--- title: Ensemble forecasting with IBM Granite time-series models description: IBM Research released Granite Timeseries Ensemble, a recipe combining four pre-trained Granite time-series models into one forecasting interface. It returns... canonical: https://daily.dev/posts/ensemble-forecasting-with-ibm-granite-time-series-models-vtj3xkopw twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Ensemble forecasting with IBM Granite time-series models | daily.dev og:description: IBM Research released Granite Timeseries Ensemble, a recipe combining four pre-trained Granite time-series models into one forecasting interface. It returns... og:url: https://daily.dev/posts/ensemble-forecasting-with-ibm-granite-time-series-models-vtj3xkopw og:image: https://api.daily.dev/og/posts/vTj3XKOPW.png og:image:alt: Ensemble forecasting with IBM Granite time-series models 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. # Ensemble forecasting with IBM Granite time-series models **[Hugging Face](https://daily.dev/sources/huggingface)** · 9 min read · 0 upvotes · 0 comments ## Summary IBM Research released Granite Timeseries Ensemble, a recipe combining four pre-trained Granite time-series models into one forecasting interface. It returns median and quantile forecasts with no retraining and no fitted ensemble weights. The default Probabilistic Ensemble pools quantile forecasts with equal member influence, while an alternative IQR Ensemble weights members by forecast spread. On 97 GIFT-Eval dataset configurations, the Probabilistic Ensemble scored the lowest MASE (0.683) among the six compared forecasts, edging out the best single member, PatchTST-FM-r2 (0.685). Install it with 'pip install granite-tsfm>=0.3.11' and load it via Hugging Face's QuantileEnsembleForecaster. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### How do I use the IBM Granite Timeseries Ensemble model for forecasting without training on my own data? Install granite-tsfm (version 0.3.11 or later) with pip, then load the recipe with QuantileEnsembleForecaster.from_pretrained('ibm-granite/granite-timeseries-ensemble-r1'). Calling the ensemble on a dataframe with timestamp and target columns returns prediction_outputs for the point forecast and quantile_outputs for quantile ranges, by default spanning 0.1 to 0.9, with no retraining needed. _Teams piloting zero-shot forecasting can track releases like this ensemble recipe on daily.dev._ ### Does combining multiple pretrained time-series forecasting models into an ensemble actually improve accuracy over the best single model? Yes, but the margin is small. On 97 GIFT-Eval dataset configurations, the Granite Probabilistic Ensemble scored a MASE of 0.683 versus 0.685 for the best individual member, PatchTST-FM-r2. The main benefit is robustness: comparable or better aggregate accuracy without needing to pick the right architecture in advance, since PatchTST-FM-r2 still has the lowest aggregate CRPS/MWQL score. _Developers comparing forecasting architectures can follow benchmark write-ups like this on daily.dev._ ### What's the difference between the Probabilistic Ensemble and the IQR Ensemble in Granite's time-series forecasting recipe? The Probabilistic Ensemble gives each of the four member models equal influence and pools their quantile forecasts in probability space, with no fitting step or hyperparameters to tune. The IQR Ensemble instead weights members by their forecast spread, giving narrower intervals more influence, approximating each member's interquartile range and using isotonic regression to prevent quantile crossing; it exposes a temperature setting and an optional weight cap. _Anyone deciding between ensemble aggregation strategies can keep up with releases like this via daily.dev._ ## Similar posts on daily.dev - [IBM releases SOTA Granite Time Series PatchTST-FM-r2 model with commercial-friendly license](https://daily.dev/posts/ibm-releases-sota-granite-time-series-patchtst-fm-r2-model-with-commercial-friendly-license-xbwkizgkr) · Hugging Face · 0 upvotes · 0 comments - [BigQuery TimesFM 3.0 with multivariate time-series forecasting](https://daily.dev/posts/bigquery-timesfm-3-0-with-multivariate-time-series-forecasting-75pgz1vwm) · Google Cloud · 0 upvotes · 0 comments - [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 --- Tags: [#python](https://daily.dev/tags/python), [#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/ensemble-forecasting-with-ibm-granite-time-series-models-vtj3xkopw) ```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":"Ensemble forecasting with IBM Granite time-series models","url":"https://daily.dev/posts/ensemble-forecasting-with-ibm-granite-time-series-models-vtj3xkopw","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/ensemble-forecasting-with-ibm-granite-time-series-models-vtj3xkopw"},"datePublished":"2026-10-22T07:51:39.830Z","dateModified":"2026-10-22T07:52:23.531Z","description":"IBM Research released Granite Timeseries Ensemble, a recipe combining four pre-trained Granite time-series models into one forecasting interface. It returns...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/97696397e759ea282096307dac0ebe36?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/97696397e759ea282096307dac0ebe36?_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/ensemble-forecasting-with-ibm-granite-time-series-models-vtj3xkopw","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"python,huggingface,time-series-forecasting","timeRequired":"PT9M"} {"@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":"Ensemble forecasting with IBM Granite time-series models"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/ensemble-forecasting-with-ibm-granite-time-series-models-vtj3xkopw#faq","mainEntity":[{"@type":"Question","name":"How do I use the IBM Granite Timeseries Ensemble model for forecasting without training on my own data?","acceptedAnswer":{"@type":"Answer","text":"Install granite-tsfm (version 0.3.11 or later) with pip, then load the recipe with QuantileEnsembleForecaster.from_pretrained('ibm-granite/granite-timeseries-ensemble-r1'). Calling the ensemble on a dataframe with timestamp and target columns returns prediction_outputs for the point forecast and quantile_outputs for quantile ranges, by default spanning 0.1 to 0.9, with no retraining needed. Teams piloting zero-shot forecasting can track releases like this ensemble recipe on daily.dev."}},{"@type":"Question","name":"Does combining multiple pretrained time-series forecasting models into an ensemble actually improve accuracy over the best single model?","acceptedAnswer":{"@type":"Answer","text":"Yes, but the margin is small. On 97 GIFT-Eval dataset configurations, the Granite Probabilistic Ensemble scored a MASE of 0.683 versus 0.685 for the best individual member, PatchTST-FM-r2. The main benefit is robustness: comparable or better aggregate accuracy without needing to pick the right architecture in advance, since PatchTST-FM-r2 still has the lowest aggregate CRPS/MWQL score. Developers comparing forecasting architectures can follow benchmark write-ups like this on daily.dev."}},{"@type":"Question","name":"What's the difference between the Probabilistic Ensemble and the IQR Ensemble in Granite's time-series forecasting recipe?","acceptedAnswer":{"@type":"Answer","text":"The Probabilistic Ensemble gives each of the four member models equal influence and pools their quantile forecasts in probability space, with no fitting step or hyperparameters to tune. The IQR Ensemble instead weights members by their forecast spread, giving narrower intervals more influence, approximating each member's interquartile range and using isotonic regression to prevent quantile crossing; it exposes a temperature setting and an optional weight cap. Anyone deciding between ensemble aggregation strategies can keep up with releases like this via daily.dev."}}]} ```