Stable LM Series by Stability AI
The Stable LM Series is Stability AI’s family of open-access large language models designed for efficient text generation, multilingual communication, reasoning tasks, and coding workflows. These models are optimized for both research and real-world applications, offering lightweight variants for low-resource environments and larger models for more complex tasks. Stability AI continues to support the development and deployment of these models through its GitHub and Hugging Face repositories.
Current Models in the Stable LM Series:
Open-Access Models:
StableLM 2 1.6B:
A compact multilingual model trained on 2 trillion tokens in seven languages (English, Spanish, German, Italian, French, Portuguese, Dutch). Offered in base and instruction-tuned variants. Optimized for research and lightweight applications with a 4K token context window.StableLM 2 12B:
A larger multilingual model designed for high performance in zero-shot and few-shot learning tasks, tool use, and retrieval-augmented generation (RAG). Offers strong generalization capabilities while maintaining efficiency. Supports 4K token context length.StableLM Zephyr 3B:
A 3B parameter instruction-tuned model fine-tuned from StableLM-2 3B using Direct Preference Optimization (DPO). Provides strong instruction-following behavior and conversational performance in a small footprint.
Coding Models:
Stable Code Instruct 3B:
A fine-tuned language model focused on code generation, bug fixing, and general software development tasks using natural language prompts.Stable Code Base (3B):
A base model version trained specifically on code datasets for flexible downstream tuning and deployment.
Licensing and Usage:
Base models are available under the Stability AI Non-Commercial Research Community License, requiring a Stability AI Membership for commercial use.
Instruction-tuned models (e.g., StableLM Zephyr) are intended for non-commercial research purposes only.
Key Features Across the Series:
Multilingual Support: Models cover English, Spanish, German, Italian, French, Portuguese, Dutch, and more.
Compact Deployment: Lightweight models like 1.6B and Zephyr 3B are optimized for local or mobile environments.
Coding Capabilities: Coding-specific models support tasks from auto-complete to debugging across multiple languages.
Open Access: Models are available via GitHub and Hugging Face for experimentation, fine-tuning, and integration.
Example Use Cases:
Building multilingual chatbots and AI assistants with StableLM 2 models.
Supporting developers with code generation and troubleshooting using Stable Code models.
Conducting academic and industrial research using instruction-tuned variants like Zephyr 3B.
Deploying efficient AI agents in resource-constrained environments with 1.6B parameter models.