0 / 0
Foundation models built by IBM
Last updated: Feb 21, 2025
Foundation models built by IBM

In IBM watsonx.ai, you can use IBM foundation models that are built with integrity and designed for business.

Granite foundation models

The Granite family of IBM foundation models includes decoder-only models that can efficiently predict and generate language.

The models were built with trusted data that has the following characteristics:

  • Sourced from quality data sets in domains such as finance (SEC Filings), law (Free Law), technology (Stack Exchange), science (arXiv, DeepMind Mathematics), literature (Project Gutenberg (PG-19)), and more.
  • Compliant with rigorous IBM data clearance and governance standards.
  • Scrubbed of hate, abuse, and profanity, data duplication, and blocklisted URLs, among other things.

IBM is committed to building AI that is open, trusted, targeted, and empowering. For more information about contractual protections that are related to IBM indemnification, see the IBM Client Relationship Agreement and IBM watsonx.ai service description.

The following Granite models are available in watsonx.ai:

The following sections provide a short description and a few resources for learning about each model. For more information, see Supported foundation models.

granite-13b-chat-v2

General use model that is optimized for dialog use cases. This version of the model is able to generate longer, higher-quality responses with a professional tone. The model can recognize mentions of people and can detect tone and sentiment. This foundation model is available for you to deploy on demand on dedicated hardware for the exclusive use of your organization.

For prompting guidelines, see Prompting the granite-13b-chat-v2 foundation model from IBM.

For model specs, see Supported foundation models.

Model card: granite-13b-chat-v2 model card

Try it out:

For more information about how to deploy the model, see Deploying foundation models on-demand.

granite-13b-instruct-v2

General use model. This version of the model is optimized for classification, extraction, and summarization tasks. The model can recognize mentions of people and can summarize longer inputs.

For model specs, see Supported foundation models.

Model card: granite-13b-instruct-v2 model card

Try it out:

granite-8b-japanese

General use model that supports the Japanese language. This version of the model is based on the Granite Instruct model and is optimized for classification, extraction, and question-answering tasks in Japanese. You can also use the model for translation between English and Japanese.

For model specs, see Supported foundation models.

Model card: granite-8b-japanese model card

Try it out:

granite-20b-multilingual

General use model that supports the English, German, Spanish, French, and Portuguese languages. This version of the model is based on the Granite Instruct model and is optimized for classification, extraction, and question-answering tasks in multiple languages. You can also use the model for translation tasks.

For model specs, see Supported foundation models.

Model card: granite-20b-multilingual model card

Try it out:

Granite Code models

Instruction fine-tuned models that support code discussion, generation, and conversion. Use these foundation models for programmatic coding tasks. The Granite Code models are fine-tuned on a combination of instruction data to enhance instruction-following capabilities including logical reasoning and problem solving.

  • granite-3b-code-instruct
  • granite-8b-code-instruct
  • granite-20b-code-instruct
  • granite-34b-code-instruct

The Granite Code foundation models support 116 programming languages.

The following Granite Code foundation models are instruction-tuned versions of the granite-20b-code-base foundation model that are designed for text-to-SQL generation tasks.

  • granite-20b-code-base-schema-linking
  • granite-20b-code-base-sql-gen

Try them out by using the sample notebook, Inferencing with Granite Text-to-SQL Models sample notebook.

For more information, see the following topics:

Model cards:

Try them out:

Granite Instruct models

Lightweight and open-source third generation Granite models that are fine tuned on a combination of permissively licensed open-source and proprietary instruction data. The Granite Instruct language models designed to excel in instruction following tasks such as summarization, problem-solving, text translation, reasoning, funcion-calling, and more.

  • granite-3-2b-instruct
  • granite-3-8b-instruct

For more information, see the following resources:

Model cards:

Try them out:

Granite Guardian models

Granite Guardian models are fine tuned third generation Granite Instruct models trained on unique data comprising human annotations and synthetic data. The foundation models are useful for risk detection use cases which are applicable across a wide-range of enterprise applications.

  • granite-guardian-3-2b
  • granite-guardian-3-8b

For more information, see the following topics:

Model cards:

Try them out:

Granite time series models

IBM Granite time series foundation models are compact, pretrained models for multivariate time series forecasting from IBM Research, also known as Tiny Time Mixers (TTM).

The Granite time series models were trained on almost a billion samples of time series data from various domains, including electricity, traffic, manufacturing, and more. You can apply one of these pretrained models on your target data to get an initial forecast without having to train the model on your data. When given a set of historic, timed data observations, the Granite time series foundation models can apply their understanding of dynamic systems to forecast future data values.

The following time series foundation models are available for use in watsonx.ai:

  • granite-ttm-512-96-r2: Requires at least 512 data points per channel in the request.
  • granite-ttm-1024-96-r2: Requires at least 1,024 data points per channel in the request.
  • granite-ttm-1536-96-r2: Requires at least 1,536 data points per channel in the request.

The Granite time series models work best with data points in minute or hour intervals and generate a forecast dataset with up to 96 data points per time series, per target channel.

Try them out:

Learn more

IBM embedding models

Foundation models built by IBM include encoder-only models that specialize in natural language processing and text embedding tasks.

The following embedding models from IBM are available in watsonx.ai:

granite-embedding-107m-multilingual
A 384-dimension embedding model that converts text in multiple languages into text embeddings.
granite-embedding-278m-multilingual
A 384-dimension embedding model that converts text in multiple languages into text embeddings.
slate-125m-english-rtrvr-v2, slate-125m-english-rtrvr
768-dimension embedding models that convert text into text embeddings.
slate-30m-english-rtrvr-v2, slate-30m-english-rtrvr
384-dimension embedding models that convert text into text embeddings.

For more information about these models, see Supported encoder foundation models.

For more information about using IBM embedding models to convert sentences and passages into text embeddings, see Text embedding generation.

Natural Language Processing capabilities

IBM Slate models also power a set of libraries that you can use for common natural language processing (NLP) tasks, such as classification, entity extraction, sentiment analysis, and more.

For more information about how to use the NLP capabilities of the Slate models, see Watson NLP library.

Parent topic: Supported foundation models