[None][feat] Interactive recipe selector with curated configs and button-grid UI

[venkywonka]

Summary

Overhaul the recipe selector on the Model Recipes documentation page into a polished, dual-source interactive widget that exposes both curated/selected configs and the database (170+ pareto-optimized configs) in one place.

What changed

• Button-grid UI: Replace cascading dropdowns with a button-grid selector (Model → GPU → ISL/OSL → Concurrency dropdown). Incompatible options are visually greyed out, valid selections highlight in NVIDIA green.
• Curated "Recommended Configs" panel: When a model is selected, curated configs appear as clickable cards between the Model row and GPU row. Clicking a card immediately resolves to a trtllm-serve command, dimming the database filter chain. Clicking again deselects.
• Model grouping by display name: Multiple HuggingFace checkpoint IDs that represent the same model (e.g. 3 DeepSeek-R1 NVFP4 variants) are collapsed into a single button. Each generated command still uses the correct HF model ID.
• Syntax highlighting: Bash highlighting for the command output (export, --config, ${TRTLLM_DIR}), YAML highlighting for the config preview panel.
• Inline export TRTLLM_DIR: The copyable command now includes export TRTLLM_DIR=/app/tensorrt_llm as the first line, so users get a working 3-liner.
• Disaggregated config filtering: Configs requiring disaggregated serving (Kimi-K2-Thinking with UCX transceiver, Qwen3 disagg-prefill) are tagged disagg: true in the curated lookup and filtered from the UI. An RST note clarifies these configs are for aggregated (in-flight batched) serving only.
• Quick Start table removed: The static 11-row RST table is fully replaced by the interactive selector.
• Performance profile in concurrency: The profile step is merged into the concurrency dropdown (e.g. "64 · Balanced").
• Friendly model names: Buttons show display names (e.g. "DeepSeek-R1") instead of raw HF IDs.

Data pipeline

lookup.yaml → CuratedRecipe (Pydantic) → build_curated_rows() → config_db.json (with curated_entries array) → JS widget

Files changed

Area	Files
Curated data	examples/configs/curated/lookup.yaml
Data models	examples/configs/database/database.py
JSON generation	scripts/generate_config_table.py
Sphinx extension	docs/source/_ext/trtllm_config_selector.py
Widget JS	docs/source/_static/config_selector.js
Widget CSS	docs/source/_static/config_selector.css
Tests	tests/unittest/tools/test_config_selector.js (18 tests)
Docs	docs/source/deployment-guide/index.rst + per-model guides

Test plan

• node --test tests/unittest/tools/test_config_selector.js — 18 tests pass
• Visual verification at the preview link below
  • Select DeepSeek-R1 → 3 curated cards (Max Throughput H100/H200, Max Throughput B200/GB200, Min Latency B200/GB200)
  • Click a curated card → command resolves, filter chain dims
  • Click card again → deselects
  • Select a curated-only model (e.g. Llama-3.3-70B) → single curated card, no database filters
  • Select DeepSeek-R1 (NVFP4) → curated cards + database filter flow both available
  • Verify dark/light mode styling
• cd docs && make html builds without errors

Preview

https://venkywonka.github.io/TensorRT-LLM/deployment-guide/index.html#recipe-selector

🤖 Generated with Claude Code

[coderabbitai]

📝 Walkthrough

Walkthrough

The changes remove the Performance profile UI component and its associated filtering logic from the config selector, consolidating profile information into concurrency option labels. A deployment guide table linking to pre-configured model scenarios is replaced with text referencing external resources.

Changes

Cohort / File(s)	Summary
Config Selector UI Refactoring docs/source/_static/config_selector.js	Removes Profile UI element, state management for profile, and profile-based filtering. Profile information is now embedded in concurrency option labels. Simplifies filter logic, state reset behavior, and error messaging by eliminating separate profile constraints and change handlers.
Deployment Guide Documentation docs/source/deployment-guide/index.rst	Replaces embedded pre-configured model scenarios table with textual overview directing users to the configuration database on GitHub and the interactive recipe selector.

Estimated code review effort

🎯 3 (Moderate) | ⏱️ ~20 minutes

🚥 Pre-merge checks | ✅ 1 | ❌ 2

❌ Failed checks (1 warning, 1 inconclusive)

Check name	Status	Explanation	Resolution
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.
Title check	❓ Inconclusive	The PR title mentions 'Interactive recipe selector with curated configs and button-grid UI' but the actual changes merge the Performance Profile into Concurrency dropdown and simplify the UI from 5 to 4 steps—the button-grid UI aspect is not evident in the provided summary.	Clarify whether 'button-grid UI' is implemented in this PR, or revise the title to accurately reflect the main change: merging Performance Profile into the Concurrency dropdown.

✅ Passed checks (1 passed)

Check name	Status	Explanation
Description check	✅ Passed	The PR description comprehensively covers what changed, why, the data pipeline, files affected, and test plan with clear examples and verification steps.

