A collection of foundation models are deployed in IBM watsonx.ai. You can prompt these models in the Prompt Lab.
| Model name | Size | Source | License | Instruction-tuned* | Use | Context length** | Details |
|---|---|---|---|---|---|---|---|
| flan-t5-xxl-11b | 11B | Hugging Face | Apache 2.0 | Yes | General use with zero- or few-shot prompts | 4096 |
• Model card • Flan paper |
| flan-ul2-20b | 20B | Hugging Face | Apache 2.0 | Yes | General use with zero- or few-shot prompts | 4096 |
• Model card • UL2 paper • Flan paper |
| gpt-neox-20b | 20B | Hugging Face | Apache 2.0 | No | • Works best with few-shot prompts • Works well with special characters, which can be useful for generating structured output |
8192 |
• Model card • Paper |
| llama-2-70b-chat | 70B | Hugging Face | License | Yes | • General use with zero- or few-shot prompts • Optimized for dialogue use cases |
4096 |
• Model card • Paper |
| mpt-7b-instruct2 | 7B | Hugging Face | Apache 2.0 | Yes | General use with few-shot prompts | 2048 | • Model card |
| mt0-xxl-13b | 13B | Hugging Face | Apache 2.0 | Yes | General use with zero- or few-shot prompts | 4096 |
• Model card • Paper |
| starcoder-15.5b | 15.5B | Hugging Face | License | No | Code generation and code translation Note: The model might return code from its training data, including licensed code that requires attribution. |
8192 |
• Model card • Paper |
* Instruction-tuned means the model has been fine-tuned with prompts that include an instruction. When a model has been instruction-tuned, that means it will probably respond well to prompts that have an instruction even if those prompts don't have examples. See: Prompt tips
** Context length, sometimes called context window length, context window, or maximum sequence length, is the maximum allowed value for: the number of tokens in the input prompt plus the number of tokens in the generated output from the model. See: Tokens and tokenization
Data center information
The following table shows the IBM Cloud data centers where each foundation model is available. A checkmark indicates that the model is hosted in the region.
| Model name | Dallas | Frankfurt |
|---|---|---|
| flan-t5-xxl-11b | ✓ | ✓ |
| flan-ul2-20b | ✓ | ✓ |
| gpt-neox-20b | ✓ | |
| llama-2-70b-chat | ✓ | |
| mpt-7b-instruct2 | ✓ | ✓ |
| mt0-xxl-13b | ✓ | |
| starcoder-15.5b | ✓ |
Learn more
Parent topic: Foundation models