Create handler.py
Browse files- handler.py +28 -0
handler.py
ADDED
@@ -0,0 +1,28 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from typing import Any, Dict, List
|
2 |
+
import torch
|
3 |
+
import transformers
|
4 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer
|
5 |
+
|
6 |
+
dtype = torch.bfloat16 if torch.cuda.get_device_capability()[0] ==8 else torch.float16
|
7 |
+
|
8 |
+
class EndpointHandler:
|
9 |
+
def __init__(self, path=""):
|
10 |
+
self.tokenizer = AutoTokenizer.from_pretrained(path)
|
11 |
+
self.model = AutoModelForCausalLM.from_pretrained(path, trust_remote_code=True, revision="main")
|
12 |
+
self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
13 |
+
self.model = self.model.to(self.device)
|
14 |
+
|
15 |
+
|
16 |
+
def __call__(self, data: Dict[str, Any]) -> Dict[str, Any]:
|
17 |
+
prompt = data["inputs"]
|
18 |
+
if "config" in data:
|
19 |
+
config = data.pop("config", None)
|
20 |
+
else:
|
21 |
+
config = {'max_new_tokens':100}
|
22 |
+
input_ids = self.tokenizer(prompt, return_tensors="pt").input_ids.to(self.device)
|
23 |
+
generated_ids = self.model.generate(input_ids, **config)
|
24 |
+
generated_text = self.tokenizer.decode(generated_ids[0], skip_special_tokens=True)
|
25 |
+
return [{"generated_text": generated_text}]
|
26 |
+
|
27 |
+
|
28 |
+
|