Spaces:
Running
Running
Update app.py
Browse files
app.py
CHANGED
|
@@ -5,8 +5,6 @@ import json
|
|
| 5 |
import base64
|
| 6 |
from PIL import Image
|
| 7 |
import io
|
| 8 |
-
import requests
|
| 9 |
-
from smolagents.mcp_client import MCPClient
|
| 10 |
|
| 11 |
ACCESS_TOKEN = os.getenv("HF_TOKEN")
|
| 12 |
print("Access token loaded.")
|
|
@@ -41,154 +39,9 @@ def encode_image(image_path):
|
|
| 41 |
print(f"Error encoding image: {e}")
|
| 42 |
return None
|
| 43 |
|
| 44 |
-
# Dictionary to store active MCP connections
|
| 45 |
-
mcp_connections = {}
|
| 46 |
-
|
| 47 |
-
def connect_to_mcp_server(server_url, server_name=None):
|
| 48 |
-
"""Connect to an MCP server and return available tools"""
|
| 49 |
-
if not server_url:
|
| 50 |
-
return None, "No server URL provided"
|
| 51 |
-
|
| 52 |
-
try:
|
| 53 |
-
# Create an MCP client and connect to the server
|
| 54 |
-
client = MCPClient({"url": server_url})
|
| 55 |
-
# Get available tools
|
| 56 |
-
tools = client.get_tools()
|
| 57 |
-
|
| 58 |
-
# Store the connection for later use
|
| 59 |
-
name = server_name or f"Server_{len(mcp_connections)}"
|
| 60 |
-
mcp_connections[name] = {"client": client, "tools": tools, "url": server_url}
|
| 61 |
-
|
| 62 |
-
return name, f"Successfully connected to {name} with {len(tools)} available tools"
|
| 63 |
-
except Exception as e:
|
| 64 |
-
print(f"Error connecting to MCP server: {e}")
|
| 65 |
-
return None, f"Error connecting to MCP server: {str(e)}"
|
| 66 |
-
|
| 67 |
-
def list_mcp_tools(server_name):
|
| 68 |
-
"""List available tools for a connected MCP server"""
|
| 69 |
-
if server_name not in mcp_connections:
|
| 70 |
-
return "Server not connected"
|
| 71 |
-
|
| 72 |
-
tools = mcp_connections[server_name]["tools"]
|
| 73 |
-
tool_info = []
|
| 74 |
-
for tool in tools:
|
| 75 |
-
tool_info.append(f"- {tool.name}: {tool.description}")
|
| 76 |
-
|
| 77 |
-
if not tool_info:
|
| 78 |
-
return "No tools available for this server"
|
| 79 |
-
|
| 80 |
-
return "\n".join(tool_info)
|
| 81 |
-
|
| 82 |
-
def call_mcp_tool(server_name, tool_name, **kwargs):
|
| 83 |
-
"""Call a specific tool from an MCP server"""
|
| 84 |
-
if server_name not in mcp_connections:
|
| 85 |
-
return f"Server '{server_name}' not connected"
|
| 86 |
-
|
| 87 |
-
client = mcp_connections[server_name]["client"]
|
| 88 |
-
tools = mcp_connections[server_name]["tools"]
|
| 89 |
-
|
| 90 |
-
# Find the requested tool
|
| 91 |
-
tool = next((t for t in tools if t.name == tool_name), None)
|
| 92 |
-
if not tool:
|
| 93 |
-
return f"Tool '{tool_name}' not found on server '{server_name}'"
|
| 94 |
-
|
| 95 |
-
try:
|
| 96 |
-
# Call the tool with provided arguments
|
| 97 |
-
result = client.call_tool(tool_name, kwargs)
|
| 98 |
-
return result
|
| 99 |
-
except Exception as e:
|
| 100 |
-
print(f"Error calling MCP tool: {e}")
|
| 101 |
-
return f"Error calling MCP tool: {str(e)}"
|
| 102 |
-
|
| 103 |
-
