Spaces:
Sleeping
Sleeping
added app.py
Browse files
app.py
ADDED
|
@@ -0,0 +1,127 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import torch
|
| 2 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer
|
| 3 |
+
import gradio as gr
|
| 4 |
+
|
| 5 |
+
# Load model and tokenizer
|
| 6 |
+
model = AutoModelForCausalLM.from_pretrained(
|
| 7 |
+
"Neo111x/Falcon3-3B-Instruct-RL-CODE-FIX",
|
| 8 |
+
trust_remote_code=True
|
| 9 |
+
)
|
| 10 |
+
tokenizer = AutoTokenizer.from_pretrained(
|
| 11 |
+
"Neo111x/Falcon3-3B-Instruct-RL-CODE-FIX",
|
| 12 |
+
trust_remote_code=True
|
| 13 |
+
)
|
| 14 |
+
model.eval()
|
| 15 |
+
|
| 16 |
+
# Inference function
|
| 17 |
+
def repair_code(faulty_code):
|
| 18 |
+
PROGRAM_REPAIR_TEMPLATE = f"""
|
| 19 |
+
You are an expert in the field of software testing.
|
| 20 |
+
You are given a buggy Python program, you are supposed to first generate testcases that can expose the bug,
|
| 21 |
+
and then generate the corresponding fixed code. The two tasks are detailed as follows.
|
| 22 |
+
|
| 23 |
+
1. **Generate a comprehensive set of test cases to expose the bug**:
|
| 24 |
+
- Each test case should include an input and the expected output.
|
| 25 |
+
- Output the test cases as a JSON list, where each entry is a dictionary with keys "test_input" and "test_output".
|
| 26 |
+
- Write in
|
| 27 |
+
|
| 28 |
+
json
|
| 29 |
+
|
| 30 |
+
block.
|
| 31 |
+
|
| 32 |
+
2. **Provide a fixed version**:
|
| 33 |
+
- Write a correct Python program to fix the bug.
|
| 34 |
+
- Write in
|
| 35 |
+
|
| 36 |
+
python
|
| 37 |
+
|
| 38 |
+
block.
|
| 39 |
+
- The code should read from standard input and write to standard output, matching the input/output format specified in the problem.
|
| 40 |
+
|
| 41 |
+
Here is an example.
|
| 42 |
+
The faulty Python program is:
|
| 43 |
+
|
| 44 |
+
python
|
| 45 |
+
\"\"\"Please write a Python program to sum two integer inputs\"\"\"
|
| 46 |
+
def add (x, y):
|
| 47 |
+
return x - y
|
| 48 |
+
x = int(input())
|
| 49 |
+
y = int(input())
|
| 50 |
+
print(add(x,y))
|
| 51 |
+
|
| 52 |
+
|
| 53 |
+
|
| 54 |
+
Testcases that can expose the bug:
|
| 55 |
+
|
| 56 |
+
json
|
| 57 |
+
[
|
| 58 |
+
{{
|
| 59 |
+
\"test_input\":\"1\n2\",
|
| 60 |
+
\"test_output\":\"3\"
|
| 61 |
+
}},
|
| 62 |
+
{{
|
| 63 |
+
\"test_input\":\"-1\n1\",
|
| 64 |
+
\"test_output\":\"0\"
|
| 65 |
+
}},
|
| 66 |
+
{{
|
| 67 |
+
\"test_input\":\"-1\n2\",
|
| 68 |
+
\"test_output\":\"1\"
|
| 69 |
+
}}
|
| 70 |
+
]
|
| 71 |
+
|
| 72 |
+
|
| 73 |
+
|
| 74 |
+
Fixed code:
|
| 75 |
+
|
| 76 |
+
python
|
| 77 |
+
def add (x, y):
|
| 78 |
+
return x + y
|
| 79 |
+
x = int(input())
|
| 80 |
+
y = int(input())
|
| 81 |
+
print(add(x,y))
|
| 82 |
+
|
| 83 |
+
|
| 84 |
+
|
| 85 |
+
Now, you are given a faulty Python function, please return:
|
| 86 |
+
1. **Testcases** that helps expose the bug.
|
| 87 |
+
2. **Fixed code** that can pass all testcases.
|
| 88 |
+
|
| 89 |
+
The faulty function is:
|
| 90 |
+
|
| 91 |
+
python
|
| 92 |
+
{faulty_code}
|
| 93 |
+
|
| 94 |
+
|
| 95 |
+
<|assistant|>
|
| 96 |
+
"""
|
| 97 |
+
messages = [
|
| 98 |
+
{
|
| 99 |
+
"role": "user",
|
| 100 |
+
"content": PROGRAM_REPAIR_TEMPLATE
|
| 101 |
+
}
|
| 102 |
+
]
|
| 103 |
+
|
| 104 |
+
text = tokenizer.apply_chat_template(messages, tokenize=False)
|
| 105 |
+
inputs = tokenizer([text], return_tensors="pt").to(model.device)
|
| 106 |
+
|
| 107 |
+
with torch.no_grad():
|
| 108 |
+
outputs = model.generate(
|
| 109 |
+
**inputs,
|
| 110 |
+
max_new_tokens=512,
|
| 111 |
+
do_sample=False
|
| 112 |
+
)
|
| 113 |
+
|
| 114 |
+
generated_ids = [
|
| 115 |
+
output_ids[len(input_ids):] for input_ids, output_ids in zip(inputs["input_ids"], outputs)
|
| 116 |
+
]
|
| 117 |
+
result = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
|
| 118 |
+
return result.strip()
|
| 119 |
+
|
| 120 |
+
# Gradio UI
|
| 121 |
+
gr.Interface(
|
| 122 |
+
fn=repair_code,
|
| 123 |
+
inputs=gr.Textbox(label="Faulty Python Function", lines=15, placeholder="Paste your buggy function here..."),
|
| 124 |
+
outputs=gr.Textbox(label="Generated Test Cases and Fixed Code", lines=30),
|
| 125 |
+
title="🧠 AI Program Repair - Falcon3 3B GRPO",
|
| 126 |
+
description="Paste a buggy Python function. The model will generate test cases and a fixed version of the code."
|
| 127 |
+
).launch()
|