Merge branch 'main' into docker/cudnn

Lightning-AI · Mar 26, 2024 · 5bb2f09 · 5bb2f09
2 parents fac798d + 79af2e3
commit 5bb2f09
Show file tree

Hide file tree

Showing 19 changed files with 273 additions and 1,685 deletions.
diff --git a/.github/workflows/docs-build.yml b/.github/workflows/docs-build.yml
@@ -15,11 +15,51 @@ defaults:
 
 jobs:
   docs-make:
-    uses: Lightning-AI/utilities/.github/workflows/check-docs.yml@v0.11.0
-    with:
-      python-version: "3.10"
-      requirements-file: "requirements/docs.txt"
-      install-tex: true
+    if: github.event.pull_request.draft == false
+    runs-on: ubuntu-22.04
+    strategy:
+      fail-fast: false
+      matrix:
+        target: ["html", "doctest", "linkcheck"]
+    env:
+      ARTIFACT_DAYS: 0
+      PYPI_LOCAL_DIR: "pypi_pkgs/"
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/setup-python@v5
+        with:
+          python-version: "3.10"
+
+      - name: Pull sphinx template
+        run: |
+          pip install -q "awscli >=1.30.0"
+          aws s3 sync --no-sign-request s3://sphinx-packages/ ${PYPI_LOCAL_DIR}
+          pip install lai-sphinx-theme -U -f ${PYPI_LOCAL_DIR}
+      - name: Install pandoc
+        timeout-minutes: 5
+        run: |
+          sudo apt-get update --fix-missing
+          sudo apt-get install -y pandoc
+      - name: Install package & dependencies
+        timeout-minutes: 20
+        run: pip install . -U -r requirements/docs.txt
+
+      - name: Make ${{ matrix.target }}
+        working-directory: docs/
+        # allow failing link check and doctest if you run with dispatch
+        continue-on-error: ${{ matrix.target == 'doctest' || matrix.target == 'linkcheck' }}
+        run: make ${{ matrix.target }} --debug --jobs $(nproc) SPHINXOPTS="-W --keep-going"
+
+      - name: Keep artifact
+        if: github.event_name == 'pull_request'
+        run: echo "ARTIFACT_DAYS=7" >> $GITHUB_ENV
+      - name: Upload built docs
+        if: ${{ matrix.target == 'html' }}
+        uses: actions/upload-artifact@v4
+        with:
+          name: docs-html-${{ github.sha }}
+          path: docs/build/html/
+          retention-days: ${{ env.ARTIFACT_DAYS }}
 
   deploy-docs:
     needs: docs-make
@@ -28,7 +68,7 @@ jobs:
     env:
       GCP_TARGET: "gs://lightning-docs-thunder"
     steps:
-      - uses: actions/download-artifact@v3
+      - uses: actions/download-artifact@v4
         with:
           name: docs-html-${{ github.sha }}
           path: docs/build/

diff --git a/.pre-commit-config.yaml b/.pre-commit-config.yaml
@@ -8,7 +8,7 @@ ci:
 
 repos:
   - repo: https://github.com/pre-commit/pre-commit-hooks
-    rev: v4.4.0
+    rev: v4.5.0
     hooks:
       - id: end-of-file-fixer
       - id: trailing-whitespace
@@ -24,22 +24,22 @@ repos:
       - id: detect-private-key
 
   - repo: https://github.com/asottile/pyupgrade
-    rev: v3.11.1
+    rev: v3.15.2
     hooks:
       - id: pyupgrade
         args: ["--py310-plus"]
         name: Upgrade code
         exclude: "examples|thunder/tests/test_interpreter.py|thunder/tests/test_jit_general.py"
 
   - repo: https://github.com/codespell-project/codespell
-    rev: v2.2.5
+    rev: v2.2.6
     hooks:
       - id: codespell
         additional_dependencies: [tomli]
         #args: ["--write-changes"] # uncomment if you want to get automatic fixing
 
   - repo: https://github.com/psf/black
-    rev: 23.9.1
+    rev: 24.3.0
     hooks:
       - id: black
         name: Black code
@@ -61,7 +61,7 @@ repos:
       - id: sphinx-lint
 