def analyze_message_for_tool_call(message, active_mcp_servers, client, model_to_use, system_message):
|
| 104 |
-
"""Analyze a message to determine if an MCP tool should be called"""
|
| 105 |
-
# Skip analysis if message is empty
|
| 106 |
-
if not message or not message.strip():
|
| 107 |
-
return None, None
|
| 108 |
-
|
| 109 |
-
# Get information about available tools
|
| 110 |
-
tool_info = []
|
| 111 |
-
for server_name in active_mcp_servers:
|
| 112 |
-
if server_name in mcp_connections:
|
| 113 |
-
server_tools = mcp_connections[server_name]["tools"]
|
| 114 |
-
for tool in server_tools:
|
| 115 |
-
tool_info.append({
|
| 116 |
-
"server_name": server_name,
|
| 117 |
-
"tool_name": tool.name,
|
| 118 |
-
"description": tool.description
|
| 119 |
-
})
|
| 120 |
-
|
| 121 |
-
if not tool_info:
|
| 122 |
-
return None, None
|
| 123 |
-
|
| 124 |
-
# Create a structured query for the LLM to analyze if a tool call is needed
|
| 125 |
-
tools_desc = []
|
| 126 |
-
for info in tool_info:
|
| 127 |
-
tools_desc.append(f"{info['server_name']}.{info['tool_name']}: {info['description']}")
|
| 128 |
-
|
| 129 |
-
tools_string = "\n".join(tools_desc)
|
| 130 |
-
|
| 131 |
-
analysis_system_prompt = f"""You are an assistant that helps determine if a user message requires using an external tool.
|
| 132 |
-
Available tools:
|
| 133 |
-
{tools_string}
|
| 134 |
-
|
| 135 |
-
Your job is to:
|
| 136 |
-
1. Analyze the user's message
|
| 137 |
-
2. Determine if they're asking to use one of the tools
|
| 138 |
-
3. If yes, respond with a JSON object with the server_name, tool_name, and parameters
|
| 139 |
-
4. If no, respond with "NO_TOOL_NEEDED"
|
| 140 |
-
|
| 141 |
-
Example 1:
|
| 142 |
-
User: "Please turn this text into speech: Hello world"
|
| 143 |
-
Response: {{"server_name": "kokoroTTS", "tool_name": "text_to_audio", "parameters": {{"text": "Hello world", "speed": 1.0}}}}
|
| 144 |
-
|
| 145 |
-
Example 2:
|
| 146 |
-
User: "What is the capital of France?"
|
| 147 |
-
Response: NO_TOOL_NEEDED"""
|
| 148 |
-
|
| 149 |
-
try:
|
| 150 |
-
# Call the LLM to analyze the message
|
| 151 |
-
response = client.chat_completion(
|
| 152 |
-
model=model_to_use,
|
| 153 |
-
messages=[
|
| 154 |
-
{"role": "system", "content": analysis_system_prompt},
|
| 155 |
-
{"role": "user", "content": message}
|
| 156 |
-
],
|
| 157 |
-
temperature=0.2, # Low temperature for more deterministic responses
|
| 158 |
-
max_tokens=300
|
| 159 |
-
)
|
| 160 |
-
|
| 161 |
-
analysis = response.choices[0].message.content
|
| 162 |
-
print(f"Tool analysis: {analysis}")
|
| 163 |
-
|
| 164 |
-
if "NO_TOOL_NEEDED" in analysis:
|
| 165 |
-
return None, None
|
| 166 |
-
|
| 167 |
-
# Try to extract JSON from the response
|
| 168 |
-
json_start = analysis.find("{")
|
| 169 |
-
json_end = analysis.rfind("}") + 1
|
| 170 |
-
|
| 171 |
-
if json_start < 0 or json_end <= 0:
|
| 172 |
-
return None, None
|
| 173 |
-
|
| 174 |
-
json_str = analysis[json_start:json_end]
|
| 175 |
-
try:
|
| 176 |
-
tool_call = json.loads(json_str)
|
| 177 |
-
return tool_call.get("server_name"), {
|
| 178 |
-
"tool_name": tool_call.get("tool_name"),
|
| 179 |
-
"parameters": tool_call.get("parameters", {})
|
| 180 |
-
}
|
| 181 |
-
except json.JSONDecodeError:
|
| 182 |
-
print(f"Failed to parse tool call JSON: {json_str}")
|
| 183 |
-
return None, None
|
| 184 |
-
|
| 185 |
-
except Exception as e:
|
| 186 |
-
print(f"Error analyzing message for tool calls: {str(e)}")
|
| 187 |
-
return None, None
|
| 188 |
-
|
| 189 |
def respond(
|
| 190 |
message,
|
| 191 |
-
image_files,
|
| 192 |
history: list[tuple[str, str]],
|
| 193 |
system_message,
|
| 194 |
max_tokens,
|
|
@@ -200,10 +53,7 @@ def respond(
|
|
| 200 |
custom_api_key,
|
| 201 |
custom_model,
|
| 202 |
model_search_term,
|
| 203 |
-
selected_model
|
| 204 |
-
mcp_enabled=False,
|
| 205 |
-
active_mcp_servers=None,
|
| 206 |
-
mcp_interaction_mode="Natural Language"
|
| 207 |
):
|
| 208 |
print(f"Received message: {message}")
|
| 209 |
print(f"Received {len(image_files) if image_files else 0} images")
|
|
@@ -216,9 +66,6 @@ def respond(
|
|
| 216 |
print(f"Selected model (custom_model): {custom_model}")
|
| 217 |
print(f"Model search term: {model_search_term}")
|
| 218 |
print(f"Selected model from radio: {selected_model}")
|
| 219 |
-
print(f"MCP enabled: {mcp_enabled}")
|
| 220 |
-
print(f"Active MCP servers: {active_mcp_servers}")
|
| 221 |
-
print(f"MCP interaction mode: {mcp_interaction_mode}")
|
| 222 |
|
| 223 |
# Determine which token to use
|
| 224 |
token_to_use = custom_api_key if custom_api_key.strip() != "" else ACCESS_TOKEN
|
|
@@ -235,58 +82,6 @@ def respond(
|
|
| 235 |
# Convert seed to None if -1 (meaning random)
|
| 236 |
if seed == -1:
|
| 237 |
seed = None
|
| 238 |
-
|
| 239 |
-
# Determine which model to use
|
| 240 |
-
model_to_use = custom_model.strip() if custom_model.strip() != "" else selected_model
|
| 241 |
-
print(f"Model selected for inference: {model_to_use}")
|
| 242 |
-
|
| 243 |
-
# Process MCP commands in command mode
|
| 244 |
-
if mcp_enabled and message:
|
| 245 |
-
if message.startswith("/mcp"): # Always handle explicit commands
|
| 246 |
-
# Handle MCP command
|
| 247 |
-
command_parts = message.split(" ", 3)
|
| 248 |
-
if len(command_parts) < 3:
|
| 249 |
-
return "Invalid MCP command. Format: /mcp <server_name> <tool_name> [arguments]"
|
| 250 |
-
|
| 251 |
-
_, server_name, tool_name = command_parts[:3]
|
| 252 |
-
args_json = "{}" if len(command_parts) < 4 else command_parts[3]
|
| 253 |
-
|
| 254 |
-
try:
|
| 255 |
-
args_dict = json.loads(args_json)
|
| 256 |
-
result = call_mcp_tool(server_name, tool_name, **args_dict)
|
| 257 |
-
if isinstance(result, dict):
|
| 258 |
-
return json.dumps(result, indent=2)
|
| 259 |
-
return str(result)
|
| 260 |
-
except json.JSONDecodeError:
|
| 261 |
-
return f"Invalid JSON arguments: {args_json}"
|
| 262 |
-
except Exception as e:
|
| 263 |
-
return f"Error executing MCP command: {str(e)}"
|
| 264 |
-
