IBM has released its new Granite Time Series PatchTST-FM-r2 model, a foundation model designed for general-purpose zero-shot time-series forecasting. This approximately 385 million parameter model offers probabilistic forecasting, handles missing values, and achieves top performance on the GIFT-Eval benchmark among replicable, zero-shot models with permissive licenses. The model's architecture combines conformer blocks with self-attention and temporal convolution, and its weights, code, and benchmark results are publicly available under Apache 2.0 and OpenMDW 1.0 licenses. AI
IMPACT This release provides a powerful, commercially-friendly open-source tool for time-series forecasting, potentially accelerating adoption in various industries.
RANK_REASON Release of a new foundation model with benchmark results and open-source licensing.
Read on Mastodon — mastodon.social →
- Apache Software License 2.0
- Confluent
- GIFT-Eval
- Granite Time Series PatchTST-FM-r2
- Granite TSFM
- Hugging Face
- IBM
- OpenMDW 1.0
- PatchTST-FM-r1
AI-generated summary · Google Gemini · from 3 sources. How we write summaries →
