Severian/ANIMA-Nectar

TEXT GENERATIONConcurrency Cost:1Model Size:7BQuant:FP8Ctx Length:4kPublished:Dec 6, 2023License:mitArchitecture:Transformer Open Weights Cold

Severian/ANIMA-Nectar is a 7 billion parameter language model, fine-tuned by Severian, specializing in biomimicry, biology, and environmental science. This model is the 9th iteration, trained for an additional 20 hours on a custom dataset combining Nectar and Biomimicry/STEM data, enabling it to form novel relationships for problem-solving. It is designed to guide users through the Biomimicry Design Process and excels at generating verbose, long-form conversational discourse for multi-turn problem-solving strategies.

Loading preview...

ANIMA-Nectar-7B: Biomimicry Enhanced LLM

ANIMA (Advanced Nature Inspired Multidisciplinary Assistant) is a 7 billion parameter language model developed by Severian, specifically fine-tuned for expertise in biomimicry, biology, and environmental science. This model represents the 9th sequential fine-tuning iteration, built upon the ANIMA-Phi-Neptune-Mistral-7B base. It underwent an additional 20 hours of training on a unique custom dataset that interwove Nectar with Biomimicry/STEM data.

Key Capabilities

  • Specialized Knowledge: Expert in biomimicry, biology, and environmental science.
  • Problem-Solving: Designed to form new and novel relationships for complex problem-solving, particularly within the Biomimicry Design Process.
  • Verbose Output: Tends to generate very detailed and long-form responses, reflecting its training on multi-turn conversational discourse for design strategies.

Good For

  • Biomimicry Design: Guiding users through the Biomimicry Design Process to solve problems by drawing inspiration from nature.
  • Scientific Inquiry: Exploring concepts and generating detailed explanations in biology, environmental science, and related STEM fields.
  • Long-form Conversational Discourse: Use cases requiring extensive, multi-turn dialogue and detailed problem-solving discussions.

Known Issues

  • The model may occasionally respond to itself, taking on both user and AI roles, a known characteristic of the underlying Mistral architecture.

Performance Highlights

Evaluated on the Open LLM Leaderboard, ANIMA-Nectar-7B achieved an average score of 50.58, with notable scores including 75.99 on HellaSwag (10-Shot) and 73.72 on Winogrande (5-shot). Detailed results are available on the Open LLM Leaderboard.

License: This model is licensed under Artistic-2.0 and is restricted to research purposes only, prohibiting commercial use.