README.md

metadata

language:
  - en
license: apache-2.0
base_model: unsloth/Qwen2.5-Coder-7B-Instruct-bnb-4bit
tags:
  - text-generation-inference
  - transformers
  - unsloth
  - qwen2
  - trl
  - nodejs
  - javascript
  - backend
  - express
  - gguf
  - ollama
pipeline_tag: text-generation

🚀 nodejs-coder-qwen25

A fine-tuned Qwen2.5-Coder-7B-Instruct model specialized for Node.js backend development, trained with LoRA adapters using Unsloth, merged into a single GGUF file for efficient local inference with Ollama.

---

🧠 Model Description

Property	Details
Base Model	Qwen2.5-Coder-7B-Instruct
Fine-tuning Method	LoRA (Low-Rank Adaptation) via Unsloth
Training Framework	TRL SFTTrainer
Quantization	GGUF Q4_K_M (~4.4 GB)
Context Length	2048 tokens
Language	JavaScript / Node.js

This model is specifically trained to write clean, production-ready Node.js backend code. It understands common backend patterns including REST APIs, database integrations, authentication, and testing.

---

🎯 Specialties

• ✅ Express.js — REST APIs, middleware, routing
• ✅ NestJS — modules, controllers, services, guards
• ✅ Sequelize / Prisma — ORM models, migrations, queries
• ✅ MongoDB / Mongoose — schemas, models, aggregations
• ✅ PostgreSQL / pg — raw queries, connection pooling
• ✅ JWT Authentication — login, token generation, guards
• ✅ Jest — unit tests, mocking, integration tests
• ✅ Async/Await — file I/O, error handling, promises

---

⚡ Quick Start with Ollama

Step 1 — Download files

hf download Sriram-214/nodejs-coder-qwen25 nodejs-coder-Q4_K_M.gguf --local-dir ./
hf download Sriram-214/nodejs-coder-qwen25 Modelfile --local-dir ./

Step 2 — Create Ollama model

ollama create nodejs-coder -f Modelfile

Step 3 — Run

ollama run nodejs-coder

---

📋 Modelfile

FROM ./nodejs-coder-Q4_K_M.gguf

PARAMETER temperature 0.7
PARAMETER top_p 0.9
PARAMETER repeat_penalty 1.1
PARAMETER num_ctx 2048
PARAMETER num_gpu 0
PARAMETER stop "<|im_end|>"
PARAMETER stop "<|im_start|>"

TEMPLATE """<|im_start|>system
{{ .System }}<|im_end|>
<|im_start|>user
{{ .Prompt }}<|im_end|>
<|im_start|>assistant
"""

SYSTEM """You are a Node.js backend development expert. You write clean, efficient, production-ready code following best practices. You specialize in Express.js, NestJS, Sequelize, Prisma, MongoDB, PostgreSQL, and Jest testing. Always include error handling and follow RESTful API design principles."""

Note: Set num_gpu 0 for CPU-only mode on GPUs with ≤6GB VRAM. For GPUs with more VRAM, increase this value.

---

💬 Example Prompts & Outputs

Express REST API

Prompt: Write a simple Express.js REST API with a GET /health endpoint

const express = require('express');
const app = express();

app.get('/health', (req, res) => {
  res.status(200).json({ status: 'ok' });
});

app.listen(3000, () => console.log('Server running on port 3000'));

Async File Reading

Prompt: Write a Node.js function that reads a JSON file using async/await

const fs = require('fs').promises;

async function readJsonFile(filePath) {
  try {
    const data = await fs.readFile(filePath, 'utf-8');
    return JSON.parse(data);
  } catch (error) {
    throw new Error(`Failed to read file: ${error.message}`);
  }
}

List Directory Files

Prompt: Write a simple Node.js code to list all the files in a directory

const fs = require('fs');

fs.readdir('./', (err, files) => {
  if (err) throw err;
  console.log(files);
});

---

🏋️ Training Details

Parameter	Value
Base Model	unsloth/Qwen2.5-Coder-7B-Instruct-bnb-4bit
LoRA Rank	16
LoRA Alpha	32
Training Data	Node.js backend code dataset
Training Framework	Unsloth + TRL SFTTrainer
Training Environment	Google Colab (T4 GPU)
Quantization	Q4_K_M via llama.cpp

---

⚠️ Limitations

• Optimized for Node.js/JavaScript — not suited for other languages
• Context window of 2048 tokens — long files may be truncated
• CPU inference is slow (~3-5 tokens/sec on modern CPUs)
• May occasionally produce outdated library syntax

---

📄 License

Apache 2.0 — see LICENSE