Determining which models and optional images to mirror to your private container registry

About this task

Some images and models are not mirrored by default when you mirror the images for a service to your private container registry. In some cases, images or models are not mirrored because they are required only to enable optional features. In other cases, models are not mirrored because there are a large number of models to choose from, and mirroring all of the models would require a significant amount of storage.

In addition, if a service depends on a model that is available through watsonx.ai™ Inference foundation models component (the watsonx_ai_ifm component), the model is not mirrored by default when you mirror the images for the service.

This task also provides the group that you must specify to mirror each model or image. Use this information to populate the IMAGE_GROUPS environment variable.

Procedure

Review the following table to determine whether the services that you plan to install use models and whether the models are optional or required.

Models for IBM Knowledge Catalog Premium

If you plan to enable gen AI based features, you must mirror the specified model.

Models for IBM Knowledge Catalog Standard

If you plan to enable gen AI based features, you must mirror the specified model.

Optional images for Orchestration Pipelines

You can choose whether to mirror the optional images for Orchestration Pipelines:

Models for Watson Speech to Text

You can choose which language models to mirror.

Models for Watson Text to Speech

You can choose which language models to mirror.

Models for watsonx.ai

You can choose which models you mirror.

• Foundation models
• Embedding models
• Reranker models
• Text extraction models
• Time series models
• LoRA and QLoRA fine tuning

