deneme0001/Qwen2.5-Coder-Deep-Instruct-LoRA

Qwen2.5-Coder-1.5B - Deep Instruction LoRA

This model is a LoRA fine-tuned version of Qwen2.5-Coder-1.5B-Instruct. It was trained on the CodeGen-Deep-5K dataset to enhance code reasoning capabilities using Chain-of-Thought (CoT) traces.

Model Details

• Base Model: Qwen/Qwen2.5-Coder-1.5B-Instruct
• Training Dataset: CodeGen-Deep-5K (Reasoning-focused)
• Method: LoRA (Rank: 32, Alpha: 64)
• Epochs: 3

Performance (Pass@1 on LiveCodeBench - AtCoder Easy)

• Base Model: 26.83%
• This Model: 34.15% (+7.32% Improvement) 🚀

This model specializes in algorithmic problems requiring multi-step reasoning and state tracking.

Usage

from peft import PeftModel
from transformers import AutoModelForCausalLM

base_model_id = "Qwen/Qwen2.5-Coder-1.5B-Instruct"
adapter_model_id = "deneme0001/Qwen2.5-Coder-Deep-Instruct-LoRA"

model = AutoModelForCausalLM.from_pretrained(base_model_id, device_map="auto")
model = PeftModel.from_pretrained(model, adapter_model_id)