Create handler.py
Browse files- handler.py +31 -0
handler.py
ADDED
@@ -0,0 +1,31 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from typing import Dict, Any, List
|
2 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer
|
3 |
+
import torch
|
4 |
+
|
5 |
+
class EndpointHandler():
|
6 |
+
def __init__(self, path=""):
|
7 |
+
# Load the model in FP16 to reduce memory usage while retaining performance.
|
8 |
+
self.model = AutoModelForCausalLM.from_pretrained(path, torch_dtype=torch.float16)
|
9 |
+
self.tokenizer = AutoTokenizer.from_pretrained(path)
|
10 |
+
|
11 |
+
def __call__(self, data: Dict[str, Any]) -> List[Dict[str, Any]]:
|
12 |
+
"""
|
13 |
+
data args:
|
14 |
+
inputs (str): The text input or prompts for the model
|
15 |
+
Return:
|
16 |
+
A list containing the generated responses.
|
17 |
+
"""
|
18 |
+
# Extract the input text from the request
|
19 |
+
inputs = data.get("inputs", "")
|
20 |
+
if not inputs:
|
21 |
+
return [{"error": "No input provided"}]
|
22 |
+
|
23 |
+
# Tokenize the input and run the model to generate output
|
24 |
+
tokens = self.tokenizer(inputs, return_tensors="pt").to(torch.float16)
|
25 |
+
output_tokens = self.model.generate(**tokens)
|
26 |
+
|
27 |
+
# Decode the generated tokens back to text
|
28 |
+
output_text = self.tokenizer.decode(output_tokens[0], skip_special_tokens=True)
|
29 |
+
|
30 |
+
# Return the generated response as a list (required format)
|
31 |
+
return [{"generated_text": output_text}]
|