Spaces:
Running
Running
DexterSptizu
commited on
Commit
•
6ded5b8
1
Parent(s):
02a6b70
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,164 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import streamlit as st
|
2 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer
|
3 |
+
import torch
|
4 |
+
from datetime import datetime
|
5 |
+
|
6 |
+
# Initialize session state for chat history
|
7 |
+
if 'messages' not in st.session_state:
|
8 |
+
st.session_state.messages = []
|
9 |
+
|
10 |
+
@st.cache_resource
|
11 |
+
def load_model():
|
12 |
+
tokenizer = AutoTokenizer.from_pretrained("amd/AMD-OLMo-1B-SFT")
|
13 |
+
model = AutoModelForCausalLM.from_pretrained("amd/AMD-OLMo-1B-SFT")
|
14 |
+
if torch.cuda.is_available():
|
15 |
+
model = model.to("cuda")
|
16 |
+
return model, tokenizer
|
17 |
+
|
18 |
+
def generate_response(prompt, model, tokenizer, history):
|
19 |
+
# Format conversation history with the template
|
20 |
+
bos = tokenizer.eos_token
|
21 |
+
conversation = ""
|
22 |
+
for msg in history:
|
23 |
+
if msg["role"] == "user":
|
24 |
+
conversation += f"<|user|>\n{msg['content']}\n"
|
25 |
+
else:
|
26 |
+
conversation += f"<|assistant|>\n{msg['content']}\n"
|
27 |
+
|
28 |
+
template = bos + conversation + f"<|user|>\n{prompt}\n<|assistant|>\n"
|
29 |
+
|
30 |
+
inputs = tokenizer([template], return_tensors='pt', return_token_type_ids=False)
|
31 |
+
if torch.cuda.is_available():
|
32 |
+
inputs = inputs.to("cuda")
|
33 |
+
|
34 |
+
outputs = model.generate(
|
35 |
+
**inputs,
|
36 |
+
max_new_tokens=1000,
|
37 |
+
do_sample=True,
|
38 |
+
top_k=50,
|
39 |
+
top_p=0.95,
|
40 |
+
temperature=0.7
|
41 |
+
)
|
42 |
+
|
43 |
+
response = tokenizer.decode(outputs[0], skip_special_tokens=True)
|
44 |
+
# Extract only the assistant's last response
|
45 |
+
response = response.split("<|assistant|>\n")[-1].strip()
|
46 |
+
return response
|
47 |
+
|
48 |
+
def main():
|
49 |
+
st.set_page_config(page_title="AMD-OLMo Chatbot", layout="wide")
|
50 |
+
|
51 |
+
# Custom CSS
|
52 |
+
st.markdown("""
|
53 |
+
<style>
|
54 |
+
.stTab {
|
55 |
+
font-size: 20px;
|
56 |
+
}
|
57 |
+
.model-info {
|
58 |
+
background-color: #f0f2f6;
|
59 |
+
padding: 20px;
|
60 |
+
border-radius: 10px;
|
61 |
+
}
|
62 |
+
.chat-message {
|
63 |
+
padding: 10px;
|
64 |
+
border-radius: 10px;
|
65 |
+
margin: 5px 0;
|
66 |
+
}
|
67 |
+
.user-message {
|
68 |
+
background-color: #e6f3ff;
|
69 |
+
}
|
70 |
+
.assistant-message {
|
71 |
+
background-color: #f0f2f6;
|
72 |
+
}
|
73 |
+
</style>
|
74 |
+
""", unsafe_allow_html=True)
|
75 |
+
|
76 |
+
# Create tabs
|
77 |
+
tab1, tab2 = st.tabs(["Model Information", "Chat Interface"])
|
78 |
+
|
79 |
+
with tab1:
|
80 |
+
st.title("AMD-OLMo-1B-SFT Model Information")
|
81 |
+
|
82 |
+
st.markdown("""
|
83 |
+
## Model Overview
|
84 |
+
AMD-OLMo-1B-SFT is a state-of-the-art language model developed by AMD[1][2]. Key features include:
|
85 |
+
|
86 |
+
### Architecture
|
87 |
+
- **Base Model**: 1.2B parameters
|
88 |
+
- **Layers**: 16
|
89 |
+
- **Attention Heads**: 16
|
90 |
+
- **Hidden Size**: 2048
|
91 |
+
- **Context Length**: 2048
|
92 |
+
- **Vocabulary Size**: 50,280
|
93 |
+
|
94 |
+
### Training Details
|
95 |
+
- Pre-trained on 1.3 trillion tokens from Dolma v1.7
|
96 |
+
- Supervised fine-tuned (SFT) in two phases:
|
97 |
+
1. Tulu V2 dataset
|
98 |
+
2. OpenHermes-2.5, WebInstructSub, and Code-Feedback datasets
|
99 |
+
|
100 |
+
### Capabilities
|
101 |
+
- General text generation
|
102 |
+
- Question answering
|
103 |
+
- Code understanding
|
104 |
+
- Reasoning tasks
|
105 |
+
- Instruction following
|
106 |
+
|
107 |
+
### Hardware Requirements
|
108 |
+
- Optimized for AMD Instinct™ MI250 GPUs
|
109 |
+
- Training performed on 16 nodes with 4 GPUs each
|
110 |
+
""")
|
111 |
+
|
112 |
+
with tab2:
|
113 |
+
st.title("Chat with AMD-OLMo")
|
114 |
+
|
115 |
+
# Load model
|
116 |
+
try:
|
117 |
+
model, tokenizer = load_model()
|
118 |
+
st.success("Model loaded successfully! You can start chatting.")
|
119 |
+
except Exception as e:
|
120 |
+
st.error(f"Error loading model: {str(e)}")
|
121 |
+
return
|
122 |
+
|
123 |
+
# Chat interface
|
124 |
+
st.markdown("### Chat History")
|
125 |
+
chat_container = st.container()
|
126 |
+
|
127 |
+
with chat_container:
|
128 |
+
for message in st.session_state.messages:
|
129 |
+
div_class = "user-message" if message["role"] == "user" else "assistant-message"
|
130 |
+
st.markdown(f"""
|
131 |
+
<div class="chat-message {div_class}">
|
132 |
+
<b>{message["role"].title()}:</b> {message["content"]}
|
133 |
+
</div>
|
134 |
+
""", unsafe_allow_html=True)
|
135 |
+
|
136 |
+
# User input
|
137 |
+
with st.container():
|
138 |
+
user_input = st.text_area("Your message:", key="user_input", height=100)
|
139 |
+
col1, col2, col3 = st.columns([1, 1, 4])
|
140 |
+
|
141 |
+
with col1:
|
142 |
+
if st.button("Send"):
|
143 |
+
if user_input.strip():
|
144 |
+
# Add user message to history
|
145 |
+
st.session_state.messages.append({"role": "user", "content": user_input})
|
146 |
+
|
147 |
+
# Generate response
|
148 |
+
with st.spinner("Thinking..."):
|
149 |
+
response = generate_response(user_input, model, tokenizer, st.session_state.messages)
|
150 |
+
|
151 |
+
# Add assistant response to history
|
152 |
+
st.session_state.messages.append({"role": "assistant", "content": response})
|
153 |
+
|
154 |
+
# Clear input
|
155 |
+
st.session_state.user_input = ""
|
156 |
+
st.experimental_rerun()
|
157 |
+
|
158 |
+
with col2:
|
159 |
+
if st.button("Clear History"):
|
160 |
+
st.session_state.messages = []
|
161 |
+
st.experimental_rerun()
|
162 |
+
|
163 |
+
if __name__ == "__main__":
|
164 |
+
main()
|