Model Overview Card for AI21: Jamba 1.5 Mini

Basic Information

• Model Name: Jamba 1.5 Mini
• Developer/Creator: AI21 Labs
• Release Date: August 2024
• Version: 1.0
• Model Type: Large Language Model (LLM)

Description

Overview:

Jamba 1.5 Mini is a state-of-the-art hybrid SSM-Transformer model designed for high efficiency and performance in instruction-following tasks. It excels in processing long contexts and generating high-quality outputs, making it suitable for a variety of applications in natural language processing.

Key Features:

• Up to 2.5 times faster inference than comparable models.
• Handles long context lengths of up to 256,000 tokens.
• Supports multiple languages, including English, Spanish, French, Portuguese, Italian, Dutch, German, Arabic, and Hebrew.
• Optimized for business applications with features like function calling and structured output (JSON).
• Utilizes the innovative ExpertsInt8 quantization technique for efficient deployment on a single 80GB GPU.

Intended Use:

The model is designed for applications such as chatbots, customer service automation, content generation, and any scenario requiring efficient processing of extensive information.

Language Support:

Jamba 1.5 Mini supports multiple languages, enhancing its usability in global contexts.

Technical Details

Architecture:

Jamba 1.5 Mini is built on a hybrid SSM-Transformer architecture that combines Transformer layers with Mamba layers and a mixture-of-experts (MoE) module. Key architectural details include:

• Active Parameters: 12 billion
• Total Parameters: Approximately 52 billion
• Context Length: Up to 256K tokens
• Attention Heads: Configurable based on the model's architecture

Training Data:

The model was trained using a diverse dataset that emphasizes instruction-following capabilities and conversational contexts.

• Data Source and Size: The training dataset includes a wide range of texts from various domains to ensure robust language understanding.
• Knowledge Cutoff: March 5, 2024
• Diversity and Bias: The training data was curated to minimize bias while maximizing diversity in topics and languages, which enhances the model's robustness across different contexts.

Performance Metrics:

Jamba 1.5 Mini has shown competitive performance across various benchmarks:

Comparison to Other Models

Usage

Code Samples:

The model is available on the AI/ML API platform as "Jamba 1.5 Mini" .

API Documentation:

Detailed API Documentation is available here.

Ethical Guidelines

AI21 Labs emphasizes ethical considerations in AI development by promoting transparency regarding the model's capabilities and limitations. They encourage responsible usage to prevent misuse or harmful applications.

Licensing

Jamba 1.5 Mini is released under the Jamba Open Model License, allowing both commercial and non-commercial usage rights while ensuring compliance with ethical standards.

‍

Get Jamba 1.5 Mini API here.