Spaces:
Runtime error
Runtime error
wjbmattingly
commited on
Commit
•
8335937
1
Parent(s):
83f93f3
init
Browse files- app.py +59 -4
- requirements.txt +3 -0
app.py
CHANGED
@@ -1,7 +1,62 @@
|
|
1 |
import gradio as gr
|
|
|
|
|
|
|
2 |
|
3 |
-
|
4 |
-
|
5 |
|
6 |
-
|
7 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
import gradio as gr
|
2 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer
|
3 |
+
import torch
|
4 |
+
import spaces
|
5 |
|
6 |
+
# Load the model and tokenizer
|
7 |
+
model_name = "Qwen/Qwen2-72B-Instruct"
|
8 |
|
9 |
+
# Load model (without moving to GPU yet)
|
10 |
+
model = AutoModelForCausalLM.from_pretrained(
|
11 |
+
model_name,
|
12 |
+
torch_dtype=torch.float16,
|
13 |
+
trust_remote_code=True
|
14 |
+
)
|
15 |
+
tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
|
16 |
+
|
17 |
+
@spaces.GPU
|
18 |
+
def generate_text(prompt):
|
19 |
+
# Move model to GPU when function is called
|
20 |
+
model.to('cuda')
|
21 |
+
|
22 |
+
messages = [
|
23 |
+
{"role": "system", "content": "You are a helpful assistant."},
|
24 |
+
{"role": "user", "content": prompt}
|
25 |
+
]
|
26 |
+
|
27 |
+
text = tokenizer.apply_chat_template(
|
28 |
+
messages,
|
29 |
+
tokenize=False,
|
30 |
+
add_generation_prompt=True
|
31 |
+
)
|
32 |
+
model_inputs = tokenizer([text], return_tensors="pt").to('cuda')
|
33 |
+
|
34 |
+
with torch.no_grad():
|
35 |
+
generated_ids = model.generate(
|
36 |
+
model_inputs.input_ids,
|
37 |
+
temperature=0.7,
|
38 |
+
max_new_tokens=500,
|
39 |
+
do_sample=True,
|
40 |
+
top_p=0.95
|
41 |
+
)
|
42 |
+
generated_ids = [
|
43 |
+
output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
|
44 |
+
]
|
45 |
+
response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
|
46 |
+
|
47 |
+
# Move model back to CPU to free up GPU resources
|
48 |
+
model.to('cpu')
|
49 |
+
|
50 |
+
return response
|
51 |
+
|
52 |
+
# Create Gradio interface
|
53 |
+
iface = gr.Interface(
|
54 |
+
fn=generate_text,
|
55 |
+
inputs=gr.Textbox(lines=5, label="Input Prompt"),
|
56 |
+
outputs=gr.Textbox(label="Generated Text"),
|
57 |
+
title="Qwen Text Generator (Spaces GPU)",
|
58 |
+
description="Enter a prompt to generate text using the Qwen model. This Space uses Spaces GPU for efficient GPU usage."
|
59 |
+
)
|
60 |
+
|
61 |
+
# Launch the app
|
62 |
+
iface.launch()
|
requirements.txt
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
torch
|
2 |
+
spaces
|
3 |
+
transformers
|