   - repo: https://github.com/asottile/yesqa
-    rev: v1.4.0
+    rev: v1.5.0
     hooks:
       - id: yesqa
 

diff --git a/Makefile b/Makefile
@@ -10,7 +10,13 @@ test: clean
 	python -m coverage run --source thunder -m pytest thunder tests -v
 	python -m coverage report
 
-docs: clean
+sphinx-theme:
+	pip install -q awscli
+	mkdir -p dist/
+	aws s3 sync --no-sign-request s3://sphinx-packages/ dist/
+	pip install lai-sphinx-theme -f dist/
+
+docs: clean sphinx-theme
 	pip install -e . --quiet -r requirements/docs.txt -f https://download.pytorch.org/whl/cpu/torch_stable.html
 	cd docs ; python -m sphinx -b html -W --keep-going source build
 

diff --git a/README.md b/README.md
@@ -14,8 +14,9 @@ ______________________________________________________________________
   <a href="#get-started">Get started</a> •
   <a href="#install-thunder">Install</a> •
   <a href="#hello-world">Examples</a> •
-  <a href="#features">Features</a> •
-  <a href="#documentation">Documentation</a> •
+  <a href="#inside-thunder-a-brief-look-at-the-core-features">Inside Thunder</a> •
+  <a href="#get-involved">Get involved!</a> •
+  <a href="#documentation">Documentation</a>
 </p>
 
 [![license](https://img.shields.io/badge/License-Apache%202.0-blue.svg)](https://github.com/Lightning-AI/lightning-thunder/blob/main/LICENSE)
@@ -30,41 +31,58 @@ ______________________________________________________________________
 
 **Thunder makes PyTorch models Lightning fast.**
 
-Thunder is a source-to-source compiler for PyTorch. It makes PyTorch programs faster by combining and using different hardware executors at once (ie: nvFuser, torch.compile, cuDNN, and TransformerEngine FP8).
+Thunder is a source-to-source compiler for PyTorch. It makes PyTorch programs faster by combining and using different hardware executors at once (for instance, [nvFuser](https://github.com/NVIDIA/Fuser), [torch.compile](https://pytorch.org/docs/stable/torch.compiler.html), [cuDNN](https://developer.nvidia.com/cudnn), and [TransformerEngine FP8](https://github.com/NVIDIA/TransformerEngine)).
 
-Works on single accelerators and in multi-GPU settings.
+It supports both single and multi-GPU configurations.
 Thunder aims to be usable, understandable, and extensible.
 
-## Performance
+&#160;
 
-Thunder can achieve significant speedups over standard PyTorch eager code, through the compounding effects of optimizations and the use of best-in-class executors. Here is an example of the pretraining throughput for Llama 2 7B as implemented in [LitGPT](https://github.com/Lightning-AI/litgpt).
+> \[!Note\]
+> Lightning Thunder is in alpha. Feel free to get involved, but expect a few bumps along the way.
+
+&#160;
+
+## Single-GPU performance
+
+Thunder can achieve significant speedups over standard non-compiled PyTorch code ("PyTorch eager"), through the compounding effects of optimizations and the use of best-in-class executors. The figure below shows the pretraining throughput for Llama 2 7B as implemented in [LitGPT](https://github.com/Lightning-AI/litgpt).
 
 <div align="center">
 <img alt="Thunder" src="docs/source/_static/images/training_throughput_single.png" width="800px" style="max-width: 100%;">
 </div>
 
-Thunder achieves a 40% speedup in training throughput compared to eager code on H100 using a combination of executors including nvFuser, torch.compile, cuDNN, and TransformerEngine FP8.
+As shown in the plot above, Thunder achieves a 40% speedup in training throughput compared to eager code on H100 using a combination of executors including nvFuser, torch.compile, cuDNN, and TransformerEngine FP8.
+
+&#160;
 
-Thunder supports distributed strategies like DDP and FSDP (ZeRO2 and ZeRO3). Here is the normalized throughput measured for Llama 2 7B (this time without FP8 mixed precision, support for FSDP is underway).
+## Multi-GPU performance
+
+Thunder also supports distributed strategies such as DDP and FSDP for training models on multiple GPUs. The following plot displays the normalized throughput measured for Llama 2 7B without FP8 mixed precision; support for FSDP is in progress.
 
