Create handler.py
Browse files- handler.py +29 -0
handler.py
ADDED
@@ -0,0 +1,29 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from typing import Dict, List, Any
|
2 |
+
from transformers import AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfig, GemmaTokenizer
|
3 |
+
import torch
|
4 |
+
from peft import PeftModel
|
5 |
+
import json
|
6 |
+
import os
|
7 |
+
|
8 |
+
class EndpointHandler():
|
9 |
+
def __init__(self, path=""):
|
10 |
+
|
11 |
+
bnb_config = BitsAndBytesConfig(
|
12 |
+
load_in_8bit=True,
|
13 |
+
bnb_8bit_quant_type="nf4",
|
14 |
+
bnb_8bit_compute_dtype=torch.bfloat16
|
15 |
+
)
|
16 |
+
|
17 |
+
tokenizer = AutoTokenizer.from_pretrained('LexiconShiftInnovations/Gemma_Dental_it_07_merged')
|
18 |
+
model = AutoModelForCausalLM.from_pretrained('LexiconShiftInnovations/Gemma_Dental_it_07_merged', quantization_config=bnb_config, device_map={"":0})
|
19 |
+
|
20 |
+
self.pipeline = pipeline("text-generation", model=model, tokenizer=tokenizer)
|
21 |
+
|
22 |
+
def __call__(self, data: Any) -> List[List[Dict[str, float]]]:
|
23 |
+
inputs = data.pop("inputs", data)
|
24 |
+
parameters = data.pop("parameters", None)
|
25 |
+
if parameters is not None:
|
26 |
+
prediction = self.pipeline(inputs, **parameters)
|
27 |
+
else:
|
28 |
+
prediction = self.pipeline(inputs)
|
29 |
+
return prediction
|