elif mcp_interaction_mode == "Natural Language" and active_mcp_servers:
|
| 265 |
-
# Use natural language processing to detect tool calls
|
| 266 |
-
server_name, tool_info = analyze_message_for_tool_call(
|
| 267 |
-
message,
|
| 268 |
-
active_mcp_servers,
|
| 269 |
-
client,
|
| 270 |
-
model_to_use,
|
| 271 |
-
system_message
|
| 272 |
-
)
|
| 273 |
-
|
| 274 |
-
if server_name and tool_info:
|
| 275 |
-
try:
|
| 276 |
-
# Call the detected tool
|
| 277 |
-
print(f"Calling tool via natural language: {server_name}.{tool_info['tool_name']} with parameters: {tool_info['parameters']}")
|
| 278 |
-
result = call_mcp_tool(server_name, tool_info['tool_name'], **tool_info['parameters'])
|
| 279 |
-
|
| 280 |
-
# Format the response to include what was done
|
| 281 |
-
if isinstance(result, dict):
|
| 282 |
-
result_str = json.dumps(result, indent=2)
|
| 283 |
-
else:
|
| 284 |
-
result_str = str(result)
|
| 285 |
-
|
| 286 |
-
return f"I used the {tool_info['tool_name']} tool from {server_name} with your request.\n\nResult:\n{result_str}"
|
| 287 |
-
except Exception as e:
|
| 288 |
-
print(f"Error executing MCP tool via natural language: {str(e)}")
|
| 289 |
-
# Continue with normal response if tool call fails
|
| 290 |
|
| 291 |
# Create multimodal content if images are present
|
| 292 |
if image_files and len(image_files) > 0:
|
|
@@ -319,25 +114,8 @@ def respond(
|
|
| 319 |
# Text-only message
|
| 320 |
user_content = message
|
| 321 |
|
| 322 |
-
# Add information about available MCP tools to the system message if MCP is enabled
|
| 323 |
-
augmented_system_message = system_message
|
| 324 |
-
if mcp_enabled and active_mcp_servers:
|
| 325 |
-
tool_info = []
|
| 326 |
-
for server_name in active_mcp_servers:
|
| 327 |
-
if server_name in mcp_connections:
|
| 328 |
-
server_tools = list_mcp_tools(server_name).split("\n")
|
| 329 |
-
tool_info.extend([f"{server_name}: {tool}" for tool in server_tools])
|
| 330 |
-
|
| 331 |
-
if tool_info:
|
| 332 |
-
mcp_tools_description = "\n".join(tool_info)
|
| 333 |
-
|
| 334 |
-
if mcp_interaction_mode == "Command Mode":
|
| 335 |
-
augmented_system_message += f"\n\nYou have access to the following MCP tools:\n{mcp_tools_description}\n\nTo use these tools, the user can type a command in the format: /mcp <server_name> <tool_name> <arguments_json>"
|
| 336 |
-
else:
|
| 337 |
-
augmented_system_message += f"\n\nYou have access to the following MCP tools:\n{mcp_tools_description}\n\nThe user can use these tools by describing what they want in natural language, and the system will automatically detect when to use a tool based on their request."
|
| 338 |
-
|
| 339 |
# Prepare messages in the format expected by the API
|
| 340 |
-
messages = [{"role": "system", "content":
|
| 341 |
print("Initial messages array constructed.")
|
| 342 |
|
| 343 |
# Add conversation history to the context
|
|
@@ -433,13 +211,19 @@ def respond(
|
|
| 433 |
|
| 434 |
print("Completed response generation.")
|
| 435 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 436 |
# GRADIO UI
|
| 437 |
with gr.Blocks(theme="Nymbo/Nymbo_Theme") as demo:
|
| 438 |
# Create the chatbot component
|
| 439 |
chatbot = gr.Chatbot(
|
| 440 |
height=600,
|
| 441 |
show_copy_button=True,
|
| 442 |
-
placeholder="Select a model and begin chatting. Now supports multiple inference providers
|
| 443 |
layout="panel"
|
| 444 |
)
|
| 445 |
print("Chatbot interface created.")
|
|
@@ -455,6 +239,8 @@ with gr.Blocks(theme="Nymbo/Nymbo_Theme") as demo:
|
|
| 455 |
sources=["upload"]
|
| 456 |
)
|
| 457 |
|
|
|
|
|
|
|
| 458 |
# Create accordion for settings
|
| 459 |
with gr.Accordion("Settings", open=False):
|
| 460 |
# System message
|
|
@@ -588,69 +374,6 @@ with gr.Blocks(theme="Nymbo/Nymbo_Theme") as demo:
|
|
| 588 |
|
| 589 |
gr.Markdown("[View all Text-to-Text models](https://huggingface.co/models?inference_provider=all&pipeline_tag=text-generation&sort=trending) | [View all multimodal models](https://huggingface.co/models?inference_provider=all&pipeline_tag=image-text-to-text&sort=trending)")
|
| 590 |
|
| 591 |
-
# Create accordion for MCP settings
|
| 592 |
-
with gr.Accordion("MCP Settings", open=False):
|
| 593 |
-
mcp_enabled_checkbox = gr.Checkbox(
|
| 594 |
-
label="Enable MCP Support",
|
| 595 |
-
value=False,
|
| 596 |
-
info="Enable Model Context Protocol support to connect to external tools and services"
|
| 597 |
-
)
|
| 598 |
-
|
| 599 |
-
with gr.Row():
|
| 600 |
-
mcp_server_url = gr.Textbox(
|
| 601 |
-
label="MCP Server URL",
|
| 602 |
-
placeholder="https://example-mcp-server.hf.space/gradio_api/mcp/sse",
|
| 603 |
-
info="URL of the MCP server to connect to"
|
| 604 |
-
)
|
| 605 |
-
|
| 606 |
-
mcp_server_name = gr.Textbox(
|
| 607 |
-
label="Server Name",
|
| 608 |
-
placeholder="Optional name for this server",
|
| 609 |
-
info="A friendly name to identify this server"
|
| 610 |
-
)
|
| 611 |
-
|
| 612 |
-
mcp_connect_button = gr.Button("Connect to MCP Server")
|
| 613 |
-
|
| 614 |
-
mcp_status = gr.Textbox(
|
| 615 |
-
label="MCP Connection Status",
|
| 616 |
-
placeholder="No MCP servers connected",
|
| 617 |
-
interactive=False
|
| 618 |
-
)
|
| 619 |
-
|
| 620 |
-
active_mcp_servers = gr.Dropdown(
|
| 621 |
-
label="Active MCP Servers",
|
| 622 |
-
choices=[],
|
| 623 |
-
multiselect=True,
|
| 624 |
-
info="Select which MCP servers to use in chat"
|
| 625 |
-
)
|
| 626 |
-
|
| 627 |
-
mcp_mode = gr.Radio(
|
| 628 |
-
label="MCP Interaction Mode",
|
| 629 |
-
choices=["Natural Language", "Command Mode"],
|
| 630 |
-
value="Natural Language",
|
| 631 |
-
info="Choose how to interact with MCP tools"
|
| 632 |
-
)
|
| 633 |
-
|
| 634 |
-
gr.Markdown("""
|
| 635 |
-
### MCP Interaction Modes
|
| 636 |
-
|
| 637 |
-
**Natural Language Mode**: Simply describe what you want in plain English. Examples:
|
| 638 |
-
```
|
| 639 |
-
Please convert the text "Hello world" to speech
|
| 640 |
-
Can you read this text aloud: "Welcome to MCP integration"
|
| 641 |
-
```
|
| 642 |
-
|
| 643 |
-
**Command Mode**: Use structured commands (for advanced users)
|
| 644 |
-
```
|
| 645 |
-
/mcp <server_name> <tool_name> {"param1": "value1", "param2": "value2"}
|
| 646 |
-
```
|
| 647 |
-
|
| 648 |
-
Example:
|
| 649 |
-
```
|
| 650 |
-
/mcp kokoroTTS text_to_audio {"text": "Hello world", "speed": 1.0}
|
| 651 |
-
```
|
| 652 |
-
""")
|
| 653 |
-
|
| 654 |
# Chat history state
|
| 655 |
chat_history = gr.State([])
|
| 656 |
|
|
@@ -666,16 +389,6 @@ with gr.Blocks(theme="Nymbo/Nymbo_Theme") as demo:
|
|
| 666 |
print(f"Featured model selected: {selected}")
|
| 667 |
return selected
|
| 668 |
|
| 669 |
-
# Function to connect to MCP server
|
| 670 |
-
def connect_mcp_server(url, name):
|
| 671 |
-
server_name, status = connect_to_mcp_server(url, name)
|
| 672 |
-
|
| 673 |
-
# Update the active servers dropdown
|
| 674 |
-
servers = list(mcp_connections.keys())
|
| 675 |
-
|
| 676 |
-
# Return the status message and updated server list
|
| 677 |
-
return status, gr.update(choices=servers)
|
| 678 |
-
|
| 679 |
# Function for the chat interface
|
| 680 |
def user(user_message, history):
|
| 681 |
# Debug logging for troubleshooting
|
|
@@ -721,7 +434,7 @@ with gr.Blocks(theme="Nymbo/Nymbo_Theme") as demo:
|
|
| 721 |
return history
|
| 722 |
|
| 723 |
# Define bot response function
|
| 724 |
-
def bot(history, system_msg, max_tokens, temperature, top_p, freq_penalty, seed, provider, api_key, custom_model, search_term, selected_model
|
| 725 |
# Check if history is valid
|
| 726 |
if not history or len(history) == 0:
|
| 727 |
print("No history to process")
|
|
@@ -772,10 +485,7 @@ with gr.Blocks(theme="Nymbo/Nymbo_Theme") as demo:
|
|
| 772 |
api_key,
|
| 773 |
custom_model,
|
| 774 |
search_term,
|
| 775 |
-
selected_model
|
| 776 |
-
mcp_enabled,
|
| 777 |
-
selected_servers,
|
| 778 |
-
mcp_interaction_mode
|
| 779 |
):
|
| 780 |
history[-1][1] = response
|
| 781 |
yield history
|
|
@@ -795,21 +505,12 @@ with gr.Blocks(theme="Nymbo/Nymbo_Theme") as demo:
|
|
| 795 |
api_key,
|
| 796 |
custom_model,
|
| 797 |
search_term,
|
| 798 |
-
selected_model
|
| 799 |
-
mcp_enabled,
|
| 800 |
-
selected_servers,
|
| 801 |
-
mcp_interaction_mode
|
| 802 |
):
|
| 803 |
history[-1][1] = response
|
| 804 |
yield history
|
| 805 |
|
| 806 |
-
#
|
| 807 |
-
def validate_provider(api_key, provider):
|
| 808 |
-
if not api_key.strip() and provider != "hf-inference":
|
| 809 |
-
return gr.update(value="hf-inference")
|
| 810 |
-
return gr.update(value=provider)
|
| 811 |
-
|
| 812 |
-
# Event handlers
|
| 813 |
msg.submit(
|
| 814 |
user,
|
| 815 |
[msg, chatbot],
|
|
@@ -819,7 +520,7 @@ with gr.Blocks(theme="Nymbo/Nymbo_Theme") as demo:
|
|
| 819 |
bot,
|
| 820 |
[chatbot, system_message_box, max_tokens_slider, temperature_slider, top_p_slider,
|
| 821 |
frequency_penalty_slider, seed_slider, provider_radio, byok_textbox, custom_model_box,
|
| 822 |
-
model_search_box, featured_model_radio
|
| 823 |
[chatbot]
|
| 824 |
).then(
|
| 825 |
lambda: {"text": "", "files": []}, # Clear inputs after submission
|
|
@@ -827,13 +528,6 @@ with gr.Blocks(theme="Nymbo/Nymbo_Theme") as demo:
|
|
| 827 |
[msg]
|
| 828 |
)
|
| 829 |
|
| 830 |
-
# Connect MCP connect button
|
| 831 |
-
mcp_connect_button.click(
|
| 832 |
-
connect_mcp_server,
|
| 833 |
-
[mcp_server_url, mcp_server_name],
|
| 834 |
-
[mcp_status, active_mcp_servers]
|
| 835 |
-
)
|
| 836 |
-
|
| 837 |
# Connect the model filter to update the radio choices
|
| 838 |
model_search_box.change(
|
| 839 |
fn=filter_models,
|
|
@@ -870,4 +564,4 @@ print("Gradio interface initialized.")
|
|
| 870 |
|
| 871 |
if __name__ == "__main__":
|
| 872 |
print("Launching the demo application.")
|
| 873 |
-
demo.launch(show_api=True
|
|
|
|
| 5 |
import base64
|
| 6 |
from PIL import Image
|
| 7 |
import io
|
|
|
|
|
|
|
| 8 |
|
| 9 |
ACCESS_TOKEN = os.getenv("HF_TOKEN")
|
| 10 |
print("Access token loaded.")
|
|
|
|
| 39 |
print(f"Error encoding image: {e}")
|
| 40 |
return None
|
| 41 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 42 |
def respond(
|
| 43 |
message,
|
| 44 |
+
image_files, # Changed parameter name and structure
|
| 45 |
history: list[tuple[str, str]],
|
| 46 |
system_message,
|
| 47 |
max_tokens,
|
|
|
|
| 53 |
custom_api_key,
|
| 54 |
custom_model,
|
| 55 |
model_search_term,
|
| 56 |
+
selected_model
|
|
|
|
|
|
|
|
|
|
| 57 |
):
|
| 58 |
print(f"Received message: {message}")
|
| 59 |
print(f"Received {len(image_files) if image_files else 0} images")
|
|
|
|
| 66 |
print(f"Selected model (custom_model): {custom_model}")
|
| 67 |
print(f"Model search term: {model_search_term}")
|
| 68 |
print(f"Selected model from radio: {selected_model}")
|
|
|
|
|
|
|
|
|
|
| 69 |
|
| 70 |
# Determine which token to use
|
| 71 |
token_to_use = custom_api_key if custom_api_key.strip() != "" else ACCESS_TOKEN
|
|
|
|
| 82 |
# Convert seed to None if -1 (meaning random)
|
| 83 |
if seed == -1:
|
| 84 |
seed = None
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 85 |
|
| 86 |
# Create multimodal content if images are present
|
| 87 |
if image_files and len(image_files) > 0:
|
|
|
|
| 114 |
# Text-only message
|
| 115 |
user_content = message
|
| 116 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 117 |
# Prepare messages in the format expected by the API
|
| 118 |
+
messages = [{"role": "system", "content": system_message}]
|
| 119 |
print("Initial messages array constructed.")
|
| 120 |
|
| 121 |
# Add conversation history to the context
|
|
|
|
| 211 |
|
| 212 |
print("Completed response generation.")
