Spaces:
Sleeping
Sleeping
zubairsamo
commited on
Commit
•
c2b2585
1
Parent(s):
cc5a14b
Reversing Code With Alpaca Implementation
Browse files
app.py
CHANGED
@@ -1,27 +1,30 @@
|
|
1 |
-
from threading import Thread
|
2 |
|
3 |
-
import torch
|
4 |
-
import gradio as gr
|
5 |
-
from transformers import AutoTokenizer,
|
6 |
|
7 |
-
|
|
|
8 |
torch_device = "cuda" if torch.cuda.is_available() else "cpu"
|
9 |
print("Running on device:", torch_device)
|
10 |
print("CPU threads:", torch.get_num_threads())
|
11 |
|
|
|
12 |
if torch_device == "cuda":
|
13 |
-
model =
|
14 |
else:
|
15 |
-
model =
|
16 |
-
tokenizer = AutoTokenizer.from_pretrained(model_id)
|
17 |
|
|
|
18 |
|
|
|
19 |
def run_generation(user_text, top_p, temperature, top_k, max_new_tokens):
|
20 |
# Get the model and tokenizer, and tokenize the user text.
|
21 |
model_inputs = tokenizer([user_text], return_tensors="pt").to(torch_device)
|
22 |
|
23 |
-
# Start generation on a separate thread, so that we don't block the UI.
|
24 |
-
#
|
25 |
streamer = TextIteratorStreamer(tokenizer, timeout=10., skip_prompt=True, skip_special_tokens=True)
|
26 |
generate_kwargs = dict(
|
27 |
model_inputs,
|
@@ -32,37 +35,42 @@ def run_generation(user_text, top_p, temperature, top_k, max_new_tokens):
|
|
32 |
temperature=float(temperature),
|
33 |
top_k=top_k
|
34 |
)
|
|
|
|
|
35 |
t = Thread(target=model.generate, kwargs=generate_kwargs)
|
36 |
t.start()
|
37 |
|
38 |
-
# Pull the generated text from the streamer, and update the model output.
|
39 |
model_output = ""
|
40 |
for new_text in streamer:
|
41 |
model_output += new_text
|
42 |
yield model_output
|
43 |
return model_output
|
44 |
|
45 |
-
|
46 |
def reset_textbox():
|
47 |
return gr.update(value='')
|
48 |
|
49 |
-
|
50 |
with gr.Blocks() as demo:
|
51 |
-
#
|
52 |
gr.Markdown(
|
53 |
"# Testing ALPACA Model \n"
|
54 |
)
|
55 |
|
56 |
with gr.Row():
|
57 |
with gr.Column(scale=4):
|
|
|
58 |
user_text = gr.Textbox(
|
59 |
placeholder="Ask Me Anything ... ",
|
60 |
label="User input"
|
61 |
)
|
|
|
62 |
model_output = gr.Textbox(label="Model output", lines=10, interactive=False)
|
|
|
63 |
button_submit = gr.Button(value="Submit")
|
64 |
|
65 |
with gr.Column(scale=1):
|
|
|
66 |
max_new_tokens = gr.Slider(
|
67 |
minimum=1, maximum=1000, value=250, step=1, interactive=True, label="Max New Tokens",
|
68 |
)
|
@@ -76,7 +84,9 @@ with gr.Blocks() as demo:
|
|
76 |
minimum=0.1, maximum=5.0, value=0.8, step=0.1, interactive=True, label="Temperature",
|
77 |
)
|
78 |
|
|
|
79 |
user_text.submit(run_generation, [user_text, top_p, temperature, top_k, max_new_tokens], model_output)
|
80 |
button_submit.click(run_generation, [user_text, top_p, temperature, top_k, max_new_tokens], model_output)
|
81 |
|
82 |
-
|
|
|
|
1 |
+
from threading import Thread # Import the Thread class from the threading module
|
2 |
|
3 |
+
import torch # Import the PyTorch library
|
4 |
+
import gradio as gr # Import Gradio for creating a UI
|
5 |
+
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM, TextIteratorStreamer # Import Hugging Face Transformers
|
6 |
|
7 |
+
# Define the Hugging Face model ID and check for available GPU (cuda)
|
8 |
+
model_id = "declare-lab/flan-alpaca-large"
|
9 |
torch_device = "cuda" if torch.cuda.is_available() else "cpu"
|
10 |
print("Running on device:", torch_device)
|
11 |
print("CPU threads:", torch.get_num_threads())
|
12 |
|
13 |
+
# Load the pre-trained model based on the device
|
14 |
if torch_device == "cuda":
|
15 |
+
model = AutoModelForSeq2SeqLM.from_pretrained(model_id, load_in_8bit=True, device_map="auto")
|
16 |
else:
|
17 |
+
model = AutoModelForSeq2SeqLM.from_pretrained(model_id)
|
|
|
18 |
|
19 |
+
tokenizer = AutoTokenizer.from_pretrained(model_id)
|
20 |
|
21 |
+
# Define a function to run model text generation
|
22 |
def run_generation(user_text, top_p, temperature, top_k, max_new_tokens):
