diff --git a/docs/source/conf.py b/docs/source/conf.py index 3e5b9d942795..0a0ef3e1231b 100644 --- a/docs/source/conf.py +++ b/docs/source/conf.py @@ -11,6 +11,7 @@ # -- Project information ----------------------------------------------------- # https://www.sphinx-doc.org/en/master/usage/configuration.html#project-information import os +import re import subprocess import sys @@ -182,6 +183,15 @@ def tag_role(name, rawtext, text, lineno, inliner, options=None, content=None): return [node], [] +def _expand_visualgen_navigation(_app, docname: str, source: list[str]) -> None: + """Keep Sphinx navigation hidden in GitHub's Markdown preview.""" + if docname == 'features/visual-generation': + source[0] = re.sub(r'', + r'\1', + source[0], + flags=re.DOTALL) + + def setup(app): from helper import (check_llmapi_reference_size, compact_llmapi_search_signature, generate_examples, @@ -209,6 +219,7 @@ def setup(app): print(f"Warning: {msg}; skipping tag_llm_params") app.add_role('tag', tag_role) + app.connect('source-read', _expand_visualgen_navigation) app.connect('autodoc-process-docstring', strip_llmapi_search_docstrings) app.connect('autodoc-process-signature', compact_llmapi_search_signature) app.connect('build-finished', check_llmapi_reference_size) diff --git a/docs/source/features/visual-generation.md b/docs/source/features/visual-generation.md new file mode 100644 index 000000000000..7a4bbcbd3110 --- /dev/null +++ b/docs/source/features/visual-generation.md @@ -0,0 +1,32 @@ + + +# Visual Generation (Beta) + +This section documents optimization features for VisualGen image and video generation. Use the guides below for configuration, supported options, and interactions between features. + +For supported models and an overview of VisualGen, see the [model overview](../models/visual-generation.md). For usage examples, see [VisualGen Examples](https://nvidia.github.io/TensorRT-LLM/examples/visual_gen_examples.html). + +- [CUDA Graphs](visualgen-cuda-graph.md) +- [Quantized Attention](visualgen-quantized-attention.md) +- [Sparse Attention](visualgen-sparse-attention.md) + + + +## Related technical blogs + +These articles explain the design choices and benchmark results behind VisualGen optimizations: + +- [Scaling video generation across NVL72](../blogs/tech_blog/blog25_Scaling_Video_Generation_Across_NVL72_Rack_with_TensorRT-LLM.md) covers multi-GPU parallelism and scaling results. +- [Quantization and skipsoftmax attention for video generation](../blogs/tech_blog/blog28_Accelerating_Video_Generation_with_GEMM_Quantization_Attention_Quantization_and_Skip_Softmax_Attention_in_TensorRT-LLM.md) covers GEMM quantization, quantized attention, and Skip Softmax Attention, including performance and visual-quality trade-offs. diff --git a/docs/source/index.rst b/docs/source/index.rst index cd3bb7f95643..24c460d0e99e 100644 --- a/docs/source/index.rst +++ b/docs/source/index.rst @@ -1,3 +1,6 @@ +.. SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved. +.. SPDX-License-Identifier: Apache-2.0 + .. TensorRT LLM documentation master file, created by sphinx-quickstart on Wed Sep 20 08:35:21 2023. You can adapt this file completely to your liking, but it should at least @@ -89,9 +92,7 @@ Welcome to TensorRT LLM's Documentation! features/helix.md features/kv-cache-connector.md features/sparse-attention.md - features/visualgen-cuda-graph.md - features/visualgen-quantized-attention.md - features/visualgen-sparse-attention.md + features/visual-generation.md .. toctree::