Spaces:
Sleeping
Sleeping
Create src/agent.py
Browse files- src/agent.py +36 -0
src/agent.py
ADDED
@@ -0,0 +1,36 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import torch
|
2 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer
|
3 |
+
|
4 |
+
class CodingAgent:
|
5 |
+
def __init__(self, model_path):
|
6 |
+
self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
7 |
+
self.model = AutoModelForCausalLM.from_pretrained(model_path).to(self.device)
|
8 |
+
self.tokenizer = AutoTokenizer.from_pretrained(model_path)
|
9 |
+
|
10 |
+
def generate_code(self, prompt, max_length=512, temperature=0.7, top_k=50, top_p=0.95):
|
11 |
+
inputs = self.tokenizer(prompt, return_tensors="pt").to(self.device)
|
12 |
+
|
13 |
+
with torch.no_grad():
|
14 |
+
outputs = self.model.generate(
|
15 |
+
**inputs,
|
16 |
+
max_length=max_length,
|
17 |
+
temperature=temperature,
|
18 |
+
top_k=top_k,
|
19 |
+
top_p=top_p,
|
20 |
+
do_sample=True,
|
21 |
+
num_return_sequences=1,
|
22 |
+
)
|
23 |
+
|
24 |
+
return self.tokenizer.decode(outputs[0], skip_special_tokens=True)
|
25 |
+
|
26 |
+
def answer_coding_question(self, question):
|
27 |
+
prompt = f"As a coding assistant, please answer the following question:\n\nQuestion: {question}\n\nAnswer:"
|
28 |
+
return self.generate_code(prompt)
|
29 |
+
|
30 |
+
def explain_code(self, code):
|
31 |
+
prompt = f"Please explain the following code:\n\n```python\n{code}\n```\n\nExplanation:"
|
32 |
+
return self.generate_code(prompt)
|
33 |
+
|
34 |
+
def suggest_improvements(self, code):
|
35 |
+
prompt = f"Please suggest improvements for the following code:\n\n```python\n{code}\n```\n\nSuggestions:"
|
36 |
+
return self.generate_code(prompt)
|