[TRTLLM-13694][feat] Add IBDB recipe provenance and refresh configs

[Mgluhovskoi]

Description

Add the public config-database support needed for IBDB-sourced recipe refreshes and include the first DeepSeek/B200 batch.

This change:

• adds optional, paired validated_trtllm_version and validated_trtllm_commit provenance metadata;
• validates release-tag-safe versions and full 40-character Git SHAs;
• adds a conflict-only profile discriminator for latency and throughput variants, with an optional balanced variant;
• propagates provenance and profile metadata through the generated config database and documentation selector;
• keeps generated test names unique when profiles exist and rejects unknown lookup fields;
• refreshes 17 nvidia/DeepSeek-R1-0528-FP4-v2 B200 recipes from IBDB Pareto records validated against TensorRT-LLM v1.3.0rc14 at commit 93cb6518b6d6dbd6095748189e626db731f44545;
• adds eight 4-GPU recipes and updates nine 8-GPU recipes for 1024/1024 and 8192/1024 workloads; and
• regenerates the customer-facing config_db.json.

IBDB record, study, and job identifiers remain private and are not written to this repository. Existing recipes remain valid without provenance metadata. There are no new dependencies or protected TensorRT-LLM API changes.

Related ticket: TRTLLM-13694.

Test Coverage

• python scripts/generate_config_table.py — passed.
• python tests/unittest/tools/test_config_database_sync.py TestConfigDatabaseSync.test_config_db_json_generation TestConfigDatabaseSync.test_profile_and_validated_commit_metadata TestConfigDatabaseSync.test_recipe_metadata_validation — 3 passed.
• node --test tests/unittest/tools/test_config_selector.js — passed.
• Live automation validation — 17 recipes selected; 8 added and 9 updated; all skipped records reported with explicit reasons.
• Repository pre-commit hooks for all changed files — passed.
• Full tests/unittest/llmapi/test_config_database.py was not run locally because the checkout lacks its TensorRT-LLM runtime dependencies; it remains required in CI.

Summary by CodeRabbit

• New Features
  • Added profile-based filtering to the configuration selector, including latency and throughput options.
  • Added validated TensorRT-LLM version and commit details to applicable configurations.
  • Added new DeepSeek-R1 B200 configurations for 1k1k and 8k1k workloads.
• Improvements
  • Updated configuration validation to enforce valid profiles and validation metadata.
  • Refined batching, token, sequence-length, and performance settings across DeepSeek-R1 configurations.
  • Improved selector messaging, focus handling, and generated configuration naming.

[coderabbitai]

📝 Walkthrough

Walkthrough

The configuration database adds workload profiles and validated TensorRT-LLM provenance. Recipe validation, database generation, B200 configurations, published JSON/YAML data, the configuration selector, and related tests are updated to propagate and display these fields.

Changes

Profile-aware configuration database

Layer / File(s)	Summary
Recipe metadata and profile validation examples/configs/database/database.py	Recipe schemas now validate profile values, TensorRT-LLM commit/version pairs, required field constraints, and profile combinations.
B200 runtime recipes and published metadata examples/configs/database/nvidia/..., examples/configs/database/lookup.yaml, docs/source/_static/config_db.json	New and revised DeepSeek B200 configurations adjust runtime limits and add validated TensorRT-LLM metadata.
Database and test generation scripts/generate_config_database_tests.py, scripts/generate_config_table.py, tests/unittest/tools/test_config_database_sync.py	Generated names, table rows, JSON entries, and synchronization tests now preserve profile and validation metadata.
Profile-aware configuration selector docs/source/_static/config_selector.js, docs/source/_static/config_selector.css, tests/unittest/tools/test_config_selector.js	The selector adds profile filtering and controls, focus handling, validated commit links, metadata rendering, and corresponding tests.

Estimated code review effort: 4 (Complex) | ~45 minutes

Sequence Diagram(s)

sequenceDiagram
  participant ProfileSelect
  participant ConfigSelector
  participant ConfigDatabase
  ProfileSelect->>ConfigSelector: Select profile
  ConfigSelector->>ConfigDatabase: Filter entries by profile
  ConfigDatabase-->>ConfigSelector: Return compatible entries
  ConfigSelector-->>ProfileSelect: Render resolved configuration and validation metadata

Loading

Possibly related PRs

• NVIDIA/TensorRT-LLM#9832: Established the related configuration database generation and synchronization workflow extended by this change.

