Spaces:
Running
Running
Update app.py
Browse files
app.py
CHANGED
@@ -6,21 +6,35 @@ import base64
|
|
6 |
from PIL import Image
|
7 |
import io
|
8 |
import requests
|
9 |
-
from
|
10 |
-
|
11 |
-
|
|
|
|
|
|
|
|
|
12 |
|
13 |
ACCESS_TOKEN = os.getenv("HF_TOKEN")
|
14 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
15 |
|
16 |
# Function to encode image to base64
|
17 |
def encode_image(image_path):
|
18 |
if not image_path:
|
19 |
-
|
20 |
return None
|
21 |
|
22 |
try:
|
23 |
-
|
24 |
|
25 |
# If it's already a PIL Image
|
26 |
if isinstance(image_path, Image.Image):
|
@@ -37,117 +51,134 @@ def encode_image(image_path):
|
|
37 |
buffered = io.BytesIO()
|
38 |
image.save(buffered, format="JPEG")
|
39 |
img_str = base64.b64encode(buffered.getvalue()).decode("utf-8")
|
40 |
-
|
41 |
return img_str
|
42 |
except Exception as e:
|
43 |
-
|
44 |
return None
|
45 |
|
46 |
-
# MCP Client
|
47 |
class MCPClient:
|
48 |
-
def __init__(self,
|
49 |
-
self.
|
50 |
-
self.
|
51 |
-
|
52 |
-
self.tools = None
|
53 |
|
54 |
-
def connect(self):
|
|
|
55 |
try:
|
56 |
-
|
57 |
-
|
58 |
-
|
59 |
-
|
60 |
-
|
61 |
-
|
62 |
-
|
63 |
-
|
64 |
-
|
65 |
-
|
66 |
-
|
67 |
-
|
68 |
except Exception as e:
|
69 |
-
|
70 |
return False
|
71 |
|
72 |
-
def
|
73 |
-
|
74 |
-
|
75 |
-
|
76 |
-
|
77 |
-
# Find the tool with the given name
|
78 |
-
tool = next((t for t in self.tools if t.name == tool_name), None)
|
79 |
-
if not tool:
|
80 |
-
print(f"Tool '{tool_name}' not found")
|
81 |
-
return None
|
82 |
|
83 |
try:
|
84 |
-
|
85 |
-
|
86 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
87 |
except Exception as e:
|
88 |
-
|
89 |
-
return
|
90 |
|
91 |
-
def
|
92 |
-
|
93 |
-
|
94 |
-
|
95 |
-
|
96 |
-
except Exception as e:
|
97 |
-
print(f"Error closing MCP client connection: {e}")
|
98 |
-
|
99 |
-
# Function to convert text to audio using Kokoro MCP server
|
100 |
-
def text_to_audio(text, speed=1.0, mcp_url=None):
|
101 |
-
"""Convert text to audio using Kokoro MCP server if available.
|
102 |
-
|
103 |
-
Args:
|
104 |
-
text (str): Text to convert to speech
|
105 |
-
speed (float): Speed multiplier for speech
|
106 |
-
mcp_url (str): URL of the Kokoro MCP server
|
107 |
|
108 |
-
|
109 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
110 |
"""
|
111 |
-
|
|
|
|
|
|
|
|
|
112 |
return None
|
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 |
return None
|
147 |
|
148 |
def respond(
|
149 |
message,
|
150 |
-
image_files,
|
151 |
history: list[tuple[str, str]],
|
152 |
system_message,
|
153 |
max_tokens,
|
@@ -160,35 +191,33 @@ def respond(
|
|
160 |
custom_model,
|
161 |
model_search_term,
|
162 |
selected_model,
|
163 |
-
mcp_server_url=None,
|
164 |
tts_enabled=False,
|
165 |
-
|
166 |
):
|
167 |
-
|
168 |
-
|
169 |
-
|
170 |
-
|
171 |
-
|
172 |
-
|
173 |
-
|
174 |
-
|
175 |
-
|
176 |
-
|
177 |
-
|
178 |
-
|
179 |
-
print(f"TTS Enabled: {tts_enabled}")
|
180 |
|
181 |
# Determine which token to use
|
182 |
token_to_use = custom_api_key if custom_api_key.strip() != "" else ACCESS_TOKEN
|
183 |
|
184 |
if custom_api_key.strip() != "":
|
185 |
-
|
186 |
else:
|
187 |
-
|
188 |
|
189 |
# Initialize the Inference Client with the provider and appropriate token
|
190 |
client = InferenceClient(token=token_to_use, provider=provider)
|
191 |
-
|
192 |
|
193 |
# Convert seed to None if -1 (meaning random)
|
194 |
if seed == -1:
|
@@ -220,14 +249,14 @@ def respond(
|
|
220 |
}
|
221 |
})
|
222 |
except Exception as e:
|
223 |
-
|
