feat: add support for T0 (theforecastingcompany/t0-alpha)#348
Conversation
|
|
There was a problem hiding this comment.
Pull request overview
Note
Copilot was unable to run its full agentic suite in this review.
Adds support for the T0 open-weights foundation forecaster, including dependency wiring, test gating by Python version, and documentation references.
Changes:
- Introduces
T0forecaster wrapper with quantile/level output support. - Adds
tfc-t0dependency with Python version markers and registers the model in test fixtures. - Documents the new model in the model hub and API docs.
Reviewed changes
Copilot reviewed 6 out of 7 changed files in this pull request and generated 4 comments.
Show a summary per file
| File | Description |
|---|---|
| timecopilot/models/foundation/t0.py | Adds the T0 forecaster implementation and forecasting logic. |
| tests/models/test_models.py | Adds an import-failure test for unsupported Python versions. |
| tests/models/conftest.py | Conditionally includes T0 in the models list for supported Python versions. |
| pyproject.toml | Adds tfc-t0 dependency with Python version markers. |
| docs/model-hub.md | Adds T0 to the model hub listing. |
| docs/api/models/foundation/models.md | Adds T0 to the API documentation module list. |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
| fcst_df[self.alias] = fcsts_np[..., median_idx].reshape(-1, 1) | ||
| if qc.quantiles is not None: | ||
| for q in qc.quantiles: | ||
| fcst_df[f"{self.alias}-q-{int(q * 100)}"] = fcsts_np[ | ||
| ..., pred_quantiles.index(q) | ||
| ].reshape(-1, 1) |
| out = model.predict( | ||
| self._to_context(batch), | ||
| horizon=h, | ||
| quantiles=pred_quantiles, | ||
| ) |
| finally: | ||
| del model | ||
| torch.cuda.empty_cache() |
| with self._get_model() as model: | ||
| for batch in tqdm(dataset): | ||
| out = model.predict( | ||
| self._to_context(batch), | ||
| horizon=h, | ||
| quantiles=pred_quantiles, | ||
| ) |
Adds The Forecasting Company's open-weights T0 foundation model as a foundation forecaster, via the tfc-t0 package (Python 3.11–3.13). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Regenerated after tfc-t0 0.1.2 (metadata-only floor relaxation) was published to PyPI. Resolves to tfc-t0 0.1.2 + einops 0.7.0 + jaxtyping 0.2.38 — the combo the T0 adapter is verified against. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|
Hi @AzulGarza — this is ready for your review when you have a moment 🙏 (I don't have permission to add you as a reviewer formally, hence the ping.) Quick status:
The Copilot auto-review left 4 comments; I checked each against the |
|
hey @GeoffNN! thanks for the pr, it looks great! could you take a look at |
Resolve uv.lock conflict by taking main's resolution (tirex>=0.1.1 bump, torchvision dropped, xlstm added) plus this branch's tfc-t0 addition. Verified content-identical to a full `uv lock` regeneration; `uv lock --check` passes. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Merge main into feat/add-t0 (resolve uv.lock conflict)
|
@AzulGarza should be good now! Thanks for the review 🙏 |
Summary
Adds T0, The Forecasting Company's open-weights time series foundation model, as a new foundation forecaster. T0 (
t0-alpha, ~102M params, Apache-2.0) is a decoder-style patch transformer producing probabilistic multi-horizon quantile forecasts — currently #1 on the fev-bench leaderboard (skill score 42.2) and CRPS 0.4941 on GIFT-Eval.(Disclosure: I work at The Forecasting Company.)
Changes
timecopilot/models/foundation/t0.py—T0forecaster wrapping thetfc-t0package, following the existing foundation-model pattern (_get_modelcontext manager,TimeSeriesDatasetbatching,QuantileConverterfor levels/quantiles). The model predicts 5 quantile knots and interpolates arbitrary requested levels in a single forward pass; the median is the point forecast. Ragged batches are left-padded with NaN, which T0 treats as missing.pyproject.toml—tfc-t0>=0.1.2for Python 3.11–3.13 (the package's supported range), mirroring the TiRex/uni2ts marker pattern.tests/models/conftest.py—T0(context_length=256, batch_size=2)added to the model matrix under a version guard, plus atest_t0_import_failsguard test mirroring TiRex/Sundial.Validation
Run locally on macOS / Python 3.12. The lock resolves to
tfc-t0 0.1.2 + einops 0.7.0 + jaxtyping 0.2.38:forecastwith defaults,quantiles=[0.1, 0.5, 0.9], andlevel=[80]on multi-series daily data: correct columns, finite values, monotone quantiles, median == point forecast.cross_validation(df, h=6): correct shape and columns.uv run pytest tests/models/test_models.py::test_t0_import_fails(skips on 3.11–3.13, asserts ImportError outside).ruff check+ruff format --checkpass on the new/changed files.Note
tfc-t00.1.2 (the version this depends on) is published to PyPI;uv.lockis regenerated and committed. Ready for review.🤖 Generated with Claude Code