 <div align="center">
 <img alt="Thunder" src="docs/source/_static/images/normalized_training_throughput_zero2.png" width="800px" style="max-width: 100%;">
 </div>
 
-**NOTE: Lightning Thunder is alpha.** Feel free to get involved, expect a few bumps along the way.
+&#160;
 
 ## Get started
 
-Try Thunder without installing by using our [Zero to Thunder Tutorial Studio](https://lightning.ai/lightning-ai/studios/zero-to-thunder-tutorial).
+The easiest way to get started with Thunder, requiring no extra installations or setups, is by using our [Zero to Thunder Tutorial Studio](https://lightning.ai/lightning-ai/studios/zero-to-thunder-tutorial).
+
+&#160;
 
 ## Install Thunder
 
-Install [nvFuser](https://github.com/NVIDIA/Fuser) nightly, and Thunder together
+To use Thunder on your local machine, first install [nvFuser](https://github.com/NVIDIA/Fuser) nightly and PyTorch nightly together as follows:
 
 ```bash
 # install nvFuser which installs the matching nightly PyTorch
 pip install --pre 'nvfuser-cu121[torch]' --extra-index-url https://pypi.nvidia.com
+```
 
+Then, install Thunder as follows:
+
+```
 # install thunder
 pip install lightning-thunder
 ```
@@ -73,26 +91,60 @@ pip install lightning-thunder
   <summary>Advanced install options</summary>
     <!-- following section will be skipped from PyPI description -->
 
+&#160;
+
 ### Install from main
 
+Alternatively, you can install the latest version of Thunder directly from this GitHub repository as follows:
+
+```
+# 1) Install nvFuser and PyTorch nightly dependencies:
+pip install --pre 'nvfuser-cu121[torch]' --extra-index-url https://pypi.nvidia.com
+```
+
 ```bash
+# 2) Install Thunder itself
 pip install git+https://github.com/Lightning-AI/lightning-thunder.git
 ```
 
+&#160;
+
 ### Install to tinker and contribute
 
-Install this way to tinker with the internals and contribute:
+If you are interested in tinkering with and contributing to Thunder, we recommend cloning the Thunder repository and installing it in pip's editable mode:
 
 ```bash
+git clone https://github.com/Lightning-AI/lightning-thunder.git
+cd lightning-thunder
 pip install -e .
 ```
 
+&#160;
+
+### Develop and run tests
+
+After cloning the lightning-thunder repository and installing it as an editable package as explained above, ou can set up your environment for developing Thunder by installing the development requirements:
+
+```bash
+pip install -r requirements/devel.txt
+```
+
+Now you run tests:
+
+```bash
+pytest thunder/tests
+```
+
+Thunder is very thoroughly tested, so expect this to take a while.
+
 </details>
 <!-- end skipping PyPI description -->
 
