111306047 commited on
Commit
0875567
·
1 Parent(s): bc6423e

fix auth: use token env; add /test

Browse files
Files changed (1) hide show
  1. app.py +29 -21
app.py CHANGED
@@ -1,55 +1,63 @@
 
 
1
  from fastapi import FastAPI
2
  from fastapi.middleware.cors import CORSMiddleware
3
  from transformers import AutoTokenizer, AutoModelForSequenceClassification
4
- import torch
5
 
6
  app = FastAPI()
7
-
8
  app.add_middleware(
9
  CORSMiddleware,
10
  allow_origins=["*"],
11
  allow_methods=["*"],
12
  allow_headers=["*"],
13
  )
14
- # ===== 載入模型 =====
15
- # 換成你自己的模型 repo 名稱
16
- MODEL_REPO = "raxhel/Malvec"
17
 
18
- # 如果是私有模型,會自動從 HF_TOKEN 抓取權限
19
- tokenizer = AutoTokenizer.from_pretrained(MODEL_REPO, use_auth_token=True)
20
- model = AutoModelForSequenceClassification.from_pretrained(MODEL_REPO, use_auth_token=True)
 
 
 
 
 
 
 
 
21
  model.eval()
22
 
23
  @app.get("/")
24
  def home():
25
  return {"message": "FastAPI Space is running!"}
26
 
 
27
  @app.post("/test")
28
  def test_model(data: dict):
29
  text = data.get("text", "")
30
  if not text:
31
  return {"error": "請提供 text"}
32
 
33
- # Tokenize
34
- inputs = tokenizer(text, return_tensors="pt", truncation=True, padding=True)
 
 
 
 
 
 
 
35
 
36
- # 模型推論
37
  with torch.no_grad():
38
  outputs = model(**inputs)
39
 
40
- # 取 logits
41
- logits = outputs.logits
42
  probs = torch.softmax(logits, dim=-1)
43
- pred_label_id = torch.argmax(probs, dim=-1).item()
44
- confidence = probs[0][pred_label_id].item()
45
-
46
- # 若 config 裡有 id2label
47
- label = model.config.id2label.get(pred_label_id, str(pred_label_id))
48
 
49
  return {
50
  "input_text": text,
51
  "predicted_label": label,
52
- "confidence": round(confidence, 4),
53
- "logits": logits.tolist()
54
  }
55
-
 
1
+ import os
2
+ import torch
3
  from fastapi import FastAPI
4
  from fastapi.middleware.cors import CORSMiddleware
5
  from transformers import AutoTokenizer, AutoModelForSequenceClassification
 
6
 
7
  app = FastAPI()
 
8
  app.add_middleware(
9
  CORSMiddleware,
10
  allow_origins=["*"],
11
  allow_methods=["*"],
12
  allow_headers=["*"],
13
  )
 
 
 
14
 
15
+ # ===== 模型設定 =====
16
+ MODEL_REPO = "raxhel/Malvec" # 換成你的模型 repo
17
+ HF_TOKEN = os.environ.get("hf_token") # 從 Space Secrets 取得;若模型是公開,None 也可
18
+
19
+ # CPU 環境(Spaces CPU 機型)
20
+ DEVICE = torch.device("cpu")
21
+
22
+ # ===== 載入模型(啟動時載一次)=====
23
+ tokenizer = AutoTokenizer.from_pretrained(MODEL_REPO, token=HF_TOKEN)
24
+ model = AutoModelForSequenceClassification.from_pretrained(MODEL_REPO, token=HF_TOKEN)
25
+ model.to(DEVICE)
26
  model.eval()
27
 
28
  @app.get("/")
29
  def home():
30
  return {"message": "FastAPI Space is running!"}
31
 
32
+ # 簡單測試路由:確認 tokenizer + model 真的能跑
33
  @app.post("/test")
34
  def test_model(data: dict):
35
  text = data.get("text", "")
36
  if not text:
37
  return {"error": "請提供 text"}
38
 
39
+ inputs = tokenizer(
40
+ text,
41
+ return_tensors="pt",
42
+ truncation=True,
43
+ padding=True,
44
+ max_length=512,
45
+ )
46
+ # 保證在 CPU
47
+ inputs = {k: v.to(DEVICE) for k, v in inputs.items()}
48
 
 
49
  with torch.no_grad():
50
  outputs = model(**inputs)
51
 
52
+ logits = outputs.logits # [1, num_labels]
 
53
  probs = torch.softmax(logits, dim=-1)
54
+ pred_id = int(torch.argmax(probs, dim=-1).item())
55
+ conf = float(probs[0, pred_id].item())
56
+ label = model.config.id2label.get(pred_id, str(pred_id))
 
 
57
 
58
  return {
59
  "input_text": text,
60
  "predicted_label": label,
61
+ "confidence": round(conf, 4),
62
+ "logits": logits.cpu().tolist(),
63
  }