Spaces:
Running
on
Zero
Running
on
Zero
File size: 14,424 Bytes
ac24bf9 a338ea6 ac24bf9 a338ea6 ac24bf9 a338ea6 ac24bf9 a338ea6 ac24bf9 a338ea6 ac24bf9 a338ea6 ac24bf9 a338ea6 ac24bf9 a338ea6 ac24bf9 5e3eb6a ac24bf9 5e3eb6a ac24bf9 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 |
import gradio as gr
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
import spaces
import re
# Model configuration
model_name = "HelpingAI/Dhanishtha-2.0-preview"
# Global variables for model and tokenizer
model = None
tokenizer = None
def load_model():
"""Load the model and tokenizer"""
global model, tokenizer
print("Loading tokenizer...")
tokenizer = AutoTokenizer.from_pretrained(model_name)
# Ensure pad token is set
if tokenizer.pad_token is None:
tokenizer.pad_token = tokenizer.eos_token
print("Loading model...")
model = AutoModelForCausalLM.from_pretrained(
model_name,
torch_dtype="auto",
device_map="auto",
trust_remote_code=True
)
print("Model loaded successfully!")
def format_thinking_text(text):
"""Format text to properly display <think> tags in Gradio with blue border styling like HelpingAI"""
if not text:
return text
# More sophisticated formatting for thinking blocks with blue styling
formatted_text = text
# Handle thinking blocks with proper HTML-like styling for Gradio
thinking_pattern = r'<think>(.*?)</think>'
def replace_thinking_block(match):
thinking_content = match.group(1).strip()
# Use HTML div with inline CSS for blue border styling like HelpingAI
return f'''
<div style="border-left: 4px solid #4a90e2; background: linear-gradient(135deg, #f0f8ff 0%, #e6f3ff 100%); padding: 16px 20px; margin: 16px 0; border-radius: 12px; font-family: 'Segoe UI', sans-serif; box-shadow: 0 2px 8px rgba(74, 144, 226, 0.15); border: 1px solid rgba(74, 144, 226, 0.2);">
<div style="color: #4a90e2; font-weight: 600; margin-bottom: 10px; display: flex; align-items: center; font-size: 14px;">
<span style="margin-right: 8px;">π§ </span> Think
</div>
<div style="color: #2c3e50; line-height: 1.6; font-size: 14px;">
{thinking_content}
</div>
</div>
'''
formatted_text = re.sub(thinking_pattern, replace_thinking_block, formatted_text, flags=re.DOTALL)
# Clean up any remaining raw tags that might not have been caught
formatted_text = re.sub(r'</?think>', '', formatted_text)
return formatted_text.strip()
@spaces.GPU()
def generate_response(message, history, max_tokens, temperature, top_p):
"""Generate streaming response without threading"""
global model, tokenizer
if model is None or tokenizer is None:
yield "Model is still loading. Please wait..."
return
# Prepare conversation history
messages = []
for user_msg, assistant_msg in history:
messages.append({"role": "user", "content": user_msg})
if assistant_msg:
messages.append({"role": "assistant", "content": assistant_msg})
# Add current message
messages.append({"role": "user", "content": message})
# Apply chat template
text = tokenizer.apply_chat_template(
messages,
tokenize=False,
add_generation_prompt=True
)
# Tokenize input
model_inputs = tokenizer([text], return_tensors="pt").to(model.device)
try:
with torch.no_grad():
# Use transformers streaming with custom approach
generated_text = ""
current_input_ids = model_inputs["input_ids"]
current_attention_mask = model_inputs["attention_mask"]
for _ in range(max_tokens):
# Generate next token
outputs = model(
input_ids=current_input_ids,
attention_mask=current_attention_mask,
use_cache=True
)
# Get logits for the last token
logits = outputs.logits[0, -1, :]
# Apply temperature
if temperature != 1.0:
logits = logits / temperature
# Apply top-p sampling
if top_p < 1.0:
sorted_logits, sorted_indices = torch.sort(logits, descending=True)
cumulative_probs = torch.cumsum(torch.softmax(sorted_logits, dim=-1), dim=-1)
sorted_indices_to_remove = cumulative_probs > top_p
sorted_indices_to_remove[1:] = sorted_indices_to_remove[:-1].clone()
sorted_indices_to_remove[0] = 0
indices_to_remove = sorted_indices[sorted_indices_to_remove]
logits[indices_to_remove] = float('-inf')
# Sample next token
probs = torch.softmax(logits, dim=-1)
next_token = torch.multinomial(probs, num_samples=1)
# Check for EOS token
if next_token.item() == tokenizer.eos_token_id:
break
# Decode the new token (preserve special tokens like <think>)
new_token_text = tokenizer.decode(next_token, skip_special_tokens=False)
generated_text += new_token_text
# Format and yield the current text
formatted_text = format_thinking_text(generated_text)
yield formatted_text
# Update inputs for next iteration
current_input_ids = torch.cat([current_input_ids, next_token.unsqueeze(0)], dim=-1)
current_attention_mask = torch.cat([current_attention_mask, torch.ones((1, 1), device=model.device)], dim=-1)
except Exception as e:
yield f"Error generating response: {str(e)}"
return
# Final yield with complete formatted text
final_text = format_thinking_text(generated_text) if generated_text else "No response generated."
