Spaces:
Sleeping
Sleeping
Update src/model_load.py
Browse files- src/model_load.py +2 -2
src/model_load.py
CHANGED
@@ -9,13 +9,13 @@ from langchain.chains import LLMChain
|
|
9 |
def load_model():
|
10 |
|
11 |
model_name="tiiuae/Falcon3-10B-Instruct"
|
12 |
-
max_memory = {0: "
|
13 |
# Cargar tokenizer y modelo de Hugging Face
|
14 |
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
15 |
model = AutoModelForCausalLM.from_pretrained(model_name,
|
16 |
device_map="auto",
|
17 |
quantization_config=BitsAndBytesConfig(load_in_8bit=True),
|
18 |
-
torch_dtype=torch.
|
19 |
max_memory=max_memory)
|
20 |
|
21 |
# Crear pipeline de generaci贸n de texto
|
|
|
9 |
def load_model():
|
10 |
|
11 |
model_name="tiiuae/Falcon3-10B-Instruct"
|
12 |
+
max_memory = {0: "24GB", "cpu": "30GB"}
|
13 |
# Cargar tokenizer y modelo de Hugging Face
|
14 |
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
15 |
model = AutoModelForCausalLM.from_pretrained(model_name,
|
16 |
device_map="auto",
|
17 |
quantization_config=BitsAndBytesConfig(load_in_8bit=True),
|
18 |
+
torch_dtype=torch.float16,
|
19 |
max_memory=max_memory)
|
20 |
|
21 |
# Crear pipeline de generaci贸n de texto
|