Update handler.py
Browse files- handler.py +1 -4
handler.py
CHANGED
|
@@ -4,10 +4,7 @@ from transformers import pipeline, AutoModelForCausalLM, AutoTokenizer, BitsAndB
|
|
| 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(
|
|
|
|
| 4 |
|
| 5 |
class EndpointHandler():
|
| 6 |
def __init__(self, path=""):
|
| 7 |
+
quantization_config = BitsAndBytesConfig(load_in_4bit=True, bnb_4bit_compute_dtype=torch.float16)
|
|
|
|
|
|
|
|
|
|
| 8 |
# load the optimized model
|
| 9 |
tokenizer = AutoTokenizer.from_pretrained(path)
|
| 10 |
model = AutoModelForCausalLM.from_pretrained(
|