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

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

📝 Coding Plan

• Generate coding plan for human review comments

---

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

[coderabbitai]

Actionable comments posted: 1

🤖 Prompt for all review comments with AI agents

Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@docs/source/_static/config_selector.js`:
- Around line 472-488: The concurrency dropdown collapses distinct performance
profiles that share the same numeric concurrency because concOpts currently keys
options only by value = String(e.concurrency); update the option generation in
the concEntries -> concOpts pipeline (the concEntries, concOpts,
filteredByState, setSelectOptions and finalEntries code paths) so each option
value is unique per profile (e.g., combine e.concurrency and
e.performance_profile into the option value) while still preserving a numeric
conc field for sorting; update the logic that checks/sets state.concurrency and
any comparisons in filteredByState/finalEntries to use the new composite value
when matching selections so finalEntries cannot return multiple rows for a
selected option.

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro

Run ID: 9818fc81-c447-4aa2-8960-ab607ea7f209

📥 Commits

Reviewing files that changed from the base of the PR and between 234eb83 and f1136e4.

📒 Files selected for processing (2)

• docs/source/_static/config_selector.js
• docs/source/deployment-guide/index.rst

[venkywonka]

/bot run

[tensorrt-cicd]

PR_Github #38333 [ run ] triggered by Bot. Commit: f1136e4 Link to invocation

[tensorrt-cicd]

PR_Github #38333 [ run ] completed with state SUCCESS. Commit: f1136e4
/LLM/main/L0_MergeRequest_PR pipeline #29709 completed with status: 'SUCCESS'

Link to invocation

[venkywonka]

/bot run

[venkywonka]

/bot run

[Copilot]

Pull request overview

This PR overhauls the Model Recipes documentation experience by replacing the static recipe tables with an interactive, dual-source selector widget that surfaces both curated “recommended” configs and the full database-generated config set.

Changes:

• Introduces a new JS/CSS widget for interactive recipe selection, including curated cards and updated command/config preview rendering.
• Extends the config database generation pipeline to optionally ingest curated recipes and emit them into config_db.json.
• Removes the legacy generated RST recipe table (config_table.rst) and updates docs + tests to align with the new selector approach.

Reviewed changes

Copilot reviewed 13 out of 14 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
tests/unittest/tools/test_config_selector.js	Adds unit tests covering selector view-model compatibility logic and curated-entry formatting/filtering.
tests/unittest/tools/test_config_database_sync.py	Updates sync tests to validate JSON payload shape/metadata now that RST table generation is removed.
scripts/generate_config_table.py	Adds curated recipe ingestion and emits curated_entries into generated config_db.json; removes RST generator.
examples/configs/database/database.py	Adds Pydantic models + loader for curated recipe YAML (including config path validation).
examples/configs/curated/lookup.yaml	Defines curated recipes, scenarios, GPU compatibility, and disaggregated-serving tagging.
docs/source/deployment-guide/index.rst	Removes the static quick-start table and anchors the new recipe selector in the docs flow.
docs/source/deployment-guide/deployment-guide-for-gpt-oss-on-trtllm.md	Removes embedded includes of the old recipe database table.
docs/source/deployment-guide/deployment-guide-for-deepseek-r1-on-trtllm.md	Removes embedded includes of the old recipe database table.
docs/source/deployment-guide/config_table.rst	Deletes the generated static table that previously mirrored the config database.
docs/source/_static/config_selector.js	Implements button-grid compatibility UI, curated cards, bash/yaml highlighting, and exports pure view-model helpers for tests.
docs/source/_static/config_selector.css	Adds styling for the new selector layout, curated cards, and syntax highlighting tokens.
docs/source/_static/config_db.json	Updates the committed static payload to include curated entries + expanded model metadata.
docs/source/_ext/trtllm_config_selector.py	Updates Sphinx build hook to generate config_db.json including curated recipes when present.
.gitignore	Removes the exception for the now-deleted config_table.rst.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[tensorrt-cicd]

PR_Github #39692 [ run ] triggered by Bot. Commit: fc947c4 Link to invocation

[tensorrt-cicd]

PR_Github #39692 [ run ] completed with state SUCCESS. Commit: fc947c4
/LLM/main/L0_MergeRequest_PR pipeline #30890 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

Link to invocation

[venkywonka]

/bot run

[tensorrt-cicd]

PR_Github #39759 [ run ] triggered by Bot. Commit: fc947c4 Link to invocation

[venkywonka]

/bot run

[tensorrt-cicd]

PR_Github #39777 [ run ] triggered by Bot. Commit: fc947c4 Link to invocation

[tensorrt-cicd]

PR_Github #39759 [ run ] completed with state ABORTED. Commit: fc947c4
LLM/main/L0_MergeRequest_PR #30954 (Blue Ocean) completed with status: ABORTED

Link to invocation

[tensorrt-cicd]

PR_Github #39777 [ run ] completed with state SUCCESS. Commit: fc947c4
/LLM/main/L0_MergeRequest_PR pipeline #30971 completed with status: 'SUCCESS'

CI Report

Link to invocation

[arysef]

lgtm