Hugging Face L3.3-70B-Lycosa-v0.2 by divinetaco is a 70 billion parameter merged language model, built using the 'sce' merge method with DeepSeek-R1-Distill-Llama-70B as its base. This model is specifically engineered to enhance intelligence, reduce positive bias, and foster creativity, making it suitable for applications requiring nuanced and imaginative responses. It integrates several Llama-3.3 based models, focusing on improved reasoning capabilities.
L3.3-70B-Lycosa-v0.2: A Merged Model for Enhanced Intelligence and Creativity
L3.3-70B-Lycosa-v0.2 is a 70 billion parameter language model developed by divinetaco, created through an 'sce' merge using mergekit. This iteration, a refinement of v0.1, specifically dropped llama-3.3-70b-instruct to further reduce positive bias and incorporated DeepSeek-R1-Distill-Llama-70B as a target model to significantly improve reasoning capabilities.
Key Characteristics
- Enhanced Intelligence: The merge prioritizes overall model intelligence, aiming for more sophisticated and coherent outputs.
- Reduced Positive Bias: Through strategic model selection and merging, v0.2 seeks to mitigate inherent positive biases often found in language models.
- Increased Creativity: Designed to excel in tasks requiring imaginative and novel responses.
- DeepSeek-R1 Influence: Utilizes
deepseek-ai/DeepSeek-R1-Distill-Llama-70Bas its base and a significant target model, influencing its reasoning and overall performance.
Recommended Usage
This model is particularly well-suited for use cases demanding high intelligence, creative text generation, and a more neutral, less biased output. The recommended chat template is the DeepSeek-R1-Distill-Llama-70B format, as the increased DeepSeek-R1 influence makes the Llama3 chat template less optimal for this version.