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 60.00% 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 matches the main change and includes the required ticket and feature type.
Description check	✅ Passed	The description covers the required Description and Test Coverage, but omits the PR Checklist section from the template.
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.}

[coderabbitai]

Actionable comments posted: 1

Caution

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

⚠️ Outside diff range comments (1)

examples/configs/database/nvidia/DeepSeek-R1-0528-FP4-v2/B200/8k1k_tp4_conc16.yaml (1)

1-20: 📐 Maintainability & Code Quality | 🟡 Minor | ⚡ Quick win

Add the required NVIDIA copyright headers to the new recipe files.

• examples/configs/database/nvidia/DeepSeek-R1-0528-FP4-v2/B200/8k1k_tp4_conc16.yaml#L1-L20: Add the SPDX copyright comment before max_batch_size.
• examples/configs/database/nvidia/DeepSeek-R1-0528-FP4-v2/B200/8k1k_tp4_conc256.yaml#L1-L25: Add the SPDX copyright comment before max_batch_size.
• examples/configs/database/nvidia/DeepSeek-R1-0528-FP4-v2/B200/8k1k_tp4_conc4.yaml#L1-L20: Add the SPDX copyright comment before max_batch_size.
• examples/configs/database/nvidia/DeepSeek-R1-0528-FP4-v2/B200/8k1k_tp4_conc8.yaml#L1-L20: Add the SPDX copyright comment before max_batch_size.

As per coding guidelines, “Add the NVIDIA copyright header to all new files.”

🤖 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
`@examples/configs/database/nvidia/DeepSeek-R1-0528-FP4-v2/B200/8k1k_tp4_conc16.yaml`
around lines 1 - 20, Prepend the standard NVIDIA SPDX copyright header before
max_batch_size in all four new recipe files:
examples/configs/database/nvidia/DeepSeek-R1-0528-FP4-v2/B200/8k1k_tp4_conc16.yaml
lines 1-20, 8k1k_tp4_conc256.yaml lines 1-25, 8k1k_tp4_conc4.yaml lines 1-20,
and 8k1k_tp4_conc8.yaml lines 1-20. Leave the existing YAML configuration
unchanged.

Source: Coding guidelines

🤖 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
`@examples/configs/database/nvidia/DeepSeek-R1-0528-FP4-v2/B200/1k1k_tp4_conc16.yaml`:
- Line 1: Add the repository-standard NVIDIA copyright header before the first
configuration key in all four files:
examples/configs/database/nvidia/DeepSeek-R1-0528-FP4-v2/B200/1k1k_tp4_conc16.yaml:1-1,
1k1k_tp4_conc256.yaml:1-1, 1k1k_tp4_conc4.yaml:1-1, and 1k1k_tp4_conc8.yaml:1-1.
Preserve each file’s existing configuration content, including cuda_graph_config
or max_batch_size.

---

Outside diff comments:
In
`@examples/configs/database/nvidia/DeepSeek-R1-0528-FP4-v2/B200/8k1k_tp4_conc16.yaml`:
- Around line 1-20: Prepend the standard NVIDIA SPDX copyright header before
max_batch_size in all four new recipe files:
examples/configs/database/nvidia/DeepSeek-R1-0528-FP4-v2/B200/8k1k_tp4_conc16.yaml
lines 1-20, 8k1k_tp4_conc256.yaml lines 1-25, 8k1k_tp4_conc4.yaml lines 1-20,
and 8k1k_tp4_conc8.yaml lines 1-20. Leave the existing YAML configuration
unchanged.

🪄 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: 6a856656-f88c-49e6-8dfc-052caf5e0c8b

📥 Commits

Reviewing files that changed from the base of the PR and between f81b9d2 and 0a29694.

📒 Files selected for processing (26)

