Frederic Marvin Abraham
commited on
Commit
•
5d526dc
1
Parent(s):
a2a8837
add custom handler
Browse files- handler.py +27 -0
handler.py
ADDED
@@ -0,0 +1,27 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from typing import Dict, List, Any
|
2 |
+
|
3 |
+
import torch
|
4 |
+
from transformers import BertModel, BertTokenizerFast
|
5 |
+
|
6 |
+
|
7 |
+
class EndpointHandler():
|
8 |
+
def __init__(self, path_to_model: str):
|
9 |
+
# Preload all the elements you are going to need at inference.
|
10 |
+
# pseudo:
|
11 |
+
self.tokenizer = BertTokenizerFast.from_pretrained(path_to_model)
|
12 |
+
self.model = BertModel.from_pretrained(path_to_model)
|
13 |
+
self.model = self.model.eval()
|
14 |
+
|
15 |
+
def __call__(self, data: Dict[str, Any]) -> List[Dict[str, Any]]:
|
16 |
+
"""
|
17 |
+
This method is called whenever a request is made to the endpoint.
|
18 |
+
:param data: { inputs [str]: list of strings to be encoded }
|
19 |
+
:return: A :obj:`list` | `dict`: will be serialized and returned
|
20 |
+
"""
|
21 |
+
|
22 |
+
inputs = self.tokenizer(data['inputs'], return_tensors = "pt", padding = True)
|
23 |
+
|
24 |
+
with torch.no_grad():
|
25 |
+
outputs = self.model(**inputs)
|
26 |
+
|
27 |
+
return outputs.pooler_output
|