File size: 6,961 Bytes
5a18dfb
8e164af
 
5a18dfb
a05c286
5a18dfb
 
 
 
 
 
 
 
a05c286
 
5a18dfb
 
 
 
 
 
 
 
 
 
 
034153d
5a18dfb
 
 
 
8e164af
 
 
 
 
 
 
034153d
7efaceb
 
 
 
034153d
 
5a18dfb
7efaceb
5a18dfb
7efaceb
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5a18dfb
 
 
 
 
 
 
 
 
 
 
 
 
7efaceb
5a18dfb
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
8e164af
 
5a18dfb
 
a05c286
5a18dfb
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
d159929
 
 
 
5a18dfb
8e164af
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5a18dfb
 
 
 
 
8e164af
5a18dfb
 
8e164af
5a18dfb
 
 
 
 
 
 
8e164af
5a18dfb
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
8e164af
5a18dfb
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
8e164af
5a18dfb
 
8e164af
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
from threading import Thread
import gradio as gr
import random
import torch
import spaces
from transformers import (
    AutoModelForCausalLM,
    AutoTokenizer,
    AutoConfig,
    TextIteratorStreamer
)

# Constants for the model and configuration
MODEL_ID = "AstroMLab/AstroSage-8B"
WINDOW_SIZE = 2048
DEVICE = "cuda"

# Load model configuration, tokenizer, and model
config = AutoConfig.from_pretrained(pretrained_model_name_or_path=MODEL_ID)
tokenizer = AutoTokenizer.from_pretrained(pretrained_model_name_or_path=MODEL_ID)
model = AutoModelForCausalLM.from_pretrained(
    pretrained_model_name_or_path=MODEL_ID,
    config=config,
    device_map="auto",
    use_safetensors=True,
    trust_remote_code=True,
    load_in_4bit=True,
    torch_dtype=torch.bfloat16
)

# Placeholder responses for when context is empty
GREETING_MESSAGES = [
    "Greetings! I am AstroSage, your guide to the cosmos. What would you like to explore today?",
    "Welcome to our cosmic journey! I am AstroSage. How may I assist you in understanding the universe?",
    "AstroSage here. Ready to explore the mysteries of space and time. How may I be of assistance?",
    "The universe awaits! I'm AstroSage. What astronomical wonders shall we discuss?",
]


def format_message(role: str, content: str) -> str:
    """Format a single message according to Llama-3 chat template."""
    return f"<|start_header_id|>{role}<|end_header_id|>\n\n{content}<|eot_id|>"


def generate_text(prompt: str, history: list, max_new_tokens=512, temperature=0.7, top_p=0.95):
    """
    Generate a response using the transformer model with proper Llama-3 chat formatting.
    """
    # Start with begin_of_text token
    formatted_messages = ["<|begin_of_text|>"]
    
    # Add formatted history
    for msg in history:
        formatted_message = format_message(msg['role'], msg['content'])
        formatted_messages.append(formatted_message)
    
    # Add the current prompt
    formatted_message = format_message('user', prompt)
    formatted_messages.append(formatted_message)
    
    # Add the start of assistant's response
    formatted_messages.append("<|start_header_id|>assistant<|end_header_id|>\n\n")
    
    # Combine all messages
    prompt_with_history = "\n".join(formatted_messages)
    
    # Encode the prompt
    inputs = tokenizer([prompt_with_history], return_tensors="pt", truncation=True).to(DEVICE)
    input_length = inputs["input_ids"].shape[-1]
    max_new_tokens = min(max_new_tokens, WINDOW_SIZE - input_length)

    # Prepare text streamer for live updates
    streamer = TextIteratorStreamer(
        tokenizer=tokenizer,
        timeout=10.0,
        skip_prompt=True,
        skip_special_tokens=True
    )
    
    generation_kwargs = dict(
        **inputs,
        streamer=streamer,
        max_new_tokens=max_new_tokens,
        do_sample=True,
        top_p=top_p,
        temperature=temperature,
    )

    # Generate the response in a separate thread for streaming
    thread = Thread(target=model.generate, kwargs=generation_kwargs)
    thread.start()

    # Collect and return the response
    response = ""
    for new_text in streamer:
        response += new_text
        yield response


def user(user_message, history):
    """
    Add the user's message to the history.
    """
    if history is None:
        history = []
    return "", history + [{"role": "user", "content": user_message}]

@spaces.GPU
def bot(history):
    """
    Generate the bot's response based on the history.
    """
    if not history:
        history = [{"role": "assistant", "content": random.choice(GREETING_MESSAGES)}]
    last_user_message = history[-1]["content"] if history else ""
    response_generator = generate_text(last_user_message, history)
    history.append({"role": "assistant", "content": ""})

    # Stream the response back
    for partial_response in response_generator:
        history[-1]["content"] = partial_response
        yield history


def initial_greeting():
    """
    Return the initial greeting message.
    """
    return [
        {"role": "system","content": "You are AstroSage, an intelligent AI assistant specializing in astronomy, astrophysics, and cosmology. Provide accurate, scientific information while making complex concepts accessible. You're enthusiastic about space exploration and maintain a sense of wonder about the cosmos. Start by introducing yourself."},
        {"role": "assistant", "content": random.choice(GREETING_MESSAGES)}
    ]


# Custom CSS for a space theme
custom_css = """
#component-0 {
    background-color: #1a1a2e;
    border-radius: 15px;
    padding: 20px;
}
.dark {
    background-color: #0f0f1a;
}
.contain {
    max-width: 1200px !important;
}
"""

# Create the Gradio interface
with gr.Blocks(css=custom_css, theme=gr.themes.Soft(primary_hue="indigo", neutral_hue="slate")) as demo:
    gr.Markdown(
        """
        # 🌌 AstroSage: Your Cosmic AI Companion
        
        Welcome to AstroSage, an advanced AI assistant specializing in astronomy, astrophysics, and cosmology. 
        Powered by the AstroSage-Llama-3.1-8B model, I'm here to help you explore the wonders of the universe!
        
        ### What Can I Help You With?
        - πŸͺ Explanations of astronomical phenomena
        - πŸš€ Space exploration and missions
        - ⭐ Stars, galaxies, and cosmology
        - 🌍 Planetary science and exoplanets
        - πŸ“Š Astrophysics concepts and theories
        - πŸ”­ Astronomical instruments and observations
        
        Just type your question below and let's embark on a cosmic journey together!
        """
    )
    
    chatbot = gr.Chatbot(
        label="Chat with AstroSage",
        bubble_full_width=False,
        show_label=True,
        height=450,
        type="messages"
    )
    
    with gr.Row():
        msg = gr.Textbox(
            label="Type your message here",
            placeholder="Ask me anything about space and astronomy...",
            scale=9
        )
        clear = gr.Button("Clear Chat", scale=1)
    
    # Example questions for quick start
    gr.Examples(
        examples=[
            "What is a black hole and how does it form?",
            "Can you explain the life cycle of a star?",
            "What are exoplanets and how do we detect them?",
            "Tell me about the James Webb Space Telescope.",
            "What is dark matter and why is it important?"
        ],
        inputs=msg,
        label="Example Questions"
    )
    
    # Set up the message chain with streaming
    msg.submit(
        user,
        [msg, chatbot],
        [msg, chatbot],
        queue=False
    ).then(
        bot,
        chatbot,
        chatbot
    )
    
    # Clear button functionality
    clear.click(lambda: None, None, chatbot, queue=False)
    
    # Initial greeting
    demo.load(initial_greeting, None, chatbot, queue=False)

# Launch the app
if __name__ == "__main__":
    demo.launch()