amkyawdev/mm-coder-agent-v1-combined

🤖 MM Coder Agent v1

A professional AI coding assistant model fine-tuned from Qwen2.5-1.5B-Instruct for software development tasks.

Model Overview

Property	Value
Base Model	Qwen/Qwen2.5-1.5B-Instruct
Architecture	LoRA (PEFT Adapter)
Parameters	1.5B (base) + 37MB (adapter)
Task	Code Generation / Software Development
Framework	Transformers, Safetensors

Model Description

MM Coder Agent v1 is a specialized coding assistant built on Qwen2.5-1.5B-Instruct. This model is optimized for:

• Code Generation - Generate clean, efficient code in multiple languages
• Bug Detection - Identify and fix common programming errors
• Algorithm Implementation - Implement sorting, searching, and data structures
• Code Review - Assist with code review and best practices

Architecture Details

{
  "peft_type": "LORA",
  "base_model_name_or_path": "Qwen/Qwen2.5-1.5B-Instruct",
  "r": 16,
  "lora_alpha": 32,
  "lora_dropout": 0.0,
  "task_type": "CAUSAL_LM",
  "target_modules": ["q_proj", "k_proj", "v_proj", "o_proj", "gate_proj", "up_proj", "down_proj"]
}

Live Demo

Try the model live at: mm-coder-v1-space

Quick Start

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch
from peft import PeftModel, PeftConfig

# Load adapter config
peft_config = PeftConfig.from_pretrained("amkyawdev/mm-coder-agent-v1-combined")

# Load base model
base_model = AutoModelForCausalLM.from_pretrained(
    peft_config.base_model_name_or_path,
    torch_dtype=torch.float16,
    device_map="auto",
    trust_remote_code=True
).eval()

# Load tokenizer
tokenizer = AutoTokenizer.from_pretrained(
    peft_config.base_model_name_or_path,
    trust_remote_code=True
)

# Load LoRA adapter
model = PeftModel.from_pretrained(base_model, "amkyawdev/mm-coder-agent-v1-combined")

# Generate code
prompt = "Write a Python function to calculate fibonacci numbers"
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)

with torch.no_grad():
    outputs = model.generate(**inputs, max_new_tokens=512, temperature=0.7)

response = tokenizer.decode(outputs[0], skip_special_tokens=True)
print(response)

Example Outputs

Prompt	Output
python hello world	print("Hello, World!")
reverse string python	s[::-1]
fibonacci function python	Full fibonacci implementation
bubble sort python	Bubble sort algorithm

Training Data

• Dataset: mm-llm-coder-dataset (4M rows)
• Additional: mm-llm-coder-agent-dataset (4M rows)
• Source: Quality coding prompts and responses

Use Cases

Ideal For

• Code completion and generation
• Bug detection and fixing
• Algorithm implementation
• Learning programming concepts
• Quick prototyping

Not Recommended For

• Production-critical systems without evaluation
• Security-sensitive applications without guardrails
• Tasks beyond software development

Limitations

• 1.5B parameter model (smaller than GPT-4 class)
• May produce incorrect code - always verify outputs
• Limited context window
• Fine-tuned primarily for English

License

Apache 2.0

Citation

@model{amkyawdev/mm-coder-agent-v1-combined,
  title={MM Coder Agent v1},
  author={amkyawdev},
  year={2024},
  url={https://huggingface.co/amkyawdev/mm-coder-agent-v1-combined}
}

---

Built with ❤️ using Transformers and PEFT