miniscript-code-helper-lora

This repository contains a LoRA adapter for Qwen/Qwen2.5-Coder-7B-Instruct, fine-tuned to help answer questions about the MiniScript programming language.

The adapter was trained on a small MiniScript Q&A corpus. On its own, it improves MiniScript awareness somewhat, but best results come when it is used together with a RAG pipeline over MiniScript reference materials.

Base model

• Qwen/Qwen2.5-Coder-7B-Instruct

What this repo contains

• PEFT/LoRA adapter weights only
• Not the full base model

Intended use

• Answering questions about MiniScript
• Assisting with MiniScript syntax and examples
• Best used with retrieval augmentation (RAG)

Limitations

• The adapter alone is not fully reliable
• It may still fall back to Python-flavored assumptions from the base model
• For best accuracy, pair it with a MiniScript documentation retriever

Example usage

from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer

base_model_id = "Qwen/Qwen2.5-Coder-7B-Instruct"
adapter_id = "JoeStrout/miniscript-code-helper-lora"

tokenizer = AutoTokenizer.from_pretrained(base_model_id)

base_model = AutoModelForCausalLM.from_pretrained(
    base_model_id,
    torch_dtype="auto",
    device_map="auto",
)

model = PeftModel.from_pretrained(base_model, adapter_id)
model.eval()

messages = [
    {"role": "system", "content": "You are a helpful assistant specializing in MiniScript programming."},
    {"role": "user", "content": "How do I iterate over a map in MiniScript?"},
]

text = tokenizer.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True,
)
inputs = tokenizer([text], return_tensors="pt").to(model.device)
output = model.generate(**inputs, max_new_tokens=512)
response = tokenizer.decode(
    output[0][len(inputs.input_ids[0]):],
    skip_special_tokens=True,
)

print(response)