Create handler.py
Browse files- handler.py +29 -0
handler.py
ADDED
@@ -0,0 +1,29 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline
|
2 |
+
import torch
|
3 |
+
|
4 |
+
class EndpointHandler:
|
5 |
+
def __init__(self, path=""):
|
6 |
+
self.tokenizer = AutoTokenizer.from_pretrained(path)
|
7 |
+
self.model = AutoModelForCausalLM.from_pretrained(path, torch_dtype=torch.float16, device_map="auto")
|
8 |
+
self.pipe = pipeline(
|
9 |
+
"text-generation",
|
10 |
+
model=self.model,
|
11 |
+
tokenizer=self.tokenizer,
|
12 |
+
max_new_tokens=512,
|
13 |
+
do_sample=True,
|
14 |
+
temperature=0.7,
|
15 |
+
top_p=0.95,
|
16 |
+
)
|
17 |
+
|
18 |
+
def __call__(self, data):
|
19 |
+
prompt = data.get("inputs", "")
|
20 |
+
if not prompt:
|
21 |
+
return {"error": "No input provided"}
|
22 |
+
|
23 |
+
# Generate response
|
24 |
+
response = self.pipe(prompt)[0]['generated_text']
|
25 |
+
|
26 |
+
# Remove the original prompt from the response
|
27 |
+
response = response[len(prompt):].strip()
|
28 |
+
|
29 |
+
return {"generated_text": response}
|