Hugging Face Overview
Tess-2.0-Llama-3-70B-v0.2 Overview
Tess-2.0-Llama-3-70B-v0.2 is a 70 billion parameter general-purpose large language model developed by migtissera, building upon the Meta-Llama-3-70B base. This model is the second iteration, with v0.2 specifically featuring an additional uncensoring step to enhance its instruction-following capabilities. The training methodology adheres to LIMA (Less-Is-More) principles, utilizing a curated dataset of approximately 100,000 high-quality code and general training samples.
Key Capabilities
- General Purpose: Designed for a wide range of natural language understanding and generation tasks.
- Highly Uncensored: The model is trained on a highly uncensored dataset, aiming to follow instructions almost always without refusal.
- Llama-3 Prompt Format: Utilizes the standard Llama-3 prompt format for optimal interaction.
- Efficient Fine-tuning: Fine-tuned for only two epochs with a low learning rate to preserve the base model's entropy.
Good For
- Applications requiring a highly instruction-following model.
- General text generation and conversational AI where uncensored responses are desired.
- Developers looking for a Llama-3 based model with enhanced instruction adherence.