maldv/badger-l3-instruct-32k
maldv/badger-l3-instruct-32k is an 8 billion parameter instruction-tuned language model based on the Llama 3 architecture. It is a recursive maximally disjoint pairwise normalized fourier interpolation of 26 different Llama 3-based models, designed for broad instruction-following capabilities. The model features an adjusted configuration for rope scale 4, supporting coherent context lengths up to 32k tokens. It achieves an average score of 69.49 on the Open LLM Leaderboard, demonstrating strong performance across various reasoning and language understanding tasks.
Loading preview...
Overview
maldv/badger-l3-instruct-32k is an 8 billion parameter instruction-following language model built upon the Llama 3 architecture. This model is notable for its unique construction method: a recursive maximally disjoint pairwise normalized fourier interpolation of 26 distinct Llama 3-based models. This merging technique aims to combine the strengths of multiple specialized models into a single, versatile instruction-tuned variant.
Key Capabilities & Features
- Broad Instruction Following: Designed to handle a wide range of prompts and tasks due to its diverse merge origins.
- Extended Context Window: Configured with rope scale 4, enabling coherent processing of context lengths up to 32k tokens.
- Strong General Performance: Achieves an average score of 69.49 on the Open LLM Leaderboard, with specific scores including 63.65 on AI2 Reasoning Challenge, 81.40 on HellaSwag, and 67.13 on MMLU.
Use Cases
This model is suitable for applications requiring a robust 8B parameter instruction model with an extended context window. Its strong performance across various benchmarks suggests utility in:
- General-purpose conversational AI.
- Reasoning and problem-solving tasks.
- Applications benefiting from longer input sequences.
Top 3 parameter combinations used by Featherless users for this model. Click a tab to see each config.