Llama-Thinker-3B-Preview Model

Introduction

Llama-Thinker-3B-Preview is a pretrained and instruction-tuned generative model designed for multilingual applications, leveraging synthetic datasets for complex reasoning tasks. It supports several languages, including English, German, French, Italian, Portuguese, Hindi, Spanish, and Thai.

Architecture

The model is based on Llama 3.2 and operates as an autoregressive language model utilizing an optimized transformer architecture. It undergoes supervised fine-tuning (SFT) and reinforcement learning with human feedback (RLHF) to align with human preferences, enhancing its helpfulness and safety.

Training

The model is trained using synthetic datasets that facilitate long chains of thought, which enable effective performance in complex reasoning tasks. It is fine-tuned with techniques like SFT and RLHF.

Guide: Running Locally

To run the Llama-Thinker-3B-Preview model locally using Ollama, follow these steps:

Step 1: Download the Model

Execute the command:

ollama run llama-thinker-3b-preview.gguf

Step 2: Model Initialization and Download

Ollama will initialize and download the necessary model files, providing feedback during the process. Ensure successful download and initialization before proceeding.

Step 3: Interact with the Model

Once the model is loaded, interact with it by sending prompts, such as:

>>> How can you assist me today?

Step 4: Exit the Program

To exit, type:

/exit

Notes on using quantized models:

• Ensure adequate VRAM/CPU resources.
• Use the .gguf model format for compatibility with Ollama.

Cloud GPU Recommendation: For enhanced performance, consider utilizing cloud GPUs from providers like AWS, Google Cloud, or Azure.

License

This model is licensed under the CreativeML OpenRAIL-M license, promoting open access while ensuring ethical use.