File size: 1,916 Bytes
0875567
fe4a64d
 
 
0875567
3466d4c
 
bc6423e
3466d4c
 
 
 
 
 
 
 
bc6423e
0875567
 
 
 
 
 
 
 
 
 
 
bc6423e
3466d4c
 
 
 
bc6423e
0875567
bc6423e
 
 
 
 
 
0875567
 
 
 
 
 
 
 
 
bc6423e
 
 
 
0875567
bc6423e
0875567
 
 
bc6423e
 
 
 
0875567
 
bc6423e
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
import os
os.environ["HF_HOME"] = "/data/hf_cache"  # /data 在 Spaces 裡是可持久化且可寫的
os.makedirs(os.environ["HF_HOME"], exist_ok=True)

import torch
from fastapi import FastAPI
from fastapi.middleware.cors import CORSMiddleware
from transformers import AutoTokenizer, AutoModelForSequenceClassification

app = FastAPI()
app.add_middleware(
    CORSMiddleware,
    allow_origins=["*"],
    allow_methods=["*"],
    allow_headers=["*"],
)

# ===== 模型設定 =====
MODEL_REPO = "raxhel/Malvec"  # 換成你的模型 repo
HF_TOKEN = os.environ.get("hf_token")  # 從 Space Secrets 取得;若模型是公開,None 也可

# CPU 環境(Spaces CPU 機型)
DEVICE = torch.device("cpu")

# ===== 載入模型(啟動時載一次)=====
tokenizer = AutoTokenizer.from_pretrained(MODEL_REPO, token=HF_TOKEN)
model = AutoModelForSequenceClassification.from_pretrained(MODEL_REPO, token=HF_TOKEN)
model.to(DEVICE)
model.eval()

@app.get("/")
def home():
    return {"message": "FastAPI Space is running!"}

# 簡單測試路由:確認 tokenizer + model 真的能跑
@app.post("/test")
def test_model(data: dict):
    text = data.get("text", "")
    if not text:
        return {"error": "請提供 text"}

    inputs = tokenizer(
        text,
        return_tensors="pt",
        truncation=True,
        padding=True,
        max_length=512,
    )
    # 保證在 CPU
    inputs = {k: v.to(DEVICE) for k, v in inputs.items()}

    with torch.no_grad():
        outputs = model(**inputs)

    logits = outputs.logits  # [1, num_labels]
    probs = torch.softmax(logits, dim=-1)
    pred_id = int(torch.argmax(probs, dim=-1).item())
    conf = float(probs[0, pred_id].item())
    label = model.config.id2label.get(pred_id, str(pred_id))

    return {
        "input_text": text,
        "predicted_label": label,
        "confidence": round(conf, 4),
        "logits": logits.cpu().tolist(),
    }