metadata
base_model: Qwen/Qwen2.5-Coder-1.5B
tags:
- lora
- sft
- code
- python
- instruction-tuning
license: apache-2.0
Track B SFT – Qwen2.5-Coder-1.5B + LoRA
Fine-tuned on ~250 synthetic coding instruction pairs generated from the verl corpus.
Results
| Metric | Baseline | Post-SFT | Δ |
|---|---|---|---|
| pass@1 | 0.565 | 0.804 | +0.239 |
| pass@3 | 0.783 | 0.848 | +0.065 |
Training
- Base model:
Qwen/Qwen2.5-Coder-1.5B - Method: LoRA (r=16, alpha=32)
- Data:
archit11/track_b_sft(~257 train examples) - Epochs: 3, LR: 2e-4, Hardware: T4 GPU
Usage
from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer
base = AutoModelForCausalLM.from_pretrained("Qwen/Qwen2.5-Coder-1.5B")
model = PeftModel.from_pretrained(base, "archit11/track_b_sft_model").merge_and_unload()
tokenizer = AutoTokenizer.from_pretrained("archit11/track_b_sft_model")