unsloth/llama-3-8b

Hugging Face
TEXT GENERATIONConcurrency Cost:1Model Size:8BQuant:FP8Ctx Length:8kPublished:Apr 18, 2024License:llama3Architecture:Transformer0.1K Warm

The unsloth/llama-3-8b model is an 8 billion parameter Llama 3-based language model developed by Unsloth, optimized for efficient fine-tuning. It offers significantly faster training speeds and reduced memory consumption compared to standard methods. This model is primarily designed for developers seeking to quickly and cost-effectively fine-tune Llama 3 for various downstream tasks, supporting a context length of 8192 tokens.

Loading preview...

Overview

The unsloth/llama-3-8b model is an 8 billion parameter language model built upon Meta's Llama 3 architecture, developed by Unsloth. Its core innovation lies in its highly optimized fine-tuning capabilities, enabling developers to train models up to 2.4 times faster while using 58% less memory compared to conventional methods.

Key Capabilities

  • Accelerated Fine-tuning: Achieves 2.4x faster fine-tuning speeds for Llama-3 8b models.
  • Memory Efficiency: Reduces memory usage by 58% during fine-tuning, making it accessible on more constrained hardware like Google Colab's Tesla T4 GPUs.
  • Beginner-Friendly Workflows: Provides easy-to-use Colab notebooks for various fine-tuning tasks, including conversational models (ShareGPT ChatML / Vicuna templates), text completion, and DPO (Direct Preference Optimization).
  • Export Options: Supports exporting fine-tuned models to GGUF, vLLM, or directly uploading to Hugging Face.
  • Broad Model Support: While this specific model is Llama-3 8b, Unsloth's framework extends similar optimizations to other models like Gemma 7b, Mistral 7b, Llama-2 7b, TinyLlama, and CodeLlama 34b.

Good For

  • Developers looking to fine-tune Llama 3 models quickly and efficiently.
  • Users with limited GPU resources (e.g., free tier Colab) who need to perform fine-tuning.
  • Experimenting with different fine-tuning tasks such as instruction following, chat, or text generation with Llama 3.

Popular Sampler Settings

Top 3 parameter combinations used by Featherless users for this model. Click a tab to see each config.

temperature
top_p
top_k
frequency_penalty
presence_penalty
repetition_penalty
min_p