Blur-7b-v1.2: A Merged 7B Language Model

Blur-7b-v1.2 is a 7 billion parameter language model developed by liminerity, created through a strategic merge of two distinct models: liminerity/Blured-Ties-7B and freecs/ThetaWave-7B. This merge was executed using the TIES (Task-agnostic Instruction-following Enhanced Slerp) method, with mlabonne/NeuralBeagle14-7B serving as its base architecture.

Key Capabilities & Performance

This model demonstrates solid performance across various benchmarks, achieving an overall average score of 67.74 on the Open LLM Leaderboard. Notable results include:

• AI2 Reasoning Challenge (25-Shot): 65.36
• HellaSwag (10-Shot): 83.88
• MMLU (5-Shot): 63.45
• TruthfulQA (0-shot): 60.30
• Winogrande (5-shot): 80.58
• GSM8k (5-shot): 52.84

These scores indicate its proficiency in general reasoning, common sense understanding, and multi-task language comprehension.

Ideal Use Cases

Blur-7b-v1.2 is well-suited for applications requiring:

• General-purpose text generation: Creating coherent and contextually relevant text.
• Reasoning tasks: Handling questions that require logical inference and problem-solving.
• Instruction following: Responding accurately to diverse prompts and instructions.
• Benchmarking and experimentation: A robust base for further fine-tuning or research in merged model architectures.