Foundation models

Model	Description	Image group
allam-1-13b-instruct Status: Available	A bilingual large language model for Arabic and English that is initialized with Llama-2 weights and is fine-tuned to support conversational tasks.	ibmwxAllam113bInstruct
codestral-2501 Status: Available	Ideal for complex tasks that require large reasoning capabilities or are highly specialized. Attention: You must purchase Mistral AI with IBM to download and use this model.	ibmwxCodestral2501
codestral-2508 Status: Available 5.2.2 and later	Ideal for code generation and high-precision fill-in-the-middle (FIM) completion. The foundation model is optimized for production engineering environments such as latency-sensitive, context-aware, and self-deployable. Attention: You must purchase Mistral AI with IBM to download and use this model.	ibmwxCodestral2508
codestral-22b Status: Deprecated	Ideal for complex tasks that require large reasoning capabilities or are highly specialized. Attention: You must purchase Mistral AI with IBM to download and use this model.	ibmwxCodestral22B
devstral-medium-2507 Status: Available 5.2.2 and later	The devstral-medium-2507 foundation model is a high-performance code generation and agentic reasoning model. Ideal for generalization across prompt styles and tool use in code agents and frameworks. Attention: You must purchase Mistral AI with IBM to download and use this model.	ibmwxDevstralMedium2507
elyza-japanese-llama-2-7b-instruct Status: Withdrawn 5.2.2 Withdrawn in Version 5.2.2.	General use with zero- or few-shot prompts. Works well for classification and extraction in Japanese and for translation between English and Japanese. Performs best when prompted in Japanese.	ibmwxElyzaJapaneseLlama27bInstruct
flan-t5-xl-3b Status: Deprecated	General use with zero- or few-shot prompts.	ibmwxGoogleFlanT5xl
flan-t5-xxl-11b Status: Withdrawn 5.2.2 Withdrawn in Version 5.2.2.	General use with zero- or few-shot prompts.	ibmwxGoogleFlanT5xxl
flan-ul2-20b Status: Withdrawn 5.2.2 Withdrawn in Version 5.2.2.	General use with zero- or few-shot prompts.	ibmwxGoogleFlanul2
gpt-oss-20b Status: Available 5.2.2 and later	The gpt-oss foundation models are OpenAI’s open-weight models designed for powerful reasoning, agentic tasks, fine-tuning, and various developer use cases.	ibmwxGptOss20B
gpt-oss-120b Status: Available 5.2.2 and later	The gpt-oss foundation models are OpenAI’s open-weight models designed for powerful reasoning, agentic tasks, fine-tuning, and various developer use cases.	ibmwxGptOss120B
granite-4-h-micro Status: Available 5.2.2 and later	The Granite 4.0 foundation models belong to the IBM Granite family of models. The granite-4-h-micro is a 3 billion parameter foundation model built for structured and long-context capabilities. The model is ideal for instruction following and tool-calling capabilities.	ibmwxGranite4HMicro
granite-4-h-small Status: Available 5.2.2 and later	The Granite 4.0 foundation models belong to the IBM Granite family of models. The granite-4-h-small is 30 billion parameter foundation model built for structured and long-context capabilities. The model is ideal for instruction following and tool-calling capabilities.	ibmwxGranite4HSmall
granite-7b-lab Status: Withdrawn 5.2.2 Withdrawn in Version 5.2.2.	InstructLab foundation model from IBM that supports knowledge and skills contributed by the open source community.	ibmwxGranite7bLab
granite-8b-japanese Status: Withdrawn 5.2.0 Withdrawn in Version 5.2.0.	This image cannot be mirrored.	This image cannot be mirrored.
granite-13b-instruct-v2 Status: Deprecated	General use model from IBM that is optimized for question and answer use cases.	ibmwxGranite13bInstructv2
granite-3-3-8b-instruct Status: Available	An IBM-trained, dense decoder-only model, which is particularly well-suited for generative tasks.	ibmwxGranite338BInstruct
granite-3-2-8b-instruct Status: Available	A text-only model that is capable of reasoning. You can choose whether reasoning is enabled, based on your use case.	ibmwxGranite328BInstruct
granite-3-2b-instruct Status: Available	Granite models are designed to be used for a wide range of generative and non-generative tasks. They employ a GPT-style decoder-only architecture, with additional innovations from IBM Research and the open-source community.	ibmwxGranite32BInstruct
granite-3-8b-instruct Status: Available	Granite models are designed to be used for a wide range of generative and non-generative tasks. They employ a GPT-style decoder-only architecture, with additional innovations from IBM Research and the open-source community.	ibmwxGranite38BInstruct
granite-guardian-3-2b Status: Available	Granite models are designed to be used for a wide range of generative and non-generative tasks. They employ a GPT-style decoder-only architecture, with additional innovations from IBM Research and the open-source community.	ibmwxGraniteGuardian32b
granite-guardian-3-8b Status: Available	Granite models are designed to be used for a wide range of generative and non-generative tasks. They employ a GPT-style decoder-only architecture, with additional innovations from IBM Research and the open-source community.	ibmwxGraniteGuardian38b
