abideen/MonarchCoder-7B
MonarchCoder-7B is a 7 billion parameter language model created by abideen, formed by slerp merging Syed-Hasan-8503/Tess-Coder-7B-Mistral-v1.0 and mlabonne/AlphaMonarch-7B. This model is designed to excel in a combination of reasoning, conversation, and coding tasks, leveraging a 4096-token context length. It demonstrates improved performance on coding benchmarks like HumanEval compared to its base models, making it suitable for applications requiring balanced general intelligence and code generation capabilities.
Loading preview...
Overview
MonarchCoder-7B is a 7 billion parameter language model developed by abideen, created through a slerp merge of two distinct models: Syed-Hasan-8503/Tess-Coder-7B-Mistral-v1.0 and mlabonne/AlphaMonarch-7B. The primary goal of this merge was to combine the strengths of a coding-focused model with a model excelling in reasoning and conversational abilities.
Key Capabilities & Performance
MonarchCoder-7B aims for a balanced performance across multiple domains, particularly in:
- Reasoning: It maintains strong reasoning capabilities inherited from AlphaMonarch.
- Conversation: Designed to handle conversational tasks effectively.
- Coding: Shows improved performance in coding benchmarks compared to its AlphaMonarch base, with a HumanEval score of 39.02 and HumanEval+ of 31.70.
Benchmarks
While its larger counterpart, MonarchCoder-MoE-2x7B, generally achieves higher average scores, MonarchCoder-7B demonstrates competitive performance:
- HumanEval: 39.02
- HumanEval+: 31.70
- AI2 Reasoning Challenge: 68.52
- MMLU: 64.65
- GSM8k: 65.13
When to Use
MonarchCoder-7B is a suitable choice for developers looking for a 7B parameter model that offers a good blend of general reasoning, conversational fluency, and enhanced code generation capabilities. It's particularly useful for applications where a single model needs to handle diverse tasks without specializing exclusively in one area.