Update handler.py
Browse files- handler.py +6 -1
handler.py
CHANGED
|
@@ -1,13 +1,18 @@
|
|
| 1 |
from typing import Dict, List, Any
|
| 2 |
-
from transformers import pipeline, AutoModelForCausalLM, AutoTokenizer
|
| 3 |
|
| 4 |
|
| 5 |
class EndpointHandler():
|
| 6 |
def __init__(self, path=""):
|
|
|
|
|
|
|
|
|
|
|
|
|
| 7 |
# load the optimized model
|
| 8 |
tokenizer = AutoTokenizer.from_pretrained(path)
|
| 9 |
model = AutoModelForCausalLM.from_pretrained(
|
| 10 |
path,
|
|
|
|
| 11 |
device_map="auto",
|
| 12 |
torch_dtype='auto'
|
| 13 |
).eval()
|
|
|
|
| 1 |
from typing import Dict, List, Any
|
| 2 |
+
from transformers import pipeline, AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
|
| 3 |
|
| 4 |
|
| 5 |
class EndpointHandler():
|
| 6 |
def __init__(self, path=""):
|
| 7 |
+
quantization_config = BitsAndBytesConfig(
|
| 8 |
+
load_in_4bit=True,
|
| 9 |
+
bnb_4bit_compute_dtype=torch.float16
|
| 10 |
+
)
|
| 11 |
# load the optimized model
|
| 12 |
tokenizer = AutoTokenizer.from_pretrained(path)
|
| 13 |
model = AutoModelForCausalLM.from_pretrained(
|
| 14 |
path,
|
| 15 |
+
quantization_config=quantization_config
|
| 16 |
device_map="auto",
|
| 17 |
torch_dtype='auto'
|
| 18 |
).eval()
|