Pythononline_evals.py
Copy
1
import json
2
import statistics
3
from collections import deque
4
from fastapi import FastAPI, BackgroundTasks
5
from openai import OpenAI
6
 
7
app = FastAPI()
8
client = OpenAI()
9
scores: deque[float] = deque(maxlen=1000)
10
 
11
def evaluate_in_background(question: str, answer: str):
12
    try:
13
        prompt = f"""Score correctness 1-5. Return JSON: {{"score": <1-5>}}
14
Q: {question}
15
A: {answer}"""
16
        resp = client.chat.completions.create(
17
            model="gpt-4o-mini",
18
            messages=[{"role": "user", "content": prompt}],
19
            response_format={"type": "json_object"},
20
        )
21
        score = json.loads(resp.choices[0].message.content or "{}").get("score", 3)
22
        scores.append(float(score))
23
    except Exception:
24
        pass  # Don't fail the request if eval errors
25
 
26
@app.post("/chat")
27
async def chat(question: str, background_tasks: BackgroundTasks):
28
    resp = client.chat.completions.create(
29
        model="gpt-4o-mini",
30
        messages=[{"role": "user", "content": question}],
31
    )
32
    answer = resp.choices[0].message.content or ""
33
    background_tasks.add_task(evaluate_in_background, question, answer)
34
    return {"answer": answer}
35
 
36
@app.get("/eval-stats")
37
async def eval_stats():
38
    if not scores:
39
        return {"message": "No evaluations yet"}
40
    score_list = list(scores)
41
    return {
42
        "count": len(score_list),
43
        "mean": round(statistics.mean(score_list), 2),
44
        "min": min(score_list),
45
        "max": max(score_list),
46
    }
47