Nitral-AI/Hathor_Stable-v0.2-L3-8B
Hugging Face
TEXT GENERATIONConcurrency Cost:1Model Size:8BQuant:FP8Ctx Length:8kPublished:Jun 9, 2024License:otherArchitecture:Transformer0.1K Warm

Hathor_Stable-v0.2-L3-8B by Nitral-AI is an 8 billion parameter language model based on the Llama 3 8B Instruct architecture. It is fine-tuned on private data, synthetic instructions, and novel data, specifically optimized for roleplaying chat pairs. This model is designed for applications requiring nuanced conversational abilities and creative text generation.

Loading preview...

Nitral-AI/Hathor_Stable-v0.2-L3-8B Overview

Hathor_Stable-v0.2-L3-8B is an 8 billion parameter language model developed by Nitral-AI, built upon the Llama 3 8B Instruct foundation. This model has undergone extensive fine-tuning over three epochs using a unique blend of private datasets, synthetic Opus instructions, and a diverse collection of light and classical novel data. Its primary optimization focuses on enhancing performance in roleplaying chat scenarios.

Key Capabilities

  • Enhanced Roleplaying: Specifically trained with roleplaying chat pairs to deliver more engaging and contextually appropriate responses in interactive scenarios.
  • Creative Text Generation: Benefits from a mix of novel data, contributing to its ability to generate varied and imaginative text.
  • Instruction Following: Incorporates synthetic Opus instructions to improve adherence to user prompts and directives.

Performance Metrics

Evaluations on the Open LLM Leaderboard indicate an average score of 25.70. Specific results include:

  • IFEval (0-Shot): 71.75
  • BBH (3-Shot): 32.83
  • MMLU-PRO (5-shot): 29.96

Good For

  • Applications requiring advanced conversational AI, particularly for roleplaying and interactive storytelling.
  • Generating creative content and engaging in nuanced dialogue.
  • Use cases where a model with a strong foundation in instruction following and diverse text generation is beneficial.
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