yield final_text
def chat_interface(message, history, max_tokens, temperature, top_p):
"""Main chat interface with improved streaming"""
if not message.strip():
return history, ""
# Add user message to history
history.append([message, ""])
# Generate response with streaming
for partial_response in generate_response(message, history[:-1], max_tokens, temperature, top_p):
history[-1][1] = partial_response
yield history, ""
return history, ""
# Load model on startup
print("Initializing model...")
load_model()
# Custom CSS for better styling and thinking blocks
custom_css = """
/* Main chatbot styling */
.chatbot {
font-size: 14px;
font-family: 'Segoe UI', Tahoma, Geneva, Verdana, sans-serif;
}
/* Enhanced thinking block styling - now handled via inline HTML */
.thinking-block {
background: linear-gradient(135deg, #f0f8ff 0%, #e6f3ff 100%);
border-left: 4px solid #4a90e2;
border-radius: 8px;
padding: 12px 16px;
margin: 12px 0;
font-family: 'Segoe UI', sans-serif;
box-shadow: 0 2px 4px rgba(0,0,0,0.1);
position: relative;
}
/* Support for HTML content in chatbot */
.chatbot .message {
overflow: visible;
}
.chatbot .message div {
max-width: none;
}
/* Message styling */
.message {
padding: 10px 14px;
margin: 6px 0;
border-radius: 12px;
line-height: 1.5;
}
.user-message {
background: linear-gradient(135deg, #e3f2fd 0%, #bbdefb 100%);
margin-left: 15%;
border-bottom-right-radius: 4px;
}
.assistant-message {
background: linear-gradient(135deg, #f5f5f5 0%, #eeeeee 100%);
margin-right: 15%;
border-bottom-left-radius: 4px;
}
/* Code block styling */
pre {
background-color: #f8f9fa;
border: 1px solid #e9ecef;
border-radius: 6px;
padding: 12px;
overflow-x: auto;
font-family: 'Consolas', 'Monaco', 'Courier New', monospace;
font-size: 13px;
line-height: 1.4;
}
/* Button styling */
.gradio-button {
border-radius: 8px;
font-weight: 500;
transition: all 0.2s ease;
}
.gradio-button:hover {
transform: translateY(-1px);
box-shadow: 0 4px 8px rgba(0,0,0,0.15);
}
/* Input styling */
.gradio-textbox {
border-radius: 8px;
border: 2px solid #e0e0e0;
transition: border-color 0.2s ease;
}
.gradio-textbox:focus {
border-color: #4a90e2;
box-shadow: 0 0 0 3px rgba(74, 144, 226, 0.1);
}
/* Slider styling */
.gradio-slider {
margin: 8px 0;
}
/* Examples styling */
.gradio-examples {
margin-top: 16px;
}
.gradio-examples .gradio-button {
background: linear-gradient(135deg, #f8f9fa 0%, #e9ecef 100%);
border: 1px solid #dee2e6;
color: #495057;
font-size: 13px;
padding: 8px 12px;
}
.gradio-examples .gradio-button:hover {
background: linear-gradient(135deg, #e9ecef 0%, #dee2e6 100%);
color: #212529;
}
"""
# Create Gradio interface
with gr.Blocks(
title="π€ Dhanishtha-2.0-preview Chat",
theme=gr.themes.Soft(),
css=custom_css
) as demo:
gr.Markdown(
"""
# π€ Dhanishtha-2.0-preview Chat
Chat with the **HelpingAI/Dhanishtha-2.0-preview** model - The world's first LLM designed to think between responses!
### β¨ Key Features:
- π§ **Multi-step Reasoning**: Unlike other LLMs that think once, Dhanishtha can think, rethink, self-evaluate, and refine using multiple `<think>` blocks
- π **Iterative Thinking**: Watch the model's thought process unfold in real-time
- π‘ **Enhanced Problem Solving**: Better reasoning capabilities through structured thinking
**Note**: The `<think>` blocks show the model's internal reasoning process and will be displayed in a formatted way below.
"""
)
with gr.Row():
with gr.Column(scale=4):
chatbot = gr.Chatbot(
[],
elem_id="chatbot",
bubble_full_width=False,
height=600,
show_copy_button=True,
show_share_button=True,
avatar_images=("π€", "π€"),
render_markdown=True,
sanitize_html=False, # Allow HTML for thinking blocks
latex_delimiters=[
{"left": "$$", "right": "$$", "display": True},
{"left": "$", "right": "$", "display": False}
]
)
with gr.Row():
msg = gr.Textbox(
container=False,
placeholder="Ask me anything! I'll show you my thinking process...",
label="Message",
autofocus=True,
scale=8,
lines=1,
max_lines=5
)
send_btn = gr.Button("π Send", variant="primary", scale=1, size="lg")
with gr.Column(scale=1, min_width=300):
gr.Markdown("### βοΈ Generation Parameters")
max_tokens = gr.Slider(
minimum=50,
maximum=8192,
value=2048,
step=50,
label="π― Max Tokens",
info="Maximum number of tokens to generate"
)
temperature = gr.Slider(
minimum=0.1,
maximum=2.0,
value=0.7,
step=0.1,
label="π‘οΈ Temperature",
info="Higher = more creative, Lower = more focused"
)
top_p = gr.Slider(
minimum=0.1,
maximum=1.0,
value=0.9,
step=0.05,
label="π² Top-p",
info="Nucleus sampling threshold"
)
with gr.Row():
clear_btn = gr.Button("ποΈ Clear Chat", variant="secondary", scale=1)
stop_btn = gr.Button("βΉοΈ Stop", variant="stop", scale=1)
gr.Markdown("### π Model Info")
gr.Markdown(
"""
**Model**: HelpingAI/Dhanishtha-2.0-preview
**Type**: Reasoning LLM with thinking blocks
**Features**: Multi-step reasoning, self-evaluation
"""
)
# Event handlers
def clear_chat():
"""Clear the chat history"""
return [], ""
# Message submission events
msg.submit(
chat_interface,
inputs=[msg, chatbot, max_tokens, temperature, top_p],
outputs=[chatbot, msg],
concurrency_limit=1,
show_progress="minimal"
)
send_btn.click(
chat_interface,
inputs=[msg, chatbot, max_tokens, temperature, top_p],
outputs=[chatbot, msg],
concurrency_limit=1,
show_progress="minimal"
)
# Clear chat event
clear_btn.click(
clear_chat,
outputs=[chatbot, msg],
show_progress=False
)
# Example prompts section
with gr.Row():
gr.Examples(
examples=[
["Hello! Can you introduce yourself and show me how you think?"],
["Solve this step by step: What is 15% of 240?"],
["Explain quantum entanglement in simple terms"],
["Write a short Python function to find the factorial of a number"],
["What are the pros and cons of renewable energy?"],
["Help me understand the difference between AI and machine learning"],
["Create a haiku about artificial intelligence"],
["Explain why the sky is blue using physics principles"]
],
inputs=msg,
label="π‘ Example Prompts - Try these to see the thinking process!",
examples_per_page=4
)
# Footer with information
gr.Markdown(
"""
---
### π§ Technical Details
- **Model**: HelpingAI/Dhanishtha-2.0-preview
- **Framework**: Transformers + Gradio
- **Features**: Real-time streaming, thinking process visualization, custom sampling
- **Reasoning**: Multi-step thinking with `<think>` blocks for transparent AI reasoning
**Note**: This interface streams responses token by token and formats thinking blocks for better readability.
The model's internal reasoning process is displayed in formatted code blocks.
---
*Built with β€οΈ using Gradio and Transformers*
"""
)
if __name__ == "__main__":
demo.queue(
max_size=20,
default_concurrency_limit=1
).launch(
server_name="0.0.0.0",
server_port=7860,
share=False,
show_error=True,
quiet=False
) |