Qwen2.5-Coder-7B-FastAPI-LoRA

A LoRA fine-tune of Qwen2.5-Coder-7B-Instruct specialized as a FastAPI documentation assistant. The model is trained to answer questions, generate code, and explain concepts related to the FastAPI framework, covering everything from basic routing to advanced topics like security and testing.

Model Details

• Developed by: LadiesMan69
• License: apache-2.0
• Base model: unsloth/Qwen2.5-Coder-7B-Instruct-bnb-4bit
• Model type: Causal decoder-only language model (Qwen2 architecture)
• Fine-tuning method: LoRA (Low-Rank Adaptation)
• Language: English
• Trained with: Unsloth + Hugging Face TRL — 2x faster training

Motivation

General-purpose code models are often imprecise or outdated when it comes to framework-specific APIs. This model was fine-tuned on a curated dataset of FastAPI-focused instruction/response pairs to produce a lightweight, deployable assistant that gives accurate, idiomatic answers for building and debugging FastAPI applications.

Training Data

The fine-tuning dataset was built specifically for this task using the ChatML format and organized into topic categories, including:

• Tutorial — core concepts: path/query parameters, request bodies, response models, dependency injection
• Advanced — background tasks, middleware, WebSockets, custom exception handlers, lifespan events
• Security — OAuth2/JWT authentication, password hashing, CORS, rate limiting
• Testing — TestClient usage, pytest fixtures, mocking dependencies, async test patterns

Examples were generated in batches per category to ensure balanced topic coverage and consistent formatting across the dataset.

Intended Use

• Answering questions about FastAPI concepts, patterns, and best practices
• Generating FastAPI route handlers, Pydantic models, and dependency-injected services
• Explaining and debugging FastAPI-related code snippets
• Acting as an in-editor or chat-based documentation assistant for developers working with FastAPI

How to Use

With transformers

from transformers import AutoTokenizer, AutoModelForCausalLM

model_id = "LadiesMan69/Qwen2.5-Coder-7B-FastAPI-LoRA"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id, device_map="auto")

messages = [
    {"role": "system", "content": "You are a helpful FastAPI documentation assistant."},
    {"role": "user", "content": "How do I add JWT-based authentication to a FastAPI route?"},
]

inputs = tokenizer.apply_chat_template(
    messages,
    add_generation_prompt=True,
    tokenize=True,
    return_dict=True,
    return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=512)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:], skip_special_tokens=True))

With unsloth

from unsloth import FastModel

model, tokenizer = FastModel.from_pretrained(
    model_name="LadiesMan69/Qwen2.5-Coder-7B-FastAPI-LoRA",
    max_seq_length=2048,
)

With vLLM

pip install vllm
vllm serve "LadiesMan69/Qwen2.5-Coder-7B-FastAPI-LoRA"

curl -X POST "http://localhost:8000/v1/chat/completions" \
  -H "Content-Type: application/json" \
  --data '{
    "model": "LadiesMan69/Qwen2.5-Coder-7B-FastAPI-LoRA",
    "messages": [
      {"role": "user", "content": "Show me a minimal FastAPI app with a health check endpoint."}
    ]
  }'

Prompt Format

This model uses the ChatML-style chat template built into the tokenizer (apply_chat_template). For best results, include a system message establishing the assistant's role as a FastAPI expert, followed by the user's question.

Limitations

• Focused specifically on FastAPI; general coding ability outside this domain is inherited from the base model and not specifically enhanced.
• As with any LLM, generated code should be reviewed and tested before use in production.
• May not reflect the very latest FastAPI releases if they postdate the training data.

Training Procedure

Fine-tuned using LoRA adapters on top of the 4-bit quantized unsloth/Qwen2.5-Coder-7B-Instruct-bnb-4bit base model, leveraging Unsloth's optimized training kernels for faster, memory-efficient fine-tuning.

Model Tree

• Base: Qwen/Qwen2.5-7B
• → Qwen/Qwen2.5-Coder-7B
• → Qwen/Qwen2.5-Coder-7B-Instruct
• → unsloth/Qwen2.5-Coder-7B-Instruct-bnb-4bit (quantized)
• → LadiesMan69/Qwen2.5-Coder-7B-FastAPI-LoRA (this model, LoRA fine-tune)

Acknowledgements

This qwen2 model was trained 2x faster with Unsloth and Hugging Face's TRL library.