224 |
else:
|
225 |
# Text-only message
|
226 |
user_content = message
|
227 |
|
228 |
# Prepare messages in the format expected by the API
|
229 |
messages = [{"role": "system", "content": system_message}]
|
230 |
-
|
231 |
|
232 |
# Add conversation history to the context
|
233 |
for val in history:
|
@@ -256,29 +285,29 @@ def respond(
|
|
256 |
}
|
257 |
})
|
258 |
except Exception as e:
|
259 |
-
|
260 |
|
261 |
messages.append({"role": "user", "content": history_content})
|
262 |
else:
|
263 |
# Regular text message
|
264 |
messages.append({"role": "user", "content": user_part})
|
265 |
-
|
266 |
|
267 |
if assistant_part:
|
268 |
messages.append({"role": "assistant", "content": assistant_part})
|
269 |
-
|
270 |
|
271 |
# Append the latest user message
|
272 |
messages.append({"role": "user", "content": user_content})
|
273 |
-
|
274 |
|
275 |
# Determine which model to use, prioritizing custom_model if provided
|
276 |
model_to_use = custom_model.strip() if custom_model.strip() != "" else selected_model
|
277 |
-
|
278 |
|
279 |
# Start with an empty string to build the response as tokens stream in
|
280 |
response = ""
|
281 |
-
|
282 |
|
283 |
# Prepare parameters for the chat completion request
|
284 |
parameters = {
|
@@ -301,7 +330,7 @@ def respond(
|
|
301 |
**parameters
|
302 |
)
|
303 |
|
304 |
-
|
305 |
|
306 |
# Process the streaming response
|
307 |
for chunk in stream:
|
@@ -314,26 +343,28 @@ def respond(
|
|
314 |
response += token_text
|
315 |
yield response
|
316 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
317 |
print()
|
318 |
except Exception as e:
|
319 |
-
|
320 |
response += f"\nError: {str(e)}"
|
321 |
yield response
|
322 |
|
323 |
-
|
324 |
-
|
325 |
-
# If TTS is enabled and we have a valid MCP server URL, convert response to audio
|
326 |
-
if tts_enabled and mcp_server_url and response:
|
327 |
-
try:
|
328 |
-
print(f"Converting response to audio using MCP server: {mcp_server_url}")
|
329 |
-
audio_data = text_to_audio(response, tts_speed, mcp_server_url)
|
330 |
-
if audio_data:
|
331 |
-
# Here we would need to handle returning both text and audio
|
332 |
-
# This would require modifying the Gradio interface to support this
|
333 |
-
print("Successfully converted text to audio")
|
334 |
-
# For now, we'll just return the text response
|
335 |
-
except Exception as e:
|
336 |
-
print(f"Error converting text to audio: {e}")
|
337 |
|
338 |
# Function to validate provider selection based on BYOK
|
339 |
def validate_provider(api_key, provider):
|
@@ -341,21 +372,10 @@ def validate_provider(api_key, provider):
|
|
341 |
return gr.update(value="hf-inference")
|
342 |
return gr.update(value=provider)
|
343 |
|
344 |
-
# Function to
|
345 |
-
def
|
346 |
-
|
347 |
-
|
348 |
-
|
349 |
-
try:
|
350 |
-
mcp_client = MCPClient(mcp_url)
|
351 |
-
if mcp_client.connect():
|
352 |
-
tools = [tool.name for tool in mcp_client.tools]
|
353 |
-
mcp_client.close()
|
354 |
-
return f"Successfully connected to MCP server. Available tools: {', '.join(tools)}"
|
355 |
-
else:
|
356 |
-
return "Failed to connect to MCP server"
|
357 |
-
except Exception as e:
|
358 |
-
return f"Error connecting to MCP server: {str(e)}"
|
359 |
|
360 |
# GRADIO UI
|
361 |
with gr.Blocks(theme="Nymbo/Nymbo_Theme") as demo:
|
@@ -363,10 +383,10 @@ with gr.Blocks(theme="Nymbo/Nymbo_Theme") as demo:
|
|
363 |
chatbot = gr.Chatbot(
|
364 |
height=600,
|
365 |
show_copy_button=True,
|
366 |
-