• docs/source/_static/config_db.json
• docs/source/_static/config_selector.css
• docs/source/_static/config_selector.js
• examples/configs/database/database.py
• examples/configs/database/lookup.yaml
• examples/configs/database/nvidia/DeepSeek-R1-0528-FP4-v2/B200/1k1k_tp4_conc16.yaml
• examples/configs/database/nvidia/DeepSeek-R1-0528-FP4-v2/B200/1k1k_tp4_conc256.yaml
• examples/configs/database/nvidia/DeepSeek-R1-0528-FP4-v2/B200/1k1k_tp4_conc4.yaml
• examples/configs/database/nvidia/DeepSeek-R1-0528-FP4-v2/B200/1k1k_tp4_conc8.yaml
• examples/configs/database/nvidia/DeepSeek-R1-0528-FP4-v2/B200/1k1k_tp8_conc128.yaml
• examples/configs/database/nvidia/DeepSeek-R1-0528-FP4-v2/B200/1k1k_tp8_conc256.yaml
• examples/configs/database/nvidia/DeepSeek-R1-0528-FP4-v2/B200/1k1k_tp8_conc32.yaml
• examples/configs/database/nvidia/DeepSeek-R1-0528-FP4-v2/B200/1k1k_tp8_conc4.yaml
• examples/configs/database/nvidia/DeepSeek-R1-0528-FP4-v2/B200/1k1k_tp8_conc64.yaml
• examples/configs/database/nvidia/DeepSeek-R1-0528-FP4-v2/B200/8k1k_tp4_conc16.yaml
• examples/configs/database/nvidia/DeepSeek-R1-0528-FP4-v2/B200/8k1k_tp4_conc256.yaml
• examples/configs/database/nvidia/DeepSeek-R1-0528-FP4-v2/B200/8k1k_tp4_conc4.yaml
• examples/configs/database/nvidia/DeepSeek-R1-0528-FP4-v2/B200/8k1k_tp4_conc8.yaml
• examples/configs/database/nvidia/DeepSeek-R1-0528-FP4-v2/B200/8k1k_tp8_conc128.yaml
• examples/configs/database/nvidia/DeepSeek-R1-0528-FP4-v2/B200/8k1k_tp8_conc256.yaml
• examples/configs/database/nvidia/DeepSeek-R1-0528-FP4-v2/B200/8k1k_tp8_conc4.yaml
• examples/configs/database/nvidia/DeepSeek-R1-0528-FP4-v2/B200/8k1k_tp8_conc64.yaml
• scripts/generate_config_database_tests.py
• scripts/generate_config_table.py
• tests/unittest/tools/test_config_database_sync.py
• tests/unittest/tools/test_config_selector.js

[Mgluhovskoi]

/bot run

[tensorrt-cicd]

PR_Github #60484 [ run ] triggered by Bot. Commit: cb49bb4 Link to invocation

[tensorrt-cicd]

PR_Github #60484 [ run ] completed with state SUCCESS. Commit: cb49bb4
/LLM/main/L0_MergeRequest_PR pipeline #48815 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

[Mgluhovskoi]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #60522 [ run ] triggered by Bot. Commit: cb49bb4 Link to invocation

[BowenFu]

LGTM — purely additive tooling: all changes are under docs/static, examples/configs/database (new recipe YAMLs + provenance), scripts/generate_config*, and tests; no tensorrt_llm/ runtime source is touched, so users not using IBDB recipes see zero behavior change. (Worth addressing CodeRabbit's ask for NVIDIA copyright headers on the new YAML recipe files.)

[tensorrt-cicd]

PR_Github #60522 [ run ] completed with state SUCCESS. Commit: cb49bb4
/LLM/main/L0_MergeRequest_PR pipeline #48845 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

[Mgluhovskoi]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #60722 [ run ] triggered by Bot. Commit: 61f21e0 Link to invocation

[tensorrt-cicd]

PR_Github #60722 [ run ] completed with state FAILURE. Commit: 61f21e0
/LLM/main/L0_MergeRequest_PR pipeline #49009 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

[Mgluhovskoi]

/bot run

[tensorrt-cicd]

PR_Github #60808 [ run ] triggered by Bot. Commit: 61f21e0 Link to invocation

[tensorrt-cicd]

PR_Github #60808 [ run ] completed with state SUCCESS. Commit: 61f21e0
/LLM/main/L0_MergeRequest_PR pipeline #49084 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

[Mgluhovskoi]

/bot run

[tensorrt-cicd]

PR_Github #60842 [ run ] triggered by Bot. Commit: 61f21e0 Link to invocation

[tensorrt-cicd]

PR_Github #60842 [ run ] completed with state SUCCESS. Commit: 61f21e0
/LLM/main/L0_MergeRequest_PR pipeline #49117 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

[Mgluhovskoi]

/bot run

[tensorrt-cicd]

PR_Github #60882 [ run ] triggered by Bot. Commit: e355249 Link to invocation

[tensorrt-cicd]

PR_Github #60882 [ run ] completed with state SUCCESS. Commit: e355249
/LLM/main/L0_MergeRequest_PR pipeline #49151 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

[Mgluhovskoi]

/bot run

[tensorrt-cicd]

PR_Github #61018 [ run ] triggered by Bot. Commit: e355249 Link to invocation

[tensorrt-cicd]

PR_Github #61018 [ run ] completed with state SUCCESS. Commit: e355249
/LLM/main/L0_MergeRequest_PR pipeline #49275 completed with status: 'SUCCESS'

CI Report

Link to invocation