granite-guardian-3-2-5b Status: Available	The Granite model series is a family of IBM-trained, dense decoder-only models, which are particularly well suited for generative tasks. This model cannot be used through the API.	ibmwxGraniteGuardian325b
granite-3b-code-instruct Status: Available	A 3-billion parameter instruction fine-tuned model from IBM that supports code discussion, generation, and conversion.	ibmwxGranite3bCodeInstruct
granite-8b-code-instruct Status: Available	A 8-billion parameter instruction fine-tuned model from IBM that supports code discussion, generation, and conversion.	ibmwxGranite8bCodeInstruct
granite-20b-code-instruct Status: Available	A 20-billion parameter instruction fine-tuned model from IBM that supports code discussion, generation, and conversion.	ibmwxGranite20bCodeInstruct
granite-20b-code-base-schema-linking Status: Available	Granite models are designed to be used for a wide range of generative and non-generative tasks. They employ a GPT-style decoder-only architecture, with additional innovations from IBM Research and the open-source community.	ibmwxGranite20bCodeBaseSchemaLinking
granite-20b-code-base-sql-gen Status: Available	Granite models are designed to be used for a wide range of generative and non-generative tasks. They employ a GPT-style decoder-only architecture, with additional innovations from IBM Research and the open-source community.	ibmwxGranite20bCodeBaseSqlGen
granite-34b-code-instruct Status: Available	A 34-billion parameter instruction fine-tuned model from IBM that supports code discussion, generation, and conversion.	ibmwxGranite34bCodeInstruct
granite-vision-3-2-2b Status: Deprecated	Granite 3.2 Vision is a image-text-in, text-out model capable of understanding images like charts for enterprise use cases for computer vision tasks.	ibmwxGraniteVision322B
granite-vision-3-3-2b Status: Available	Granite 3.2 Vision is an image-text-in, text-out model capable of understanding images like charts for enterprise use cases for computer vision tasks.	ibmwxGraniteVision322B
ibm-defense-3-3-8b-instruct Status: Available 5.2.2 and later	The IBM watsonx.ai Defense Model is a specialized fine-tuned version of IBM’s granite-3-3-8b-instruct base model. The model is developed through Janes trusted open-source defense data to support defense and intelligence operations.	ibmwxDefensemodel
jais-13b-chat Status: Deprecated	General use foundation model for generative tasks in Arabic.	ibmwxCore42Jais13bChat
llama-4-maverick-17b-128e-instruct-fp8 Status: Available	The Llama 4 collection of models are natively multimodal AI models that enable text and multimodal experiences. These models leverage a mixture-of-experts architecture to offer industry-leading performance in text and image understanding.	ibmwxLlama4Maverick17B128EInstructFp8
llama-4-maverick-17b-128e-instruct-int4 Status: Available	The Llama 4 collection of models are multimodal AI models that enable text and multimodal experiences. These models leverage a mixture-of-experts architecture to offer industry-leading performance in text and image understanding.	ibmwxLlama4Maverick17B128EInstructInt4
llama-4-scout-17b-16e-instruct Status: Deprecated	The Llama 4 collection of models are natively multimodal AI models that enable text and multimodal experiences. These models leverage a mixture-of-experts architecture to offer industry-leading performance in text and image understanding.	ibmwxLlama4Scout17B16EInstruct
llama-4-scout-17b-16e-instruct-int4 Status: Available	The Llama 4 collection of models are natively multimodal AI models that enable text and multimodal experiences. These models leverage a mixture-of-experts architecture to offer industry-leading performance in text and image understanding.	ibmwxLlama4Scout17b16eInstructInt4
llama-3-3-70b-instruct Status: Available	A state-of-the-art refresh of the Llama 3.1 70B Instruct model that uses the latest advancements in post-training techniques.	ibmwxLlama3370BInstruct
llama-3-2-1b-instruct Status: Available	A pretrained and fine-tuned generative text model with 1 billion parameters, optimized for multilingual dialogue use cases and code output.	ibmwxLlama321bInstruct
llama-3-2-3b-instruct Status: Available	A pretrained and fine-tuned generative text model with 3 billion parameters, optimized for multilingual dialogue use cases and code output.	ibmwxLlama323bInstruct
llama-3-2-11b-vision-instruct Status: Available	A pretrained and fine-tuned generative text model with 11 billion parameters, optimized for multilingual dialogue use cases and code output.	ibmwxLlama3211bVisionInstruct
llama-3-2-90b-vision-instruct Status: Available	A pretrained and fine-tuned generative text model with 90 billion parameters, optimized for multilingual dialogue use cases and code output.	ibmwxLlama3290bVisionInstruct
llama-guard-3-11b-vision Status: Available	A pretrained and fine-tuned generative text model with 11 billion parameters, optimized for multilingual dialogue use cases and code output.	ibmwxLlamaGuard311bVision
