This model card describes a multi-modal AI system for context-aware code review that combines contrastive learning, fine-tuning, and semantic indexing to understand repository-specific patterns and provide code review assistance.
The Repository Learning Models consist of three specialized components that work together to provide context-aware code review assistance:
1# Install dependencies
2pip install transformers sentence-transformers faiss-cpu torch
3
4# Load the fine-tuned review model
5from transformers import AutoTokenizer, AutoModelForCausalLM
6import torch
7
8tokenizer = AutoTokenizer.from_pretrained("kotlarmilos/repository-learning-models")
9model = AutoModelForCausalLM.from_pretrained(
10 "kotlarmilos/repository-learning-models",
11 torch_dtype=torch.bfloat16,
12 device_map="auto"
13)
14
15# Generate a code review
16prompt = """Code diff:
17diff
18+def calculate_average(numbers):
19+ return sum(numbers) / len(numbers)
20
21
22Please write a code review comment:"""
23
24inputs = tokenizer(prompt, return_tensors="pt")
25outputs = model.generate(**inputs, max_new_tokens=150, temperature=0.7)
26review = tokenizer.decode(outputs[0], skip_special_tokens=True)
27print(review)
Training data consists of curated datasets from 15 high-quality open-source repositories:
