macadeliccc/OmniCorso-7B

OmniCorso-7B Overview

OmniCorso-7B is a 7 billion parameter language model developed by macadeliccc. It was created through a Slerp merge of two base models: macadeliccc/MBX-7B-v3-DPO and mlabonne/OmniBeagle-7B. This merging strategy aims to combine the strengths of its constituent models.

Key Capabilities & Performance

• General Reasoning: Achieves an average score of 61.73% across a custom evaluation suite including AGIEval (45.89%), GPT4All (77.66%), TruthfulQA (74.12%), and Bigbench (49.24%).
• Open LLM Leaderboard: Demonstrates competitive performance with an average score of 75.74% on the Hugging Face Open LLM Leaderboard, including:
  • AI2 Reasoning Challenge (25-Shot): 72.70%
  • HellaSwag (10-Shot): 88.70%
  • MMLU (5-Shot): 64.91%
  • TruthfulQA (0-shot): 73.43%
  • Winogrande (5-shot): 83.74%
  • GSM8k (5-shot): 70.96%
• Context Length: Supports a context length of 4096 tokens.
• Quantization Support: Available in various quantization formats, including GGUF and Exllamav2 (e.g., 6.5-bit recommended for good size/performance trade-off).

When to Use OmniCorso-7B

OmniCorso-7B is suitable for applications requiring a capable 7B parameter model with strong general reasoning and conversational abilities. Its balanced performance across diverse benchmarks makes it a versatile choice for tasks such as:

• General-purpose chatbots and assistants
• Question answering and information retrieval
• Code generation and understanding (as suggested by the example, though not explicitly benchmarked for code)
• Reasoning-intensive tasks where its AGIEval and Bigbench scores are relevant.