nbeerbower/llama3.1-cc-8B
nbeerbower/llama3.1-cc-8B is an 8 billion parameter language model, finetuned by nbeerbower from mlabonne/Meta-Llama-3.1-8B-Instruct-abliterated. It was specifically trained on the flammenai/casual-conversation-DPO dataset to enhance its conversational abilities. This model is designed for generating sequential, casual conversation following the Llama 3 template, making it suitable for dialogue-focused applications. It features a context length of 32768 tokens.
Loading preview...
Model Overview
nbeerbower/llama3.1-cc-8B is an 8 billion parameter language model, finetuned from mlabonne/Meta-Llama-3.1-8B-Instruct-abliterated. Its primary distinction lies in its specialized training on the flammenai/casual-conversation-DPO dataset, focusing on generating natural and sequential casual conversations. The model was finetuned using an A100 GPU on Google Colab over 3 epochs, employing a method detailed in a blog post on fine-tuning Llama 3 with ORPO.
Key Capabilities
- Casual Conversation Generation: Optimized for producing human-like, sequential dialogue in informal settings.
- Llama 3 Template Adherence: Formats conversation data according to the Llama 3 template, ensuring consistent output structure.
- Experimental Finetune: Represents an experimental approach to enhancing conversational flow through specific dataset training.
Performance Metrics
Evaluated on the Open LLM Leaderboard, the model achieved an average score of 20.13. Specific benchmark results include:
- IFEval (0-Shot): 50.68
- BBH (3-Shot): 26.48
- MATH Lvl 5 (4-Shot): 6.34
- GPQA (0-shot): 4.70
- MuSR (0-shot): 6.50
- MMLU-PRO (5-shot): 26.08
Good For
- Dialogue Systems: Ideal for chatbots or applications requiring natural, flowing casual conversations.
- Interactive AI: Suitable for scenarios where the model needs to maintain context and coherence in sequential exchanges.
- Research in Conversational AI: Provides a base for further experimentation in dialogue generation and finetuning techniques.