- add custom endpoint handler
Browse files- handler.py +36 -0
handler.py
ADDED
|
@@ -0,0 +1,36 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
from typing import Dict, List, Any
|
| 2 |
+
|
| 3 |
+
import torch as torch
|
| 4 |
+
from transformers import pipeline
|
| 5 |
+
|
| 6 |
+
|
| 7 |
+
class EndpointHandler():
|
| 8 |
+
|
| 9 |
+
|
| 10 |
+
|
| 11 |
+
def __init__(self, path=""):
|
| 12 |
+
device = 0 if torch.cuda.is_available() else "cpu"
|
| 13 |
+
self.pipe = pipeline(
|
| 14 |
+
task="automatic-speech-recognition",
|
| 15 |
+
model="openai/whisper-large",
|
| 16 |
+
chunk_length_s=30,
|
| 17 |
+
device=device,
|
| 18 |
+
)
|
| 19 |
+
self.pipe.model.config.forced_decoder_ids = self.pipe.tokenizer.get_decoder_prompt_ids(language="nl", task="transcribe")
|
| 20 |
+
|
| 21 |
+
def __call__(self, data: Dict[str, Any]) -> List[Dict[str, Any]]:
|
| 22 |
+
"""
|
| 23 |
+
data args:
|
| 24 |
+
inputs (:obj: `str`)
|
| 25 |
+
date (:obj: `str`)
|
| 26 |
+
Return:
|
| 27 |
+
A :obj:`list` | `dict`: will be serialized and returned
|
| 28 |
+
"""
|
| 29 |
+
#print request
|
| 30 |
+
print("request")
|
| 31 |
+
print(data)
|
| 32 |
+
# get inputs
|
| 33 |
+
inputs = data.pop("inputs", data)
|
| 34 |
+
|
| 35 |
+
text = self.pipe(inputs)["text"]
|
| 36 |
+
return text
|