nbeerbower/mistral-nemo-narwhal-12B
nbeerbower/mistral-nemo-narwhal-12B is a 12-billion parameter language model, fine-tuned by nbeerbower, based on the Mahou-1.5-mistral-nemo-12B-lorablated model. It was further optimized using ORPO (Odds Ratio Preference Optimization) on the reddit-dpo dataset, leveraging QLoRA for efficient training. This model is designed for general language generation tasks, benefiting from its 32768 token context length and specialized fine-tuning on Reddit data.
Loading preview...
nbeerbower/mistral-nemo-narwhal-12B Overview
nbeerbower/mistral-nemo-narwhal-12B is a 12-billion parameter language model developed by nbeerbower. It is built upon the Mahou-1.5-mistral-nemo-12B-lorablated base model and has been further fine-tuned using the ORPO (Odds Ratio Preference Optimization) method. The training utilized the reddit-dpo dataset, suggesting a specialization in conversational or Reddit-style text generation.
Key Capabilities
- ORPO Fine-tuning: Leverages ORPO for improved alignment and performance, potentially excelling in tasks requiring nuanced preference modeling.
- QLoRA Efficiency: Employs QLoRA (Quantized LoRA) for efficient fine-tuning, making it accessible for deployment and further adaptation.
- Extended Context Window: Features a substantial 32768 token context length, allowing it to process and generate longer, more coherent texts.
- Reddit Data Training: Fine-tuned on a Reddit dataset, which may enhance its ability to understand and generate informal, community-driven, or discussion-oriented content.
Good for
- Conversational AI: Its training on Reddit data makes it suitable for chatbots, forum moderation, or generating human-like dialogue.
- Content Generation: Effective for creating text that mimics online discussions, reviews, or social media posts.
- Research and Experimentation: Provides a robust base for further fine-tuning or experimentation with ORPO and QLoRA techniques on a 12B parameter model.