|
| 213 |
|
| 214 |
+
# Function to validate provider selection based on BYOK
|
| 215 |
+
def validate_provider(api_key, provider):
|
| 216 |
+
if not api_key.strip() and provider != "hf-inference":
|
| 217 |
+
return gr.update(value="hf-inference")
|
| 218 |
+
return gr.update(value=provider)
|
| 219 |
+
|
| 220 |
# GRADIO UI
|
| 221 |
with gr.Blocks(theme="Nymbo/Nymbo_Theme") as demo:
|
| 222 |
# Create the chatbot component
|
| 223 |
chatbot = gr.Chatbot(
|
| 224 |
height=600,
|
| 225 |
show_copy_button=True,
|
| 226 |
+
placeholder="Select a model and begin chatting. Now supports multiple inference providers and multimodal inputs",
|
| 227 |
layout="panel"
|
| 228 |
)
|
| 229 |
print("Chatbot interface created.")
|
|
|
|
| 239 |
sources=["upload"]
|
| 240 |
)
|
| 241 |
|
| 242 |
+
# Note: We're removing the separate submit button since MultimodalTextbox has its own
|
| 243 |
+
|
| 244 |
# Create accordion for settings
|
| 245 |
with gr.Accordion("Settings", open=False):
|
| 246 |
# System message
|
|
|
|
| 374 |
|
| 375 |
gr.Markdown("[View all Text-to-Text models](https://huggingface.co/models?inference_provider=all&pipeline_tag=text-generation&sort=trending) | [View all multimodal models](https://huggingface.co/models?inference_provider=all&pipeline_tag=image-text-to-text&sort=trending)")
|
| 376 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 377 |
# Chat history state
|
| 378 |
chat_history = gr.State([])
|
| 379 |
|
|
|
|
| 389 |
print(f"Featured model selected: {selected}")
|
| 390 |
return selected
|
| 391 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 392 |
# Function for the chat interface
|
| 393 |
def user(user_message, history):
|
| 394 |
# Debug logging for troubleshooting
|
|
|
|
| 434 |
return history
|
| 435 |
|
| 436 |
# Define bot response function
|
| 437 |
+
def bot(history, system_msg, max_tokens, temperature, top_p, freq_penalty, seed, provider, api_key, custom_model, search_term, selected_model):
|
| 438 |
# Check if history is valid
|
| 439 |
if not history or len(history) == 0:
|
| 440 |
print("No history to process")
|
|
|
|
| 485 |
api_key,
|
| 486 |
custom_model,
|
| 487 |
search_term,
|
| 488 |
+
selected_model
|
|
|
|
|
|
|
|
|
|
| 489 |
):
|
| 490 |
history[-1][1] = response
|
| 491 |
yield history
|
|
|
|
| 505 |
api_key,
|
| 506 |
custom_model,
|
| 507 |
search_term,
|
| 508 |
+
selected_model
|
|
|
|
|
|
|
|
|
|
| 509 |
):
|
| 510 |
history[-1][1] = response
|
| 511 |
yield history
|
| 512 |
|
| 513 |
+
# Event handlers - only using the MultimodalTextbox's built-in submit functionality
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 514 |
msg.submit(
|
| 515 |
user,
|
| 516 |
[msg, chatbot],
|
|
|
|
| 520 |
bot,
|
| 521 |
[chatbot, system_message_box, max_tokens_slider, temperature_slider, top_p_slider,
|
| 522 |
frequency_penalty_slider, seed_slider, provider_radio, byok_textbox, custom_model_box,
|
| 523 |
+
model_search_box, featured_model_radio],
|
| 524 |
[chatbot]
|
| 525 |
).then(
|
| 526 |
lambda: {"text": "", "files": []}, # Clear inputs after submission
|
|
|
|
| 528 |
[msg]
|
| 529 |
)
|
| 530 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 531 |
# Connect the model filter to update the radio choices
|
| 532 |
model_search_box.change(
|
| 533 |
fn=filter_models,
|
|
|
|
| 564 |
|
| 565 |
if __name__ == "__main__":
|
| 566 |
print("Launching the demo application.")
|
| 567 |
+
demo.launch(show_api=True)
|