llama-3-1-8b-instruct Status: Available	An auto-regressive language model that uses an optimized transformer architecture.	ibmwxLlama318bInstruct
llama-3-1-70b-instruct Status: Available	An auto-regressive language model that uses an optimized transformer architecture.	ibmwxLlama3170bInstruct
llama-3-405b-instruct Status: Deprecated	Meta's largest open-sourced foundation model to date, with 405 billion parameters, and optimized for dialogue use cases.	ibmwxLlama3405bInstruct
llama-3-8b-instruct Status: Withdrawn 5.2.2 Withdrawn in Version 5.2.2.	Pretrained and instruction tuned generative text model optimized for dialogue use cases.	ibmwxMetaLlamaLlama38bInstruct
llama-3-70b-instruct Status: Withdrawn 5.2.2 Withdrawn in Version 5.2.2.	Pretrained and instruction tuned generative text model optimized for dialogue use cases.	ibmwxMetaLlamaLlama370bInstruct
llama-2-13b-chat Status: Deprecated	General use with zero- or few-shot prompts. Optimized for dialogue use cases.	ibmwxMetaLlamaLlama213bChat
llama2-13b-dpo-v7 Status: Withdrawn 5.2.0 Withdrawn in Version 5.2.0.	This image cannot be mirrored.	This image cannot be mirrored.
ministral-8b-instruct Status: Available	Ideal for complex tasks that require large reasoning capabilities or are highly specialized. Attention: You must purchase Mistral AI with IBM to download and use this model.	ibmwxMinistral8BInstruct
mistral-medium-2505 Status: Available 5.2.2 and later	Mistral Medium 3 features multimodal capabilities and an extended context length of up to 128k. The model can process and understand visual inputs, long documents and supports many languages.	ibmwxMistralMedium2505
mistral-small-3-1-24b-instruct-2503 Status: Available	Building upon Mistral Small 3 (2501), Mistral Small 3.1 (2503) adds state-of-the-art vision understanding and enhances long context capabilities and is suitable for function calling and agents.	ibmwxMistralSmall3124BInstruct2503
mistral-small-3-2-24b-instruct-2506 Status: Available 5.2.2 and later	The mistral-small-3-2-24b-instruct-2506 foundation model is an enhancement to mistral-small-3-1-24b-instruct-2503, with better instruction following and tool calling performance.	ibmwxMistralSmall3224BInstruct2506
mistral-small-24b-instruct-2501 Status: Deprecated	Mistral Small 3 (2501) sets a new benchmark in the small Large Language Models category with less than 70 billion parameters. With a size of 24 billion parameters, the model achieves state-of-the-art capabilities comparable to larger models.	ibmwxMistralSmall24BInstruct2501
mistral-small-instruct Status: Deprecated	Ideal for complex tasks that require large reasoning capabilities or are highly specialized. Attention: You must purchase Mistral AI with IBM to download and use this model.	ibmwxMistralSmallInstruct
mistral-large-instruct-2411 Status: Available	The most advanced Large Language Model (LLM) developed by Mistral Al with state-of-the-art reasoning capabilities that can be applied to any language-based task, including the most sophisticated ones. Attention: You must purchase Mistral AI with IBM to download and use this model.	ibmwxMistralLargeInstruct2411
mistral-large Status: Deprecated	The most advanced Large Language Model (LLM) developed by Mistral Al with state-of-the-art reasoning capabilities that can be applied to any language-based task, including the most sophisticated ones. Attention: You must purchase Mistral AI with IBM to download and use this model.	ibmwxMistralLarge
mistral-medium-2505 Status: Available	Mistral Medium 3 features multimodal capabilities and an extended context length of up to 128k. The model can process and understand visual inputs, long documents and supports many languages. Attention: You must purchase Mistral AI with IBM to download and use this model.	ibmwxMistralMedium2505
mixtral-8x7b-instruct-v01 Status: Deprecated	The Mixtral-8x7B Large Language Model (LLM) is a pretrained generative Sparse Mixture of Experts.	ibmwxMistralaiMixtral8x7bInstructv01
mt0-xxl-13b Status: Withdrawn 5.2.2 Withdrawn in Version 5.2.2.	General use with zero- or few-shot prompts. Supports prompts in languages other than English and multilingual prompts.	ibmwxBigscienceMt0xxl
pixtral-large-instruct-2411 Status: Available	A a 124-billion multimodal model built on top of Mistral Large 2, and demonstrates frontier-level image understanding. Attention: You must purchase Mistral AI with IBM to download and use this model.	ibmwxPixtralLargeInstruct
pixtral-12b Status: Available	A 12-billion parameter model pretrained and fine-tuned for generative tasks in text and image domains. The model is optimized for multilingual use cases and provides robust performance in creative content generation.	ibmwxPixtral12b
voxtral-small-2507 Status: Available 5.2.2 and later	Voxtral Small is an enhancement of Mistral Small 3.1, incorporating state-of-the-art audio capabilities and text performance, capable of processing up to 30 minutes of audio.	ibmwxVoxtralSmall24B2507