|
23 |
# Get the model and tokenizer, and tokenize the user text.
|
24 |
model_inputs = tokenizer([user_text], return_tensors="pt").to(torch_device)
|
25 |
|
26 |
+
# Start generation on a separate thread, so that we don't block the UI.
|
27 |
+
# Adds timeout to the streamer to handle exceptions in the generation thread.
|
28 |
streamer = TextIteratorStreamer(tokenizer, timeout=10., skip_prompt=True, skip_special_tokens=True)
|
29 |
generate_kwargs = dict(
|
30 |
model_inputs,
|
|
|
35 |
temperature=float(temperature),
|
36 |
top_k=top_k
|
37 |
)
|
38 |
+
|
39 |
+
# Create a new thread for model generation
|
40 |
t = Thread(target=model.generate, kwargs=generate_kwargs)
|
41 |
t.start()
|
42 |
|
|
|
43 |
model_output = ""
|
44 |
for new_text in streamer:
|
45 |
model_output += new_text
|
46 |
yield model_output
|
47 |
return model_output
|
48 |
|
49 |
+
# Define a function to reset the user input textbox
|
50 |
def reset_textbox():
|
51 |
return gr.update(value='')
|
52 |
|
53 |
+
# Create a Gradio UI interface
|
54 |
with gr.Blocks() as demo:
|
55 |
+
# Display a title
|
56 |
gr.Markdown(
|
57 |
"# Testing ALPACA Model \n"
|
58 |
)
|
59 |
|
60 |
with gr.Row():
|
61 |
with gr.Column(scale=4):
|
62 |
+
# Create a textbox for user input
|
63 |
user_text = gr.Textbox(
|
64 |
placeholder="Ask Me Anything ... ",
|
65 |
label="User input"
|
66 |
)
|
67 |
+
# Create a textbox for model output
|
68 |
model_output = gr.Textbox(label="Model output", lines=10, interactive=False)
|
69 |
+
# Create a submit button
|
70 |
button_submit = gr.Button(value="Submit")
|
71 |
|
72 |
with gr.Column(scale=1):
|
73 |
+
# Create sliders for adjusting generation parameters
|
74 |
max_new_tokens = gr.Slider(
|
75 |
minimum=1, maximum=1000, value=250, step=1, interactive=True, label="Max New Tokens",
|
76 |
)
|
|
|
84 |
minimum=0.1, maximum=5.0, value=0.8, step=0.1, interactive=True, label="Temperature",
|
85 |
)
|
86 |
|
87 |
+
# Set up the submission of user input
|
88 |
user_text.submit(run_generation, [user_text, top_p, temperature, top_k, max_new_tokens], model_output)
|
89 |
button_submit.click(run_generation, [user_text, top_p, temperature, top_k, max_new_tokens], model_output)
|
90 |
|
91 |
+
# Launch the Gradio interface
|
92 |
+
demo.queue(max_size=32).launch(enable_queue=True)
|