Spaces:
Sleeping
Sleeping
Upload app.py
Browse files
app.py
CHANGED
@@ -1,5 +1,5 @@
|
|
1 |
# Install the necessary packages
|
2 |
-
# pip install accelerate transformers fastapi pydantic torch
|
3 |
|
4 |
from transformers import AutoModelForCausalLM, AutoTokenizer
|
5 |
import torch
|
@@ -9,14 +9,15 @@ from fastapi import FastAPI
|
|
9 |
# Initialize the FastAPI app
|
10 |
app = FastAPI(docs_url="/")
|
11 |
|
12 |
-
#
|
13 |
-
device = "cuda"
|
14 |
|
|
|
15 |
model = AutoModelForCausalLM.from_pretrained(
|
16 |
"Qwen/Qwen1.5-0.5B-Chat",
|
17 |
torch_dtype="auto",
|
18 |
device_map="auto"
|
19 |
-
)
|
20 |
tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen1.5-0.5B-Chat")
|
21 |
|
22 |
# Define the request model
|
|
|
1 |
# Install the necessary packages
|
2 |
+
# pip install accelerate transformers fastapi pydantic torch jinja2
|
3 |
|
4 |
from transformers import AutoModelForCausalLM, AutoTokenizer
|
5 |
import torch
|
|
|
9 |
# Initialize the FastAPI app
|
10 |
app = FastAPI(docs_url="/")
|
11 |
|
12 |
+
# Determine the device to use
|
13 |
+
device = "cuda" if torch.cuda.is_available() else "cpu"
|
14 |
|
15 |
+
# Load the model and tokenizer once at startup
|
16 |
model = AutoModelForCausalLM.from_pretrained(
|
17 |
"Qwen/Qwen1.5-0.5B-Chat",
|
18 |
torch_dtype="auto",
|
19 |
device_map="auto"
|
20 |
+
).to(device)
|
21 |
tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen1.5-0.5B-Chat")
|
22 |
|
23 |
# Define the request model
|