MysteriousAI/Mia-1B: An Uncensored Text Generation Model
Mia-1B is a 1.1 billion parameter text generation model developed by MysteriousAI, designed to produce coherent and contextually relevant text without censorship. It leverages FP16 tensor types for efficient computation and is fine-tuned from the OEvortex/HelpingAI-Lite dataset, enhancing its performance and adaptability across diverse tasks.
Key Capabilities
- Uncensored Text Generation: Generates text without content limitations, suitable for a wide range of applications.
- Efficient Computation: Utilizes FP16 tensor types for optimized performance and scalability.
- Fine-tuned Performance: Benefits from fine-tuning on the OEvortex/HelpingAI-Lite dataset, improving its adaptability.
- Broad Application: Capable of generating diverse content, engaging in conversational AI, and supporting creative writing endeavors.
Performance Metrics
Evaluations on the Open LLM Leaderboard show Mia-1B achieving an average score of 36.82. Notable scores include 61.02 on HellaSwag (10-Shot) and 60.38 on Winogrande (5-shot), indicating its ability in common sense reasoning tasks. However, its MMLU (5-Shot) score is 25.43 and GSM8k (5-shot) is 1.44, suggesting areas for further development in complex reasoning and mathematical problem-solving.
Good For
- Content Generation: Creating articles, stories, and dialogues.
- Conversational AI: Developing chatbots and virtual assistants.
- Creative Writing: Exploring new ideas and narrative structures.