placeholder="Select a model and begin chatting. Now supports multiple inference providers and
|
367 |
layout="panel"
|
368 |
)
|
369 |
-
|
370 |
|
371 |
# Multimodal textbox for messages (combines text and file uploads)
|
372 |
msg = gr.MultimodalTextbox(
|
@@ -511,94 +531,76 @@ with gr.Blocks(theme="Nymbo/Nymbo_Theme") as demo:
|
|
511 |
)
|
512 |
|
513 |
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)")
|
514 |
-
|
515 |
-
# New Accordion for MCP Settings
|
516 |
-
with gr.Accordion("MCP Server Settings", open=False):
|
517 |
-
mcp_server_url = gr.Textbox(
|
518 |
-
value="",
|
519 |
-
label="MCP Server URL",
|
520 |
-
info="Enter the URL of an MCP server to connect to (e.g., https://example-kokoro-mcp.hf.space/gradio_api/mcp/sse)",
|
521 |
-
placeholder="https://fdaudens-kokoro-mcp.hf.space/gradio_api/mcp/sse"
|
522 |
-
)
|
523 |
|
524 |
-
|
525 |
-
|
526 |
-
|
527 |
-
|
528 |
-
|
529 |
-
|
530 |
-
|
531 |
-
|
532 |
-
|
533 |
-
|
534 |
-
|
535 |
-
|
536 |
-
|
537 |
-
|
538 |
-
|
539 |
-
|
540 |
-
|
541 |
-
|
542 |
-
|
543 |
-
|
544 |
-
|
545 |
-
|
546 |
-
|
547 |
-
|
548 |
-
|
549 |
-
|
550 |
-
|
551 |
-
|
552 |
-
|
553 |
-
|
554 |
-
|
555 |
-
|
556 |
-
|
557 |
-
Note: TTS functionality requires an active connection to a Kokoro MCP server.
|
558 |
-
""")
|
559 |
|
560 |
# Chat history state
|
561 |
chat_history = gr.State([])
|
562 |
|
563 |
-
# Connect the test connection button
|
564 |
-
test_connection_btn.click(
|
565 |
-
fn=test_mcp_connection,
|
566 |
-
inputs=[mcp_server_url],
|
567 |
-
outputs=[connection_status]
|
568 |
-
)
|
569 |
-
|
570 |
# Function to filter models
|
571 |
def filter_models(search_term):
|
572 |
-
|
573 |
filtered = [m for m in models_list if search_term.lower() in m.lower()]
|
574 |
-
|
575 |
return gr.update(choices=filtered)
|
576 |
|
577 |
# Function to set custom model from radio
|
578 |
def set_custom_model_from_radio(selected):
|
579 |
-
|
580 |
return selected
|
581 |
|
582 |
# Function for the chat interface
|
583 |
def user(user_message, history):
|
584 |
# Debug logging for troubleshooting
|
585 |
-
|
586 |
|
587 |
# Skip if message is empty (no text and no files)
|
588 |
if not user_message or (not user_message.get("text") and not user_message.get("files")):
|
589 |
-
|
590 |
return history
|
591 |
|
592 |
# Prepare multimodal message format
|
593 |
text_content = user_message.get("text", "").strip()
|
594 |
files = user_message.get("files", [])
|
595 |
|
596 |
-
|
597 |
-
|
598 |
|
599 |
# If both text and files are empty, skip
|
600 |
if not text_content and not files:
|
601 |
-
|
602 |
return history
|
603 |
|
604 |
# Add message with images to history
|
@@ -606,33 +608,33 @@ with gr.Blocks(theme="Nymbo/Nymbo_Theme") as demo:
|
|
606 |
# Add text message first if it exists
|
607 |
if text_content:
|
608 |
# Add a separate text message
|
609 |
-
|
610 |
history.append([text_content, None])
|
611 |
|
612 |
# Then add each image file separately
|
613 |
for file_path in files:
|
614 |
if file_path and isinstance(file_path, str):
|
615 |
-
|
616 |
# Add image as a separate message with no text
|
617 |
history.append([f"", None])
|
618 |
|
619 |
return history
|
620 |
else:
|
621 |
# For text-only messages
|
622 |
-
|
623 |
history.append([text_content, None])
|
624 |
return history
|
625 |
|
626 |
# Define bot response function
|
627 |
-
def bot(history, system_msg, max_tokens, temperature, top_p, freq_penalty, seed, provider, api_key, custom_model, search_term, selected_model,
|
628 |
# Check if history is valid
|
