Trelis/Llama-3.2-1B-Instruct-MATH

Hugging Face
TEXT GENERATIONPricing:Input $0.108 / Output $0.804Concurrent Unit Cost:1Model Size:1BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Oct 7, 2024License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Warm

Trelis/Llama-3.2-1B-Instruct-MATH is a 1 billion parameter instruction-tuned Llama model developed by Trelis. This model is specifically fine-tuned for mathematical tasks, building upon the Trelis/Llama-3.2-1B-Instruct-MATH-synthetic base. It leverages Unsloth and Huggingface's TRL library for accelerated training, making it suitable for efficient mathematical reasoning applications.

Loading preview...

Overview

Trelis/Llama-3.2-1B-Instruct-MATH is a 1 billion parameter Llama-based instruction-tuned model developed by Trelis. It is fine-tuned from the Trelis/Llama-3.2-1B-Instruct-MATH-synthetic model, indicating a specialized focus on mathematical instruction following. The model was trained using Unsloth and Huggingface's TRL library, which enabled a 2x faster training process.

Key Capabilities

  • Mathematical Instruction Following: Specialized for understanding and responding to mathematical prompts.
  • Efficient Training: Benefits from Unsloth's optimizations for faster fine-tuning.
  • Llama Architecture: Based on the Llama model family, providing a familiar and robust foundation.

Good For

  • Applications requiring a compact model for mathematical problem-solving.
  • Scenarios where efficient inference and deployment of a math-focused LLM are critical.
  • Developers looking for a Llama-based model with enhanced mathematical reasoning capabilities.