Overview

ColorShadow-7B is a 7 billion parameter language model developed by nlpguy. It is a product of a Gradient-SLERP merge, combining the capabilities of diffnamehard/Mistral-CatMacaroni-slerp-7B and cookinai/Valkyrie-V1 using the mergekit tool. This merging technique, specifically using a slerp method with varying t parameters for self-attention and MLP layers, aims to create a balanced model by blending the characteristics of its constituent models.

Key Capabilities

• General Reasoning: Achieves 67.83% on the AI2 Reasoning Challenge (25-Shot).
• Common Sense Inference: Scores 85.15% on HellaSwag (10-Shot) and 80.58% on Winogrande (5-Shot).
• Knowledge & Understanding: Demonstrates 61.69% on MMLU (5-Shot).
• Mathematical Reasoning: Performs at 55.19% on GSM8k (5-Shot).
• Factuality: Scores 59.56% on TruthfulQA (0-shot).

Performance Highlights

Evaluated on the Open LLM Leaderboard, ColorShadow-7B achieved an average score of 68.34. Detailed results are available here, showcasing its balanced performance across a range of academic benchmarks.

Good for

• Applications requiring a general-purpose language model with solid reasoning and common-sense abilities.
• Use cases where a merged model approach is preferred for combining specific strengths of different base models.