amanmoon/leetcode_finetuned_Qwen2.5-Coder-0.5B-bnb-4bit

LeetCode-Finetuned-Qwen2.5-Coder-0.5B

This model is a fine-tuned version of unsloth/Qwen2.5-Coder-0.5B-bnb-4bit, specialized for solving competitive programming problems, specifically from the LeetCode platform.

Model Details

• Model Type: Causal Language Model
• Base Model: Qwen2.5-Coder-0.5B (4-bit quantized)
• Fine-tuning Technique: Supervised Fine-Tuning (SFT)
• Dataset: newfacade/LeetCodeDataset
• Primary Focus: Algorithmic problem solving and Python/C++ code generation.

Training Procedure

The model was trained using the SFT (Supervised Fine-Tuning) method to transform the base completion model into a helpful coding assistant.

Intended Use

This model is intended for:

• Solving LeetCode-style algorithmic challenges.
• Explaining code logic for data structures and algorithms.
• Providing optimized solutions (Time/Space complexity) in Python.

Performance & Comparison

In internal evaluations, this fine-tuned version significantly outperformed the base model in:

1. Instruction Following: The base model frequently entered repetition loops, whereas this SFT version terminates generation correctly after providing the solution.
2. Algorithm Efficiency: Prefers O(n) or O(n \log n) solutions over brute-force approaches.
3. Consistency: Reduced noise and "hallucinated" characters compared to early LoRA checkpoints.