Skip to content

Add TabICLv2 inference-acceleration benchmark (c0-c19, c21) + serving docs #727

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.

Sign up for GitHub

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

Open
puririshi98 wants to merge 1 commit into from
Open

Add TabICLv2 inference-acceleration benchmark (c0-c19, c21) + serving docs #727

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 4 additions & 0 deletions .github/workflows/test-gpu.yml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -103,6 +103,10 @@ jobs:
print("pyg-lib", version("pyg-lib"))
PY

- name: Benchmark import smoke
run: |
PYTHONPATH=examples python -c 'import benchmark_tabiclv2'

- name: Run tests
env:
FULL_TEST: ${{ needs.trigger.outputs.full_test == 'true' && '1' || '0' }}
Expand Down
2 changes: 1 addition & 1 deletion docs/source/icl.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -74,7 +74,7 @@ Use one-shot {py:meth}`~sdm.models.ICLModel.forward` calls for one-time calls wh
Unlike {py:meth}`~sdm.models.ICLModel.forward`, {py:meth}`~sdm.models.ICLModel.predict` does not support gradient-based fine-tuning and raises if the model is in train mode.

Both calls accept padded inputs: pad the in-context rows and the columns to a fixed set of shapes and pass the valid counts as `seqused_train` (per batch element) and `seqused_cols` (shared) `torch.int32` tensors together with a pass-through `sdm.Recipe()`.
The {py:attr}`~sdm.models.ICLModel.supports_seqused` attribute denotes whether an {py:class}`~sdm.models.ICLModel` accepts padded inputs; padded rows and columns are masked from attention, so a stream of differently sized tables can reuse the same compiled graphs.
The {py:attr}`~sdm.models.ICLModel.supports_seqused` attribute denotes whether an {py:class}`~sdm.models.ICLModel` accepts padded inputs; padded rows and columns are masked from attention, so a stream of differently sized tables can reuse the same compiled graphs; [`examples/benchmark_tabiclv2.py`](https://github.com/NVIDIA/structured-data-models/blob/main/examples/benchmark_tabiclv2.py) measures this serving recipe.

## Model Concepts

Expand Down
4 changes: 4 additions & 0 deletions examples/README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -5,4 +5,8 @@ This folder contains runnable examples of `structured-data-models`:
- [**`tabular/`**](tabular/): Tabular foundation model examples
- [**`relational/`**](relational/): Relational foundation model examples

Inference-serving benchmark drivers (the dataset benchmark suites live under [`benchmark/`](../benchmark/)):

- [**`benchmark_tabiclv2.py`**](benchmark_tabiclv2.py): inference acceleration benchmark for `sdm.models.TabICLv2`, sweeping NVIDIA-recommended serving recipes. Its module docstring doubles as the serving notes (bucketed shapes, regional compilation and its raised `recompile_limit`, compile-cache artifacts).

See [`tabular/quickstart.py`](tabular/quickstart.py) for a minimal runnable `structured-data-models` example with `TabICLv2`.
Loading
Loading