qwen3-webdev-0.6b

A fine-tuned version of Qwen/Qwen3-0.6B on a curated dataset of real-world web development Q&A.

Model Description

This model is fine-tuned to answer junior-to-mid-level web development questions covering HTML, CSS, JavaScript, React, APIs, and common frontend/backend concepts.

• Base model: Qwen/Qwen3-0.6B
• Fine-tuning method: Supervised Fine-Tuning (SFT) with TRL
• Dataset: 307 real web development Q&A pairs (interview-style)
• Training: 3 epochs, final loss 0.7072
• Hardware: NVIDIA RTX 4090 Mobile (16GB)

Intended Use

• Learning tool for web development concepts
• Junior dev quick-reference assistant
• Demo of efficient small-model fine-tuning pipeline

Training Details

Parameter	Value
Base model	Qwen3-0.6B
Dataset size	307 examples
Epochs	3
Final train loss	0.7072
Precision	bfloat16

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model = AutoModelForCausalLM.from_pretrained("PacificDev/qwen3-webdev-0.6b", dtype=torch.bfloat16)
tokenizer = AutoTokenizer.from_pretrained("PacificDev/qwen3-webdev-0.6b")

prompt = "What is the difference between flexbox and CSS grid?"
inputs = tokenizer(f"Question: {prompt}\nAnswer:", return_tensors="pt")
output = model.generate(**inputs, max_new_tokens=300, temperature=0.7, do_sample=True)
print(tokenizer.decode(output[0], skip_special_tokens=True))

Limitations

• Small model (0.6B params) — answers are concise/simplified
• Dataset is limited to 307 examples — may not cover all topics
• Outputs <think> reasoning tags (Qwen3 chain-of-thought)
• Not suitable for production use without further evaluation

License

Apache 2.0 (same as base model)