Jamba 1.5 Large (Preview)
AI21's most powerful instruction-tuned Foundation Model with 256K context window that's optimized for long-form input, superior accuracy, and speed.AI21 Labs Jamba 1.5 Large is a foundation model built from groundbreaking hybrid architecture, leveraging both the novel Mamba architecture and traditional Transformer architecture to achieve leading quality at the best price.
By drawing on its SSM-Transformer hybrid architecture, as well as its impressive 256K context window, Jamba 1.5 Large efficiently solves a variety of text generation and comprehension use cases for the enterprise. Its 94B active parameters and 398B total parameters lead to superior accuracy in responses.
Jamba 1.5 Large is ideal for enterprise workflows with tasks that are data-heavy and require a model to be able to ingest a large amount of information in order to produce an accurate and thorough response, such as summarizing lengthy documents or enabling question answering across an extensive organizational knowledge base. Jamba 1.5 Large is a model designed for superior quality responses, high throughput and attractive price compared to other models in its size class.
AI21 Labs has other models. View the options:
Characteristic | Jamba 1.5 Mini | Jamba 1.5 Large |
---|---|---|
Model size | Small Language Model (SLM) | Large Language Model (LLM) |
Architecture | Mamba+Transformer hybrid | Mamba+Transformer hybrid |
Context window length | Up to 256K | Up to 256K |
# Parameters | 12B active/52B total | 94B active/398B total |
Accuracy | High | Highest |
Speed | High | Highest |
Cost | Most cost-efficient | Cost-efficient |
Best for | Use cases needing large input context, quality results, low latency but cost and TCO are a major consideration. Attractive price point vs other non-AI21 models in small model category. | Use cases needing large input context, superior quality results, low latency, and high throughput & scalability. Attractive price point vs other non-AI21 models in large model category. |
A long context model, like AI21 Labs’ Jamba 1.5 Large, which has an effective context window of 256K, can help enterprises with the following:
Jamba 1.5 Large supports such use cases well, especially where high quality answers, response time, and throughput scalability are paramount.
Enable the Vertex AI API.
Authenticate with one of the standard mechanisms documented here.
Choose how you will call Jamba:
Configure environment variables by entering the following. Replace YOUR_PROJECT_ID with the ID of your Google Cloud project.
Send a prompt request by leveraging the following format:
The Request JSON body should be in this format:
Combine the post request and body in the Shell:
An example response body:
Sample Input
Sample Output
Category | Metric | Score |
---|---|---|
General | Arena Hard | 65.4 |
MMLU (CoT) | 81.2 | |
MMLU Pro (CoT) | 53.5 | |
IFEval | 81.5 | |
BBH | 65.5 | |
WildBench | 48.4 | |
Reasoning | ARC-C | 93 |
GPQA | 36.9 | |
Math, Code & Tool use | GSM8K | 87 |
HumanEval+ | 71.3 | |
BFCL | 85.5 |
Resource ID | Release date | Release stage | Description |
---|---|---|---|
jamba-1.5-large@001 | 2024-08-22 | Public Preview | Model release |
Google Cloud Console has failed to load JavaScript sources from www.gstatic.com.
Possible reasons are: