--- title: IBM releases SOTA Granite Time Series PatchTST-FM-r2... description: IBM has released Granite Time Series PatchTST-FM-r2, a ~385M-parameter zero-shot forecasting model that ranks #2 overall on the GIFT-Eval leaderboard and #1... canonical: https://daily.dev/posts/ibm-releases-sota-granite-time-series-patchtst-fm-r2-model-with-commercial-friendly-license-xbwkizgkr twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: IBM releases SOTA Granite Time Series PatchTST-FM-r2 model with commercial-friendly license | daily.dev og:description: IBM has released Granite Time Series PatchTST-FM-r2, a ~385M-parameter zero-shot forecasting model that ranks #2 overall on the GIFT-Eval leaderboard and #1... og:url: https://daily.dev/posts/ibm-releases-sota-granite-time-series-patchtst-fm-r2-model-with-commercial-friendly-license-xbwkizgkr og:image: https://api.daily.dev/og/posts/XbWKizgKr.png og:image:alt: IBM releases SOTA Granite Time Series PatchTST-FM-r2 model with commercial-friendly license og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # IBM releases SOTA Granite Time Series PatchTST-FM-r2 model with commercial-friendly license **[Hugging Face](https://daily.dev/sources/huggingface)** · 8 min read · 0 upvotes · 0 comments ## Summary IBM has released Granite Time Series PatchTST-FM-r2, a ~385M-parameter zero-shot forecasting model that ranks #2 overall on the GIFT-Eval leaderboard and #1 among permissively licensed, commercial-friendly models (Apache 2.0 and OpenMDW 1.0). It builds on PatchTST-FM-r1 with a conformer-based architecture combining self-attention and temporal convolution, supports context lengths up to 8,192 steps, produces probabilistic forecasts via 99 quantiles, and handles missing-value imputation. Training data sources are documented (GiftEvalPretrain, KernelSynth-based synthetic data, TSMixup, and CauKer sequences) for transparency. Weights, architecture, inference pipeline, and reproduction code are open, and the model is usable via a simple Python pipeline or through streaming integration with Confluent Cloud via Apache Flink. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### What is new in IBM's PatchTST-FM-r2 time series model compared to PatchTST-FM-r1? PatchTST-FM-r2 replaces standard transformer blocks with conformer blocks that combine multi-head self-attention with temporal convolution (alternating kernel sizes 5,5,3,3), uses 50% overlapping patches with Hamming-window weighting and overlap-and-add forecasting, and expands from 20 to 30 blocks. The result is a ~385M-parameter model supporting context lengths up to 8,192 steps and 99-quantile probabilistic forecasts. _Track architecture upgrades like this conformer redesign as forecasting models evolve, via daily.dev._ ### How does IBM's PatchTST-FM-r2 rank on the GIFT-Eval time series forecasting benchmark? PatchTST-FM-r2 ranks second overall for both CRPS and MASE among replicable, zero-shot models on GIFT-Eval as of September 8, 2026, achieving a geometric-mean CRPS of 0.467 and MASE of 0.6846. It is the top-performing model in that category among models released under a permissive, commercial-friendly license (Apache 2.0 or OpenMDW 1.0), placing just behind TimesFM-3. _Compare forecasting model benchmarks like GIFT-Eval before picking one for production on daily.dev._ ### What license is IBM's Granite Time Series PatchTST-FM-r2 model released under? It is dual-licensed under Apache 2.0 and OpenMDW 1.0, and users may choose either license. OpenMDW is a Linux Foundation licensing framework designed specifically for AI models and related materials, and both options grant broad, permissive rights to use, modify, and distribute the model for commercial purposes. _Check licensing terms like these before adopting an open model on daily.dev._ ## Similar posts on daily.dev - [Real-Time Intelligence with IBM Time Series Models on Confluent](https://daily.dev/posts/real-time-intelligence-with-ibm-time-series-models-on-confluent-korfuoerp) · Hugging Face · 0 upvotes · 0 comments - [Toto 2.0: Time series forecasting enters the scaling era](https://daily.dev/posts/toto-2-0-time-series-forecasting-enters-the-scaling-era-t2iz3dkjh) · Datadog · 3 upvotes · 0 comments --- Tags: [#ai](https://daily.dev/tags/ai), [#ibm](https://daily.dev/tags/ibm), [#huggingface](https://daily.dev/tags/huggingface), [#time-series-forecasting](https://daily.dev/tags/time-series-forecasting) [View this post on daily.dev](https://daily.dev/posts/ibm-releases-sota-granite-time-series-patchtst-fm-r2-model-with-commercial-friendly-license-xbwkizgkr) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"IBM releases SOTA Granite Time Series PatchTST-FM-r2 model with commercial-friendly license","url":"https://daily.dev/posts/ibm-releases-sota-granite-time-series-patchtst-fm-r2-model-with-commercial-friendly-license-xbwkizgkr","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/ibm-releases-sota-granite-time-series-patchtst-fm-r2-model-with-commercial-friendly-license-xbwkizgkr"},"datePublished":"2026-09-09T15:41:09.068Z","dateModified":"2026-09-14T09:16:06.787Z","description":"IBM has released Granite Time Series PatchTST-FM-r2, a ~385M-parameter zero-shot forecasting model that ranks #2 overall on the GIFT-Eval leaderboard and #1...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/6b9ff8f9bd3153957c001657f6dc631a?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/6b9ff8f9bd3153957c001657f6dc631a?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"Hugging Face","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Hugging Face","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/f1f55c67d81a4330acf5b90b26b0c8e1","url":"https://daily.dev/sources/huggingface"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/ibm-releases-sota-granite-time-series-patchtst-fm-r2-model-with-commercial-friendly-license-xbwkizgkr","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"ai,ibm,huggingface,time-series-forecasting","timeRequired":"PT8M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Hugging Face","item":"https://daily.dev/sources/huggingface"},{"@type":"ListItem","position":3,"name":"IBM releases SOTA Granite Time Series PatchTST-FM-r2 model with commercial-friendly license"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/ibm-releases-sota-granite-time-series-patchtst-fm-r2-model-with-commercial-friendly-license-xbwkizgkr#faq","mainEntity":[{"@type":"Question","name":"What is new in IBM's PatchTST-FM-r2 time series model compared to PatchTST-FM-r1?","acceptedAnswer":{"@type":"Answer","text":"PatchTST-FM-r2 replaces standard transformer blocks with conformer blocks that combine multi-head self-attention with temporal convolution (alternating kernel sizes 5,5,3,3), uses 50% overlapping patches with Hamming-window weighting and overlap-and-add forecasting, and expands from 20 to 30 blocks. The result is a ~385M-parameter model supporting context lengths up to 8,192 steps and 99-quantile probabilistic forecasts. Track architecture upgrades like this conformer redesign as forecasting models evolve, via daily.dev."}},{"@type":"Question","name":"How does IBM's PatchTST-FM-r2 rank on the GIFT-Eval time series forecasting benchmark?","acceptedAnswer":{"@type":"Answer","text":"PatchTST-FM-r2 ranks second overall for both CRPS and MASE among replicable, zero-shot models on GIFT-Eval as of September 8, 2026, achieving a geometric-mean CRPS of 0.467 and MASE of 0.6846. It is the top-performing model in that category among models released under a permissive, commercial-friendly license (Apache 2.0 or OpenMDW 1.0), placing just behind TimesFM-3. Compare forecasting model benchmarks like GIFT-Eval before picking one for production on daily.dev."}},{"@type":"Question","name":"What license is IBM's Granite Time Series PatchTST-FM-r2 model released under?","acceptedAnswer":{"@type":"Answer","text":"It is dual-licensed under Apache 2.0 and OpenMDW 1.0, and users may choose either license. OpenMDW is a Linux Foundation licensing framework designed specifically for AI models and related materials, and both options grant broad, permissive rights to use, modify, and distribute the model for commercial purposes. Check licensing terms like these before adopting an open model on daily.dev."}}]} ```