-
Notifications
You must be signed in to change notification settings - Fork 3.9k
Reorganize TXT2KG and introduce torch_geometric.llm
#10436
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Conversation
Co-authored-by: riship <[email protected]>
for more information, see https://pre-commit.ci
for more information, see https://pre-commit.ci
for more information, see https://pre-commit.ci
for more information, see https://pre-commit.ci
for more information, see https://pre-commit.ci
for more information, see https://pre-commit.ci
Co-authored-by: riship <[email protected]> Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
for more information, see https://pre-commit.ci
for more information, see https://pre-commit.ci
… since NVIDIA owns this i want to make the linting a little more lax
for more information, see https://pre-commit.ci
akihironitta
left a comment
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM as per offline discussion. I left some final comments before merging.
Co-authored-by: Akihiro Nitta <[email protected]>
torch_geometric.llm
for more information, see https://pre-commit.ci
akihironitta
left a comment
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
As per our offline discussion amongst @rusty1s @wsad1 @puririshi98 and myself, we have decided to create a new subpackage torch_geometric.llm and assign @puririshi98 as its codeowner so that NVIDIA and the PyG community can iterate on the integration much more quickly even without needing to guarantee the same standard as the PyG core, e.g., thorough test coverage, documentation and code quality.
Thanks again @puririshi98 and the team for your patience and the exciting work! 🚀
replaces #9992
adds new torch_Geometric.llm for NVIDIA to mantain
example of how to use this in nvidia docker container:
git config --global credential.helper store; huggingface-cli login --token <insert_token>; cd /opt/pyg; pip uninstall -y torch-geometric; rm -rf pytorch_geometric; git clone -b latest-txt2kg https://github.com/pyg-team/pytorch_geometric.git; cd /opt/pyg/pytorch_geometric; pip install .; pip install openai
example to run:
python3 examples/llm/txt2kg_rag.py
thanks to @Kh4L @zaristei @rlratzel @rliu

for their contributions
add the RAG CI job back after figuring out why tests pass but the CI gives a red X
add a non-toy dataset for the default data (needs to be released by NVIDIA on huggingface, waiting on that)
further improvements to user friendliness of example
I will work with @zaristei on refining RAGQueryLoader and Feature/Graph Store workflow as directed by @wsad1
Note: NVIDIA CI will run all unit tests (including these rag related ones) as well as the full txt2kg_rag.py example everytime we update our NVIDIA container. This will run on H100, B100, and A100 a few diff skus. Then NVIDIA QA also runs on almost every hardware SKU before each bi-monthly release.
(closed subPR: #10368)
I had broken this PR down but then matthias said I could merge if he hasnt reviewed before i came back from vacay since we have tested and reviewed this so thoroughly at NVIDIA and externally on top of my involvement in most of the llm features of PyG.
NVIDIA's backing of this project spans many orgs all the way to the top and has fortune500 names adopting it left and right. Getting this merged in is the first step towards cementing PyG as THE framework for anything "Graph" at NVIDIA and I am leading this charge. We will continue to optimize as time goes and treat any issues that come up as p0s to fix asap. But merging this is an important first step.