Spaces:
Paused
Paused
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,108 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import streamlit as st
|
2 |
+
from transformers import AutoTokenizer, AutoModelForCausalLM
|
3 |
+
import torch
|
4 |
+
from typing import List, Dict
|
5 |
+
import time
|
6 |
+
|
7 |
+
class LlamaDemo:
|
8 |
+
def __init__(self):
|
9 |
+
self.model_name = "meta-llama/Llama-2-7b-chat-hf"
|
10 |
+
# Initialize in lazy loading fashion
|
11 |
+
self._model = None
|
12 |
+
self._tokenizer = None
|
13 |
+
|
14 |
+
@property
|
15 |
+
def model(self):
|
16 |
+
if self._model is None:
|
17 |
+
self._model = AutoModelForCausalLM.from_pretrained(
|
18 |
+
self.model_name,
|
19 |
+
torch_dtype=torch.float16,
|
20 |
+
device_map="auto"
|
21 |
+
)
|
22 |
+
return self._model
|
23 |
+
|
24 |
+
@property
|
25 |
+
def tokenizer(self):
|
26 |
+
if self._tokenizer is None:
|
27 |
+
self._tokenizer = AutoTokenizer.from_pretrained(self.model_name)
|
28 |
+
return self._tokenizer
|
29 |
+
|
30 |
+
def generate_response(self, prompt: str, max_length: int = 512) -> str:
|
31 |
+
inputs = self.tokenizer(prompt, return_tensors="pt").to(self.model.device)
|
32 |
+
|
33 |
+
# Generate response
|
34 |
+
with torch.no_grad():
|
35 |
+
outputs = self.model.generate(
|
36 |
+
**inputs,
|
37 |
+
max_length=max_length,
|
38 |
+
num_return_sequences=1,
|
39 |
+
temperature=0.7,
|
40 |
+
do_sample=True,
|
41 |
+
pad_token_id=self.tokenizer.eos_token_id
|
42 |
+
)
|
43 |
+
|
44 |
+
response = self.tokenizer.decode(outputs[0], skip_special_tokens=True)
|
45 |
+
return response.replace(prompt, "").strip()
|
46 |
+
|
47 |
+
def main():
|
48 |
+
st.set_page_config(
|
49 |
+
page_title="Llama 3.1 Demo",
|
50 |
+
page_icon="🦙",
|
51 |
+
layout="wide"
|
52 |
+
)
|
53 |
+
|
54 |
+
st.title("🦙 Llama 3.1 Demo")
|
55 |
+
|
56 |
+
# Initialize session state
|
57 |
+
if 'llama' not in st.session_state:
|
58 |
+
st.session_state.llama = LlamaDemo()
|
59 |
+
|
60 |
+
if 'chat_history' not in st.session_state:
|
61 |
+
st.session_state.chat_history = []
|
62 |
+
|
63 |
+
# Chat interface
|
64 |
+
with st.container():
|
65 |
+
# Display chat history
|
66 |
+
for message in st.session_state.chat_history:
|
67 |
+
role = message["role"]
|
68 |
+
content = message["content"]
|
69 |
+
|
70 |
+
with st.chat_message(role):
|
71 |
+
st.write(content)
|
72 |
+
|
73 |
+
# Input for new message
|
74 |
+
if prompt := st.chat_input("What would you like to discuss?"):
|
75 |
+
# Add user message to chat history
|
76 |
+
st.session_state.chat_history.append({
|
77 |
+
"role": "user",
|
78 |
+
"content": prompt
|
79 |
+
})
|
80 |
+
|
81 |
+
with st.chat_message("user"):
|
82 |
+
st.write(prompt)
|
83 |
+
|
84 |
+
# Show assistant response
|
85 |
+
with st.chat_message("assistant"):
|
86 |
+
message_placeholder = st.empty()
|
87 |
+
|
88 |
+
with st.spinner("Generating response..."):
|
89 |
+
response = st.session_state.llama.generate_response(prompt)
|
90 |
+
message_placeholder.write(response)
|
91 |
+
|
92 |
+
# Add assistant response to chat history
|
93 |
+
st.session_state.chat_history.append({
|
94 |
+
"role": "assistant",
|
95 |
+
"content": response
|
96 |
+
})
|
97 |
+
|
98 |
+
# Sidebar with settings
|
99 |
+
with st.sidebar:
|
100 |
+
st.header("Settings")
|
101 |
+
max_length = st.slider("Maximum response length", 64, 1024, 512)
|
102 |
+
|
103 |
+
if st.button("Clear Chat History"):
|
104 |
+
st.session_state.chat_history = []
|
105 |
+
st.experimental_rerun()
|
106 |
+
|
107 |
+
if __name__ == "__main__":
|
108 |
+
main()
|