Create handler.py
Browse files- handler.py +57 -0
handler.py
ADDED
@@ -0,0 +1,57 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from transformers import AutoTokenizer, AutoModelForCausalLM, TextGenerationPipeline
|
2 |
+
import torch
|
3 |
+
|
4 |
+
def load_model(model_id):
|
5 |
+
tokenizer = AutoTokenizer.from_pretrained(model_id)
|
6 |
+
model = AutoModelForCausalLM.from_pretrained(
|
7 |
+
model_id,
|
8 |
+
device_map="auto",
|
9 |
+
torch_dtype=torch.float16,
|
10 |
+
load_in_4bit=True
|
11 |
+
)
|
12 |
+
return model, tokenizer
|
13 |
+
|
14 |
+
class EndpointHandler:
|
15 |
+
def __init__(self, path=""):
|
16 |
+
self.model, self.tokenizer = load_model(path)
|
17 |
+
self.pipeline = TextGenerationPipeline(
|
18 |
+
model=self.model,
|
19 |
+
tokenizer=self.tokenizer
|
20 |
+
)
|
21 |
+
|
22 |
+
def __call__(self, data):
|
23 |
+
# Extraire le texte d'entrée
|
24 |
+
if isinstance(data, dict):
|
25 |
+
text = data.get("inputs", "")
|
26 |
+
else:
|
27 |
+
text = data
|
28 |
+
|
29 |
+
# Paramètres de génération par défaut
|
30 |
+
generation_kwargs = {
|
31 |
+
"max_new_tokens": 512,
|
32 |
+
"temperature": 0.7,
|
33 |
+
"top_p": 0.95,
|
34 |
+
"repetition_penalty": 1.15,
|
35 |
+
"do_sample": True,
|
36 |
+
"pad_token_id": self.tokenizer.pad_token_id,
|
37 |
+
"eos_token_id": self.tokenizer.eos_token_id,
|
38 |
+
}
|
39 |
+
|
40 |
+
# Mettre à jour avec les paramètres de la requête si fournis
|
41 |
+
if isinstance(data, dict) and "parameters" in data:
|
42 |
+
generation_kwargs.update(data["parameters"])
|
43 |
+
|
44 |
+
try:
|
45 |
+
# Générer la réponse
|
46 |
+
outputs = self.pipeline(
|
47 |
+
text,
|
48 |
+
**generation_kwargs
|
49 |
+
)
|
50 |
+
|
51 |
+
# Formater la sortie en tableau comme requis par l'API
|
52 |
+
if isinstance(outputs, list):
|
53 |
+
return [{"generated_text": output["generated_text"]} for output in outputs]
|
54 |
+
return [{"generated_text": outputs["generated_text"]}]
|
55 |
+
|
56 |
+
except Exception as e:
|
57 |
+
return [{"error": str(e)}]
|