026

Uh oh!

tensorrt-cicd commented Jul 15, 2026

Uh oh!

asfiyab-nvidia commented Jul 15, 2026

Uh oh!

tensorrt-cicd commented Jul 15, 2026

Uh oh!

tensorrt-cicd commented Jul 15, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

asfiyab-nvidia commented Jul 10, 2026 •

edited by coderabbitai Bot

Loading

asfiyab-nvidia commented Jul 15, 2026 •

edited

Loading