Hathor_Sofit-L3-8B-v1, developed by Nitral-AI, is a language model built upon the Llama 3 architecture. This model is trained on a diverse dataset including STEM instructions, creative writing, and roleplaying chat pairs. It is designed to integrate creativity, intelligence, and robust performance, making it suitable for tasks requiring nuanced understanding and generation.
Loading preview...
Model Overview
Nitral-AI's Hathor_Sofit-L3-8B-v1 is a language model based on the Llama 3 architecture, representing the final iteration in the Hathor series. This version is specifically trained on a single epoch of a comprehensive dataset, which includes private STEM instruction/dialogs, the swallow-magpie-ultra-v0.1 dataset, Hathor_0.5 instructions, and a mixture of creative writing data encompassing light, web, and classical novels, alongside roleplaying chat pairs over Hathor_0.85.
Key Capabilities
- Integrated Qualities: Designed to combine creativity, intelligence, and robust performance.
- Diverse Training: Benefits from a broad training regimen covering STEM, creative writing, and conversational roleplay.
- Llama 3 Foundation: Built upon the Llama 3 architecture, leveraging its underlying capabilities.
Good For
- Applications requiring a blend of creative text generation and intelligent responses.
- Use cases involving STEM-related instruction following and dialogue.
- Roleplaying and conversational AI scenarios due to its specific training on chat pairs.
Community Contributions
- GGUF Quantizations: Available thanks to mradermacher. GGUF Here
- EXL2 Quantizations: Provided by RiveRiPH, including 8bpw and 6.3bpw versions. Exl2 8bpw Here Exl2 6.3bpw Here
Recommended ST Presets are also available for optimal performance. Hathor Presets(Updated)
Top 3 parameter combinations used by Featherless users for this model. Click a tab to see each config.