Model Overview

uukuguy/speechless-code-mistral-7b-v2.0 is a 7 billion parameter language model built upon the mistralai/Mistral-7B-v0.1 architecture. This iteration, developed by uukuguy, focuses on significantly improving the model's reasoning, planning, and code generation abilities through a targeted fine-tuning process.

Training Data & Enhancements

The model was fine-tuned on a substantial dataset of 343,370 samples (603 MB), meticulously curated from various sources to bolster its core strengths:

• Coding & Reasoning: Datasets like jondurbin/airoboros-2.2 (filtered for coding, reasoning, planning), WizardLM/WizardLM_evol_instruct_V2_196k (coding conversations), TokenBender/python_eval_instruct_51k (Python output), OpenHermes (code blocks), and ise-uiuc/Magicoder-OSS-Instruct-75K were used.
• Logical & Mathematical Reasoning: Contributions from Open-Orca/OpenOrca (COT category), garage-bAInd/Open-Platypus, and meta-math/MetaMathQA (20% of 395K samples) were integrated to enhance its problem-solving prowess.

Performance & Benchmarks

While specific humaneval-python and lm-evaluation-harness scores are not detailed in the provided README, the model's training regimen suggests a strong focus on competitive performance in code-related tasks. For context, the README references the Big Code Models Leaderboard and the Open LLM Leaderboard, indicating an ambition to compete with established code-centric models like CodeLlama variants.

Ideal Use Cases

This model is particularly well-suited for applications requiring:

• Code Generation and Completion: Especially for Python-based tasks.
• Complex Problem Solving: Where logical reasoning and planning are crucial.
• Mathematical and Scientific Inquiry: Benefiting from its MetaMathQA training.
• Instruction Following: For tasks demanding precise and structured outputs.