Update app.py
Browse files
app.py
CHANGED
@@ -1,38 +1,51 @@
|
|
1 |
-
from fastapi import FastAPI, HTTPException
|
2 |
from onnxruntime import InferenceSession
|
3 |
from transformers import AutoTokenizer
|
4 |
import numpy as np
|
5 |
import os
|
|
|
6 |
|
7 |
app = FastAPI()
|
8 |
|
9 |
-
# Initialize tokenizer
|
10 |
tokenizer = AutoTokenizer.from_pretrained(
|
11 |
"Xenova/multi-qa-mpnet-base-dot-v1",
|
12 |
-
use_fast=True,
|
13 |
legacy=False
|
14 |
)
|
15 |
|
16 |
# Load ONNX model
|
17 |
-
|
|
|
|
|
|
|
|
|
|
|
18 |
|
19 |
@app.get("/")
|
20 |
-
def
|
21 |
-
return {"status": "
|
22 |
|
23 |
@app.post("/api/predict")
|
24 |
-
async def predict(
|
25 |
try:
|
26 |
-
#
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
27 |
inputs = tokenizer(
|
28 |
text,
|
29 |
-
return_tensors="np",
|
30 |
-
padding=
|
31 |
truncation=True,
|
32 |
-
max_length=32
|
33 |
)
|
34 |
|
35 |
-
# Prepare ONNX inputs
|
36 |
onnx_inputs = {
|
37 |
"input_ids": inputs["input_ids"].astype(np.int64),
|
38 |
"attention_mask": inputs["attention_mask"].astype(np.int64)
|
@@ -41,21 +54,21 @@ async def predict(text: str):
|
|
41 |
# Run inference
|
42 |
outputs = session.run(None, onnx_inputs)
|
43 |
|
44 |
-
# Convert to
|
|
|
|
|
45 |
return {
|
46 |
-
"embedding":
|
47 |
"tokens": tokenizer.convert_ids_to_tokens(inputs["input_ids"][0])
|
48 |
}
|
49 |
|
50 |
except Exception as e:
|
51 |
-
raise HTTPException(status_code=
|
52 |
|
53 |
if __name__ == "__main__":
|
54 |
uvicorn.run(
|
55 |
-
app,
|
56 |
host="0.0.0.0",
|
57 |
port=7860,
|
58 |
-
|
59 |
-
proxy_headers=True,
|
60 |
-
forwarded_allow_ips="*"
|
61 |
)
|
|
|
1 |
+
from fastapi import FastAPI, HTTPException, Request
|
2 |
from onnxruntime import InferenceSession
|
3 |
from transformers import AutoTokenizer
|
4 |
import numpy as np
|
5 |
import os
|
6 |
+
import uvicorn
|
7 |
|
8 |
app = FastAPI()
|
9 |
|
10 |
+
# Initialize tokenizer
|
11 |
tokenizer = AutoTokenizer.from_pretrained(
|
12 |
"Xenova/multi-qa-mpnet-base-dot-v1",
|
13 |
+
use_fast=True,
|
14 |
legacy=False
|
15 |
)
|
16 |
|
17 |
# Load ONNX model
|
18 |
+
try:
|
19 |
+
session = InferenceSession("model.onnx")
|
20 |
+
print("Model loaded successfully")
|
21 |
+
except Exception as e:
|
22 |
+
print(f"Failed to load model: {str(e)}")
|
23 |
+
raise
|
24 |
|
25 |
@app.get("/")
|
26 |
+
def health_check():
|
27 |
+
return {"status": "OK", "model": "ONNX"}
|
28 |
|
29 |
@app.post("/api/predict")
|
30 |
+
async def predict(request: Request):
|
31 |
try:
|
32 |
+
# Get JSON input
|
33 |
+
data = await request.json()
|
34 |
+
text = data.get("text", "")
|
35 |
+
|
36 |
+
if not text:
|
37 |
+
raise HTTPException(status_code=400, detail="No text provided")
|
38 |
+
|
39 |
+
# Tokenize input
|
40 |
inputs = tokenizer(
|
41 |
text,
|
42 |
+
return_tensors="np",
|
43 |
+
padding="max_length",
|
44 |
truncation=True,
|
45 |
+
max_length=32
|
46 |
)
|
47 |
|
48 |
+
# Prepare ONNX inputs with correct shapes
|
49 |
onnx_inputs = {
|
50 |
"input_ids": inputs["input_ids"].astype(np.int64),
|
51 |
"attention_mask": inputs["attention_mask"].astype(np.int64)
|
|
|
54 |
# Run inference
|
55 |
outputs = session.run(None, onnx_inputs)
|
56 |
|
57 |
+
# Convert outputs to list and handle numpy types
|
58 |
+
embedding = outputs[0][0].astype(float).tolist() # First output, first batch
|
59 |
+
|
60 |
return {
|
61 |
+
"embedding": embedding,
|
62 |
"tokens": tokenizer.convert_ids_to_tokens(inputs["input_ids"][0])
|
63 |
}
|
64 |
|
65 |
except Exception as e:
|
66 |
+
raise HTTPException(status_code=500, detail=str(e))
|
67 |
|
68 |
if __name__ == "__main__":
|
69 |
uvicorn.run(
|
70 |
+
"app:app",
|
71 |
host="0.0.0.0",
|
72 |
port=7860,
|
73 |
+
reload=False
|
|
|
|
|
74 |
)
|