Embedding models

Model	Description	Image group
all-minilm-l6-v2 Status: Available	Use all-minilm-l6-v2 as a sentence and short paragraph encoder. Given an input text, the model generates a vector that captures the semantic information in the text.	ibmwxAllMinilmL6V2
all-minilm-l12-v2 Status: Available	Use all-minilm-l12-v2 as a sentence and short paragraph encoder. Given an input text, the model generates a vector that captures the semantic information in the text.	ibmwxAllMinilmL12V2
granite-embedding-107m-multilingual Status: Available	A 107 million parameter model from the Granite Embeddings suite provided by IBM. The model can be used to generate high quality text embeddings for a given input like a query, passage, or document.	ibmwxGranite107MMultilingualRtrvr
granite-embedding-278m-multilingual Status: Available	A 278 million parameter model from the Granite Embeddings suite provided by IBM. The model can be used to generate high quality text embeddings for a given input like a query, passage, or document.	ibmwxGranite278MMultilingualRtrvr
granite-embedding-english-reranker-r2 Status: Available 5.2.2 and later	A 149 million parameter model from the Granite Embeddings suite provided by IBM. The model has been trained for passage reranking, based on the granite-embedding-english-r2 to use in RAG pipelines.	ibmwxGraniteEmbeddingEnglishRerankerR2
multilingual-e5-large Status: Available	An embedding model built by Microsoft and provided by Hugging Face. The multilingual-e5-large model is useful for tasks such as passage or information retrieval, semantic similarity, bitext mining, and paraphrase retrieval.	ibmwxMultilingualE5Large
slate-30m-english-rtrvr Status: Available	The IBM provided slate embedding models are built to generate embeddings for various inputs such as queries, passages, or documents.	ibmwxSlate30mEnglishRtrvr
slate-125m-english-rtrvr Status: Available	The IBM provided slate embedding models are built to generate embeddings for various inputs such as queries, passages, or documents.	ibmwxSlate125mEnglishRtrvr

Reranker models

Model	Description	Image group
ms-marco-MiniLM-L-12-v2 Status: Available	A reranker model built by Microsoft and provided by Hugging Face. Given query text and a set of document passages, the model ranks the list of passages from most-to-least related to the query.	ibmwxMsMarcoMinilmL12V2

Text extraction models

Model	Description	Image group
wdu Status: Available	A set of text extraction models that are represented by the "wdu" identifier.	The models are always included when you mirror the images for watsonx.ai.

Time series models

Model	Description	Image group
granite-ttm-512-96-r2 Status: Available	The Granite time series models are compact pretrained models for multivariate time series forecasting from IBM Research, also known as Tiny Time Mixers (TTM). The models work best with data points in minute or hour intervals and generate a forecast dataset with 96 data points per channel by default.	ibmwxGraniteTimeseriesTtmV1
granite-ttm-1024-96-r2 Status: Available	The Granite time series models are compact pretrained models for multivariate time series forecasting from IBM Research, also known as Tiny Time Mixers (TTM). The models work best with data points in minute or hour intervals and generate a forecast dataset with 96 data points per channel by default.	ibmwxGraniteTimeseriesTtmV1
granite-ttm-1536-96-r2 Status: Available	The Granite time series models are compact pretrained models for multivariate time series forecasting from IBM Research, also known as Tiny Time Mixers (TTM). The models work best with data points in minute or hour intervals and generate a forecast dataset with 96 data points per channel by default.	ibmwxGraniteTimeseriesTtmV1

LoRA and QLoRA fine tuning

Model	Description	Image group
granite-3-1-8b-base Status: Available	Granite 3.1 8b base is a pretrained autoregressive foundation model with a context length of 128k intended for tuning. Fine tuning method: LoRA	ibmwxGranite318BBase
llama-3-1-8b-instruct Status: Available	An auto-regressive language model that uses an optimized transformer architecture. Fine tuning method: LoRA	ibmwxLlama318bInstruct
llama-3-1-70b-instruct Status: Available	An auto-regressive language model that uses an optimized transformer architecture. Fine tuning method: LoRA	ibmwxLlama3170bInstruct
llama-3-1-70b-gptq Status: Available	Llama 3.1 70b is a pretrained and fine-tuned generative text base model with 70 billion parameters, optimized for multilingual dialogue use cases and code output. Fine tuning method: QLoRA	ibmwxLlama3170BGptq

Foundation models for watsonx Assistant

You can choose which foundation models you mirror based on the features that you want to use:

Foundation models for watsonx BI

You must mirror the required foundation models.

Foundation models for watsonx Code Assistant

You must mirror the required foundation models.

Foundation models for watsonx Code Assistant for Red Hat Ansible Lightspeed

You can choose whether to mirror the optional model.

Foundation models for watsonx.data Premium

You must mirror all of the required models.

Foundation models for watsonx.data intelligence

If you plan to enable gen AI based features on CPU or GPU, you must mirror the specified model. (You do not need to mirror the specified model if you plan to connect to a remote instance of watsonx.ai where the model is running.)

Foundation models for watsonx Orchestrate

You can choose which foundation models you mirror based on the features that you want to use: