Triangulum 1 B G G U F

Introduction

Triangulum-1B is a collection of pretrained and instruction-tuned generative models designed for multilingual applications. They are optimized for complex reasoning tasks, leveraging synthetic datasets based on long chains of thought.

Architecture

• Foundation Model: Built on LLaMA's autoregressive language model using an optimized transformer architecture for enhanced performance.
• Instruction Tuning: Incorporates supervised fine-tuning (SFT) and reinforcement learning with human feedback (RLHF) to ensure outputs align with human preferences for helpfulness and safety.
• Multilingual Support: Capable of handling multiple languages, applicable across diverse linguistic contexts.

Training

• Synthetic Datasets: Utilizes long chain-of-thought synthetic data to boost reasoning capabilities.
• Supervised Fine-Tuning (SFT): Aligns the model to specific tasks using curated datasets.
• Reinforcement Learning with Human Feedback (RLHF): Ensures adherence to human values and safety through iterative training.

Guide: Running Locally

Example: Running the Triangulum-1B Model

Step 1: Download the Model

Download the Triangulum-1B-F16.gguf model:

ollama run triangulum-1b-f16.gguf

Step 2: Model Initialization and Download

Upon execution, Ollama will initialize and download the necessary model files.

Step 3: Interact with the Model

Send a prompt to interact with the model:

>>> What can you do for me?

You will receive a response outlining the model's capabilities, such as answering questions, generating ideas, and more.

Step 4: Exit the Program

To exit, type:

/exit

Cloud GPUs

For enhanced performance, consider using cloud GPUs like AWS, Google Cloud, or Azure to run the model.

License

Triangulum-1B is distributed under the CreativeML OpenRAIL-M license.