629 |
if not history or len(history) == 0:
|
630 |
-
|
631 |
return history
|
632 |
|
633 |
# Get the most recent message and detect if it's an image
|
634 |
user_message = history[-1][0]
|
635 |
-
|
636 |
|
637 |
is_image = False
|
638 |
image_path = None
|
@@ -643,7 +645,7 @@ with gr.Blocks(theme="Nymbo/Nymbo_Theme") as demo:
|
|
643 |
is_image = True
|
644 |
# Extract image path from markdown format 
|
645 |
image_path = user_message.replace(".replace(")", "")
|
646 |
-
|
647 |
text_content = "" # No text for image-only messages
|
648 |
|
649 |
# Look back for text context if this is an image
|
@@ -653,7 +655,7 @@ with gr.Blocks(theme="Nymbo/Nymbo_Theme") as demo:
|
|
653 |
prev_message = history[-2][0]
|
654 |
if isinstance(prev_message, str) and not prev_message.startswith(":
|
655 |
text_context = prev_message
|
656 |
-
|
657 |
|
658 |
# Process message through respond function
|
659 |
history[-1][1] = ""
|
@@ -676,9 +678,8 @@ with gr.Blocks(theme="Nymbo/Nymbo_Theme") as demo:
|
|
676 |
custom_model,
|
677 |
search_term,
|
678 |
selected_model,
|
679 |
-
|
680 |
-
|
681 |
-
tts_spd
|
682 |
):
|
683 |
history[-1][1] = response
|
684 |
yield history
|
@@ -699,9 +700,8 @@ with gr.Blocks(theme="Nymbo/Nymbo_Theme") as demo:
|
|
699 |
custom_model,
|
700 |
search_term,
|
701 |
selected_model,
|
702 |
-
|
703 |
-
|
704 |
-
tts_spd
|
705 |
):
|
706 |
history[-1][1] = response
|
707 |
yield history
|
@@ -716,7 +716,7 @@ with gr.Blocks(theme="Nymbo/Nymbo_Theme") as demo:
|
|
716 |
bot,
|
717 |
[chatbot, system_message_box, max_tokens_slider, temperature_slider, top_p_slider,
|
718 |
frequency_penalty_slider, seed_slider, provider_radio, byok_textbox, custom_model_box,
|
719 |
-
model_search_box, featured_model_radio,
|
720 |
[chatbot]
|
721 |
).then(
|
722 |
lambda: {"text": "", "files": []}, # Clear inputs after submission
|
@@ -730,7 +730,7 @@ with gr.Blocks(theme="Nymbo/Nymbo_Theme") as demo:
|
|
730 |
inputs=model_search_box,
|
731 |
outputs=featured_model_radio
|
732 |
)
|
733 |
-
|
734 |
|
735 |
# Connect the featured model radio to update the custom model box
|
736 |
featured_model_radio.change(
|
@@ -738,7 +738,7 @@ with gr.Blocks(theme="Nymbo/Nymbo_Theme") as demo:
|
|
738 |
inputs=featured_model_radio,
|
739 |
outputs=custom_model_box
|
740 |
)
|
741 |
-
|
742 |
|
743 |
# Connect the BYOK textbox to validate provider selection
|
744 |
byok_textbox.change(
|
@@ -746,7 +746,7 @@ with gr.Blocks(theme="Nymbo/Nymbo_Theme") as demo:
|
|
746 |
inputs=[byok_textbox, provider_radio],
|
747 |
outputs=provider_radio
|
748 |
)
|
749 |
-
|
750 |
|
751 |
# Also validate provider when the radio changes to ensure consistency
|
752 |
provider_radio.change(
|
@@ -754,10 +754,17 @@ with gr.Blocks(theme="Nymbo/Nymbo_Theme") as demo:
|
|
754 |
inputs=[byok_textbox, provider_radio],
|
755 |
outputs=provider_radio
|
756 |
)
|
757 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
758 |
|
759 |
-
|
760 |
|
761 |
if __name__ == "__main__":
|
762 |
-
|
763 |
demo.launch(show_api=True)
|
|
|
6 |
from PIL import Image
|
7 |
import io
|
8 |
import requests
|
9 |
+
from typing import Dict, List, Optional, Any, Union
|
10 |
+
import time
|
11 |
+
import logging
|
12 |
+
|
13 |
+
# Setup logging
|
14 |
+
logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(name)s - %(levelname)s - %(message)s')
|
15 |
+
logger = logging.getLogger(__name__)
|
16 |
|
17 |
ACCESS_TOKEN = os.getenv("HF_TOKEN")
|
18 |
+
logger.info("Access token loaded.")
