中文Sign up

Datadog/Toto-2.0-2.5B-FT

🤗 Hugging Face sourcetime-series-forecastingapache-2.02.5B params9.8 GBsafetensors✓ 2 checksumsupdated today
Submit in one command

Run it next to your model folder. It makes the torrent, checks your files against Hugging Face, and submits it. You just start seeding and paste your key from your account. It only reads your files and never changes them. Read the script first if you like.

curl -fsSL https://pirateface.co/package.sh | bash -s -- --repo Datadog/Toto-2.0-2.5B-FT ./model-folder
Needs a seeder →

Toto-2.0-2.5B-FT

[!WARNING] This is a benchmarking checkpoint, not a general-purpose model. Toto-2.0-2.5B-FT is the Toto 2.0 2.5B base model finetuned on the GIFT-Eval training split for our #2-on-GIFT-Eval-leaderboard submission. It is released for reproducibility only.

For real workloads, please use the base Toto 2.0 collection. The base checkpoints are pretrained without any public data, generalize to every benchmark we have evaluated, and are what we recommend deploying.

✨ What this is?

A single Toto 2.0 2.5B base checkpoint finetuned on a mix that includes the GIFT-Eval training split, used to probe how far the base model can be pushed on a single in-distribution benchmark.

On the full GIFT-Eval leaderboard (foundation models + finetuned + ensemble + agentic), Toto-2.0-2.5B-FT places #2 on CRPS rank, MASE rank, and #3 on raw CRPS / MASE, behind only the Toto 2.0 Family-and-Friends ensemble.

🔁 Finetuning recipe

Starting from a fully-decayed Toto-2.0-2.5B base checkpoint, we finetuned for 10,000 steps on a mix designed to expose the model to in-distribution structure without overfitting to GIFT-Eval alone:

Source Share
GIFT-Eval Pretrain 45%
Datadog 5-minute+ observability metrics 25%
GIFT-Eval train split 15%
Synthetic (TempoPFN) 10%
Datadog 10s observability metrics 2.5%
Datadog 60s observability metrics 2.5%

The public portion (45% GIFT-Eval Pretrain) is drawn from the Toto 1.0 mix of GIFT-Eval Pretrain and the Chronos pretraining corpus, and is non-leaking with respect to the GIFT-Eval test split.

NorMuon and AdamW learning rates were both dropped by roughly an order of magnitude from pretraining (to 0.05 and 0.001 respectively). All other architecture and inference settings match the base 2.5B model.

🔗 Additional Resources

📖 Citation

@misc{khwaja2026toto20timeseries,
      title={Toto 2.0: Time Series Forecasting Enters the Scaling Era}, 
      author={Emaad Khwaja and Chris Lettieri and Gerald Woo and Eden Belouadah and Marc Cenac and Guillaume Jarry and Enguerrand Paquin and Xunyi Zhao and Viktoriya Zhukov and Othmane Abou-Amal and Chenghao Liu and Ameet Talwalkar and David Asker},
      year={2026},
      eprint={2605.20119},
      archivePrefix={arXiv},
      primaryClass={cs.LG},
      url={https://arxiv.org/abs/2605.20119}, 
}