Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -1,4 +1,60 @@
|
|
|
|
|
|
1 |
import streamlit as st
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2 |
|
3 |
-
|
4 |
-
|
|
|
1 |
+
# app.py
|
2 |
+
|
3 |
import streamlit as st
|
4 |
+
from unsloth import FastLanguageModel
|
5 |
+
from transformers import TextStreamer
|
6 |
+
|
7 |
+
# To speed up model loading in repeated queries, you can use st.cache_resource (Streamlit 1.18+).
|
8 |
+
@st.cache_resource
|
9 |
+
def load_unsloth_model(
|
10 |
+
model_name="azizsi/model2",
|
11 |
+
max_seq_length=4096,
|
12 |
+
dtype="float16",
|
13 |
+
load_in_4bit=False
|
14 |
+
):
|
15 |
+
"""
|
16 |
+
Loads and prepares the model for inference using FastLanguageModel from Unsloth.
|
17 |
+
Returns (model, tokenizer).
|
18 |
+
"""
|
19 |
+
model, tokenizer = FastLanguageModel.from_pretrained(
|
20 |
+
model_name=model_name,
|
21 |
+
max_seq_length=max_seq_length,
|
22 |
+
dtype=dtype,
|
23 |
+
load_in_4bit=load_in_4bit
|
24 |
+
)
|
25 |
+
|
26 |
+
# Enable 2x faster inference (per Unsloth docs)
|
27 |
+
FastLanguageModel.for_inference(model)
|
28 |
+
|
29 |
+
return model, tokenizer
|
30 |
+
|
31 |
+
|
32 |
+
def main():
|
33 |
+
st.title("Unsloth Model Demo")
|
34 |
+
|
35 |
+
# Provide a text input area for the user
|
36 |
+
user_input = st.text_area("Enter your prompt:", "")
|
37 |
+
|
38 |
+
# Generate button
|
39 |
+
if st.button("Generate"):
|
40 |
+
with st.spinner("Generating response..."):
|
41 |
+
# Load the model & tokenizer
|
42 |
+
model, tokenizer = load_unsloth_model()
|
43 |
+
|
44 |
+
# Create a TextStreamer to stream tokens or capture final text
|
45 |
+
streamer = TextStreamer(tokenizer)
|
46 |
+
|
47 |
+
# Tokenize user prompt and move to GPU (or the model's device)
|
48 |
+
inputs = tokenizer(user_input, return_tensors="pt").to(model.device)
|
49 |
+
|
50 |
+
# Generate up to 128 new tokens (modify as desired)
|
51 |
+
outputs = model.generate(**inputs, streamer=streamer, max_new_tokens=128)
|
52 |
+
|
53 |
+
# If you want to display the entire response at once:
|
54 |
+
generated_text = tokenizer.decode(outputs[0], skip_special_tokens=True)
|
55 |
+
|
56 |
+
st.markdown("**Response:**")
|
57 |
+
st.write(generated_text)
|
58 |
|
59 |
+
if __name__ == "__main__":
|
60 |
+
main()
|