+&#160;
+
 ## Hello World
 
-Here is a simple example of how Thunder lets you compile and run PyTorch code:
+Below is a simple example of how Thunder allows you to compile and run PyTorch code:
 
 ```python
 import torch
@@ -120,15 +172,19 @@ print(result)
 
 The compiled function `jfoo` takes and returns PyTorch tensors, just like the original function, so modules and functions compiled by Thunder can be used as part of larger PyTorch programs.
 
+&#160;
+
 ## Train models
 
 Thunder is in its early stages and should not be used for production runs yet.
 
-However, it can already deliver outstanding performance on LLM model supported by [LitGPT](https://github.com/Lightning-AI/lit-gpt), such as Mistral, Llama 2, Gemma, Falcon, and others.
+However, it can already deliver outstanding performance for pretraining and finetuning LLMs supported by [LitGPT](https://github.com/Lightning-AI/lit-gpt), such as Mistral, Llama 2, Gemma, Falcon, and others.
 
 Check out [the LitGPT integration](https://github.com/Lightning-AI/litgpt/tree/main/extensions/thunder) to learn about running LitGPT and Thunder together.
 
-## Features
+&#160;
+
+## Inside Thunder: A brief look at the core features
 
 Given a Python callable or PyTorch module, Thunder can generate an optimized program that:
 
@@ -140,13 +196,13 @@ Given a Python callable or PyTorch module, Thunder can generate an optimized pro
 To do so, Thunder ships with:
 
 - A JIT for acquiring Python programs targeting PyTorch and custom operations
-- A multi-level IR to represent operations as a trace of a reduced op-set
-- An extensible set of transformations on the trace, such as `grad`, fusions, distributed (like `ddp`, `fsdp`), functional (like `vmap`, `vjp`, `jvp`)
+- A multi-level intermediate representation (IR) to represent operations as a trace of a reduced operation set
+- An extensible set of transformations on the trace of a computational graph, such as `grad`, fusions, distributed (like `ddp`, `fsdp`), functional (like `vmap`, `vjp`, `jvp`)
 - A way to dispatch operations to an extensible collection of executors
 
 Thunder is written entirely in Python. Even its trace is represented as valid Python at all stages of transformation. This allows unprecedented levels of introspection and extensibility.
 
-Thunder doesn't generate code for accelerators directly. It acquires and transforms user programs so that it's possible to optimally select or generate device code using fast executors like:
+Thunder doesn't generate code for accelerators, such as GPUs, directly. It acquires and transforms user programs so that it's possible to optimally select or generate device code using fast executors like:
 
 - [torch.compile](https://pytorch.org/get-started/pytorch-2.0/)
 - [nvFuser](https://github.com/NVIDIA/Fuser)
@@ -159,6 +215,8 @@ Thunder doesn't generate code for accelerators directly. It acquires and transfo
 
 Modules and functions compiled with Thunder fully interoperate with vanilla PyTorch and support PyTorch's autograd. Also, Thunder works alongside torch.compile to leverage its state-of-the-art optimizations.
 
+&#160;
+
 ## Documentation
 
 Docs are currently not hosted publicly. However you can build them locally really quickly:
@@ -169,27 +227,15 @@ make docs
 
 and point your browser to the generated docs at `docs/build/index.html`.
 
-## Develop and run tests
-
-You can set up your environment for developing Thunder by installing the development requirements:
-
-```bash
-pip install -r requirements/devel.txt
-```
+&#160;
 
-Install Thunder as an editable package (optional):
-
-```bash
-pip install -e .
-```
+## Get involved!
 
-Now you run tests:
+We appreciate your feedback and contributions. If you have feature requests, questions, or want to contribute code or config files, please don't hesitate to use the [GitHub Issue](https://github.com/Lightning-AI/lightning-thunder/issues) tracker.
 
-```bash
-pytest thunder/tests
-```
+We welcome all individual contributors, regardless of their level of experience or hardware. Your contributions are valuable, and we are excited to see what you can accomplish in this collaborative and supportive environment.
 
-Thunder is very thoroughly tested, so expect this to take a while.
+&#160;
 
 ## License
 

diff --git a/docs/source/conf.py b/docs/source/conf.py
@@ -11,15 +11,13 @@
 # add these directories to sys.path here. If the directory is relative to the
 # documentation root, use os.path.abspath to make it absolute, like shown here.
 
-import glob
 import inspect
 import os
-import re
 import shutil
 import sys
 from importlib.util import module_from_spec, spec_from_file_location
 
-import pt_lightning_sphinx_theme
+import lai_sphinx_theme
 
 _PATH_HERE = os.path.abspath(os.path.dirname(__file__))
 _PATH_ROOT = os.path.realpath(os.path.join(_PATH_HERE, "..", ".."))
@@ -99,6 +97,7 @@ def _transform_changelog(path_in: str, path_out: str) -> None:
     "sphinx_copybutton",
     "sphinx_paramlinks",
     "sphinx_togglebutton",
+    "lai_sphinx_theme.extensions.lightning",
 ]
 
 # Add any paths that contain templates here, relative to this directory.
@@ -152,8 +151,8 @@ def _transform_changelog(path_in: str, path_out: str) -> None:
 # The theme to use for HTML and HTML Help pages.  See the documentation for
 # a list of builtin themes.
 #
-html_theme = "pt_lightning_sphinx_theme"
-html_theme_path = [pt_lightning_sphinx_theme.get_html_theme_path()]
+html_theme = "lai_sphinx_theme"
+html_theme_path = [lai_sphinx_theme.get_html_theme_path()]
 
 # Theme options are theme-specific and customize the look and feel of a theme
 # further.  For a list of options available for each theme, see the