[AzulGarza]

Adds a TimeH2O wrapper for H2O.ai's TimeH2O via the h2ofm package (fev-bench #3 by skill score, #1 by win rate; zero-shot on GIFT-Eval). Also adds replication on GIFT-Eval and fev-bench against the official results.

Library

• foundationforecast/models/timeh2o.py: forecast / CV / anomalies, quantiles and level, and native known-future exog (future_covariates) on h2oai/timeh2o-v1. Non-numeric covariates keep their dtype, so h2ofm encodes them as categorical. h2oai/timeh2o-v1-tiny has no covariate support and raises with X_df. Requested quantiles are passed straight through: h2ofm extrapolates outside its 0.1 to 0.9 grid with Gaussian tails.
• clip_nonnegative and snap_integer options (off by default, as in h2ofm). The official runs enable both.
• Dependencies: h2ofm>=0.43.0 (Python >= 3.12 only) and timecopilot-timesfm>=0.4.1, which drops the safetensors<0.6 cap that conflicted with h2ofm (TimeCopilot/timesfm#1).
• Tests: both checkpoints are registered in tests/models/conftest.py (py >= 3.12) for the cross-model suite. tests/models/test_timeh2o.py has mocked tests plus a live smoke test on tiny.
• Docs: model hub, API reference, exog table, README matrix and license notes.

Replication

• GIFT-Eval: h2oai--timeh2o-v1 and -tiny in configs/models.yaml, against results/TimeH2O-v1* in the official Space. Settings match the official notebook: context 16384, batch 64 / 128, clip_nonnegative and snap_integer on. CI subset: v1 on m4_weekly/short.
• fev-bench: h2oai--timeh2o-v1 against timeh2o-v1.csv, with the settings from the official models/timeh2o-v1/model.py. CI subset: epf_pjm and entsoe_1H.
• Modal Python per model: a models.yaml entry can set python: "3.12". Both harnesses build a 3.12 image with its own Modal function and route jobs by that field. Every other model stays on the existing 3.11 image, so their environment is unchanged.

Notes

• Gated, non-commercial weights. The HF_TOKEN accounts used by GitHub CI and by the Modal hf-secret must accept the terms on timeh2o-v1 and timeh2o-v1-tiny.
• Not run locally against real weights (no HF token here). Local runs covered the wrapper's mocked tests, the experiment config tests, the Modal app imports and pre-commit. The rest, including the Modal replication runs, is left to CI.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has insufficient remaining quota.

An organization owner can choose the repository organization as the billing source for reviews requested by members with a paid Copilot license. Learn about Copilot code review billing.

[codspeed]

Merging this PR will improve performance by 10.86%

1 warning, 1 note

⚠️ Base and head ran on different runner types

Some benchmarks with significant performance changes were compared across different runner types, so the changes might not come from the code. Run base and head on the same runner type, or use CodSpeed Macro Runners for more stable results.

Learn more about Macro Runners

Open the report in CodSpeed to investigate

ℹ️ 13 benchmarks have no flamegraph

The flamegraph wasn't generated: the symbols of the profiling trace could not be resolved, because no information about the language frames was collected.

See affected benchmarks

⚡ 1 improved benchmark
✅ 16 untouched benchmarks

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
⚡	CPU Simulation	test_timeseries_dataset_from_df_with_panel	45.5 µs	41 µs	+10.86%

Tip

Curious why performance improved? Comment @codspeedbot explain why performance improved on this PR, or directly use the CodSpeed MCP with your agent.

---

_{Comparing feat/timeh2o (4775113) with main (5d833ee)}

[codecov]

Codecov Report

❌ Patch coverage is 97.77778% with 2 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
foundationforecast/models/timeh2o.py	97.61%	2 Missing ⚠️

📢 Thoughts on this report? Let us know!

[codspeed]

Unable to generate the performance report

Your organization has exceeded its Macro Runners budget. Because this run uses tokenless mode, CodSpeed needs to wait for the GitHub Actions workflow to complete to validate the upload. Since the walltime jobs will never run, the workflow stays pending and we are unable to process this run.

To run these benchmarks again, update your plan or increase your Macro Runners budget.

If you need help, please contact us at support@codspeed.io or on Discord.