fwittel
commited on
Commit
·
040e104
1
Parent(s):
4c61288
Added tokenizer to handler.py
Browse files- handler.py +14 -4
handler.py
CHANGED
@@ -1,16 +1,26 @@
|
|
1 |
from typing import Dict, List, Any
|
2 |
-
from transformers import AutoModel, pipeline
|
|
|
3 |
|
4 |
class EndpointHandler:
|
5 |
def __init__(self, path=""):
|
6 |
# load the model
|
|
|
7 |
model = AutoModel.from_pretrained(path, low_cpu_mem_usage=True)
|
8 |
# create inference pipeline
|
9 |
# Do I have to check device?
|
10 |
-
self.pipeline = pipeline(
|
|
|
11 |
|
12 |
# (Might have to adjust typing)
|
13 |
def __call__(self, data: Any) -> List[List[Dict[str, float]]]:
|
14 |
-
inputs = data.pop("inputs", data)
|
15 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
16 |
return prediction
|
|
|
1 |
from typing import Dict, List, Any
|
2 |
+
from transformers import AutoModel, AutoTokenizer, pipeline
|
3 |
+
|
4 |
|
5 |
class EndpointHandler:
|
6 |
def __init__(self, path=""):
|
7 |
# load the model
|
8 |
+
tokenizer = AutoTokenizer.from_pretrained(path)
|
9 |
model = AutoModel.from_pretrained(path, low_cpu_mem_usage=True)
|
10 |
# create inference pipeline
|
11 |
# Do I have to check device?
|
12 |
+
self.pipeline = pipeline(
|
13 |
+
"text-generation", model=model, tokenizer=tokenizer)
|
14 |
|
15 |
# (Might have to adjust typing)
|
16 |
def __call__(self, data: Any) -> List[List[Dict[str, float]]]:
|
17 |
+
inputs = data.pop("inputs", data)
|
18 |
+
parameters = data.pop("parameters", None)
|
19 |
+
|
20 |
+
# pass inputs with all kwargs in data
|
21 |
+
if parameters is not None:
|
22 |
+
prediction = self.pipeline(inputs, **parameters)
|
23 |
+
else:
|
24 |
+
prediction = self.pipeline(inputs)
|
25 |
+
# postprocess the prediction
|
26 |
return prediction
|