feeltheAGI/Maverick-7B

Maverick-7B Overview

Maverick-7B is a 7 billion parameter language model developed by feeltheAGI, formed through a merge of two base models: mlabonne/Marcoro14-7B-slerp and mlabonne/NeuralBeagle14-7B. This merging approach aims to combine the strengths of its constituent models, resulting in a versatile general-purpose LLM.

Key Capabilities & Performance

Maverick-7B has been evaluated across a range of benchmarks, showcasing its ability in various cognitive tasks:

• Truthfulness: Achieves a TruthfulQA mc2 score of 0.6661, indicating a good capacity for generating factually correct responses.
• General Reasoning: Demonstrates solid performance on GPT4ALL tasks, with an acc_norm of 0.6570 on ARC Challenge and 0.8460 on PIQA.
• Advanced Reasoning: Scores on AGIEval tasks include 0.5216 acc_norm on LSAT Logical Reasoning and 0.8010 acc_norm on SAT English, suggesting capabilities in complex problem-solving and comprehension.
• Bigbench Tasks: Shows proficiency in areas like sports understanding (0.7424 multiple_choice_grade) and reasoning about colored objects (0.7230 multiple_choice_grade).

When to Use This Model

Maverick-7B is a strong candidate for use cases requiring a 7B parameter model with a balanced performance across general knowledge, reasoning, and truthfulness. Its benchmark results suggest it is well-suited for:

• General-purpose chatbots and assistants: Capable of handling diverse queries and generating coherent responses.
• Content generation: For tasks where factual accuracy and logical consistency are important.
• Educational applications: Assisting with comprehension and problem-solving in various subjects.
• Research and development: As a base model for further fine-tuning on specific tasks.