ChaoticNeutrals/Pasta-Lake-7b
ChaoticNeutrals/Pasta-Lake-7b is a 7 billion parameter language model created by ChaoticNeutrals, formed by merging Test157t/Pasta-PrimaMaid-7b and macadeliccc/WestLake-7B-v2-laser-truthy-dpo using a slerp merge method. It features a 4096-token context length and achieves an average score of 73.07 on the Open LLM Leaderboard, demonstrating solid performance across various reasoning and language understanding benchmarks. This model is suitable for general-purpose text generation and understanding tasks where a balanced performance in its size class is desired.
Loading preview...
Pasta-Lake-7b: A Merged 7B Language Model
Pasta-Lake-7b is a 7 billion parameter language model developed by ChaoticNeutrals, created through a strategic merge of two existing models: Test157t/Pasta-PrimaMaid-7b and macadeliccc/WestLake-7B-v2-laser-truthy-dpo. This model leverages a slerp (spherical linear interpolation) merge method, combining the strengths of its constituent models to offer a balanced performance profile.
Key Capabilities
- General-Purpose Language Understanding: Achieves an average score of 73.07 on the Open LLM Leaderboard, indicating strong capabilities across a range of tasks.
- Reasoning: Scores 70.82 on the AI2 Reasoning Challenge (25-Shot) and 64.37 on GSM8k (5-shot), demonstrating proficiency in logical and mathematical reasoning.
- Common Sense & World Knowledge: Performs well on HellaSwag (10-Shot) with 87.91 and Winogrande (5-shot) with 82.64.
- Factuality: Achieves 68.28 on TruthfulQA (0-shot), suggesting a reasonable ability to generate factual responses.
- Instruction Following: While not explicitly detailed, its performance on various benchmarks implies good instruction-following capabilities for diverse prompts.
Good for
- Text Generation: Suitable for creative writing, content generation, and conversational AI applications.
- Reasoning Tasks: Can be applied to problems requiring logical deduction and mathematical understanding.
- General NLP Applications: A versatile choice for tasks like summarization, question answering, and text completion where a 7B model with balanced performance is required.
- Resource-Efficient Deployment: Available in various quantized formats (exl2, GGUF) thanks to community contributions, making it suitable for deployment on a wider range of hardware.