Saikrishna2511/java2py-qwen

Java→Python fine-tuned Qwen2.5-Coder-0.5B-Instruct checkpoint (Stage 1 LoRA, merged for inference).

Demo

Related multi-task demo: https://huggingface.co/spaces/Saikrishna2511/qwen-multitask-demo

Task

Java → Python (java2py)

### Translate Java to Python:
```java
{java code}

Python:

Training

• Base model: Qwen/Qwen2.5-Coder-0.5B-Instruct
• Data: AVATAR-TC / Java→Python pairs
• Method: LoRA (r=16, alpha=32), merged weights for inference

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model_id = "Saikrishna2511/java2py-qwen"
tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    trust_remote_code=True,
    torch_dtype=torch.float16,
    device_map="auto",
)

java = "public class Hello { public static void main(String[] args) { System.out.println(\"hi\"); } }"
prompt = f"### Translate Java to Python:\\n```java\\n{java}\\n```\\n### Python:\\n```python\\n"
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
outputs = model.generate(**inputs, max_new_tokens=512, temperature=0.2, top_p=0.95)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

Limitations

• Small 0.5B model; translation quality varies with input complexity
• Prefer the multi-task checkpoint for NL→Python / Code2Doc: Saikrishna2511/qwen-multitask
• Not intended for production use without further evaluation