Model Details

• Base model: Qwen/Qwen2.5-Coder-7B-Instruct
• Fine-tuning method: QLoRA (4-bit)
• LoRA rank: 64
• Precision: bf16
• Training samples: 7,731
• Epochs: 3
• Max length: 1024 tokens

Training Data

• Hoglet-33/webdev-coding-dataset
• sahil2801/CodeAlpaca-20k (web-filtered)
• HuggingFaceH4/CodeAlpaca_20K (web-filtered)
• HuggingFaceM4/WebSight (local cache)

Usage

python
Copy code
from transformers import AutoTokenizer, AutoModelForCausalLM
from peft import PeftModel
import torch

base = AutoModelForCausalLM.from_pretrained(
    "Qwen/Qwen2.5-Coder-7B-Instruct",
    torch_dtype=torch.bfloat16,
    device_map="auto"
)
tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen2.5-Coder-7B-Instruct")
model = PeftModel.from_pretrained(base, "lhordking/webcoder-7b")

prompt = "Create a responsive dark mode landing page for a SaaS product"
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
output = model.generate(**inputs, max_new_tokens=1024)
print(tokenizer.decode(output[0], skip_special_tokens=True))

Example Prompts

• "Create a responsive navbar with dark mode toggle"
• "Build a SaaS landing page with hero section and pricing table"
• "Make a login form with email and password validation"
• "Create a portfolio page with project cards and animations"

Limitations

• Best results with HTML/CSS/JS prompts
• Output quality improves with specific, detailed prompts
• May need more training data for complex full-stack applications