|
19 |
+
|
20 |
+
# MCP Client Configuration
|
21 |
+
MCP_SERVERS = {}
|
22 |
+
try:
|
23 |
+
mcp_config = os.getenv("MCP_CONFIG")
|
24 |
+
if mcp_config:
|
25 |
+
MCP_SERVERS = json.loads(mcp_config)
|
26 |
+
logger.info(f"Loaded MCP configuration: {len(MCP_SERVERS)} servers defined")
|
27 |
+
except Exception as e:
|
28 |
+
logger.error(f"Error loading MCP configuration: {e}")
|
29 |
|
30 |
# Function to encode image to base64
|
31 |
def encode_image(image_path):
|
32 |
if not image_path:
|
33 |
+
logger.warning("No image path provided")
|
34 |
return None
|
35 |
|
36 |
try:
|
37 |
+
logger.info(f"Encoding image from path: {image_path}")
|
38 |
|
39 |
# If it's already a PIL Image
|
40 |
if isinstance(image_path, Image.Image):
|
|
|
51 |
buffered = io.BytesIO()
|
52 |
image.save(buffered, format="JPEG")
|
53 |
img_str = base64.b64encode(buffered.getvalue()).decode("utf-8")
|
54 |
+
logger.info("Image encoded successfully")
|
55 |
return img_str
|
56 |
except Exception as e:
|
57 |
+
logger.error(f"Error encoding image: {e}")
|
58 |
return None
|
59 |
|
60 |
+
# MCP Client implementation
|
61 |
class MCPClient:
|
62 |
+
def __init__(self, server_url: str):
|
63 |
+
self.server_url = server_url
|
64 |
+
self.session_id = None
|
65 |
+
logger.info(f"Initialized MCP Client for server: {server_url}")
|
|
|
66 |
|
67 |
+
def connect(self) -> bool:
|
68 |
+
"""Establish connection with the MCP server"""
|
69 |
try:
|
70 |
+
response = requests.post(
|
71 |
+
f"{self.server_url}/connect",
|
72 |
+
json={"client": "Serverless-TextGen-Hub", "version": "1.0.0"}
|
73 |
+
)
|
74 |
+
if response.status_code == 200:
|
75 |
+
result = response.json()
|
76 |
+
self.session_id = result.get("session_id")
|
77 |
+
logger.info(f"Connected to MCP server with session ID: {self.session_id}")
|
78 |
+
return True
|
79 |
+
else:
|
80 |
+
logger.error(f"Failed to connect to MCP server: {response.status_code} - {response.text}")
|
81 |
+
return False
|
82 |
except Exception as e:
|
83 |
+
logger.error(f"Error connecting to MCP server: {e}")
|
84 |
return False
|
85 |
|
86 |
+
def list_tools(self) -> List[Dict]:
|
87 |
+
"""List available tools from the MCP server"""
|
88 |
+
if not self.session_id:
|
89 |
+
if not self.connect():
|
90 |
+
return []
|
|
|
|
|
|
|
|
|
|
|
91 |
|
92 |
try:
|
93 |
+
response = requests.get(
|
94 |
+
f"{self.server_url}/tools/list",
|
95 |
+
headers={"X-MCP-Session": self.session_id}
|
96 |
+
)
|
97 |
+
if response.status_code == 200:
|
98 |
+
result = response.json()
|
99 |
+
tools = result.get("tools", [])
|
100 |
+
logger.info(f"Retrieved {len(tools)} tools from MCP server")
|
101 |
+
return tools
|
102 |
+
else:
|
103 |
+
logger.error(f"Failed to list tools: {response.status_code} - {response.text}")
|
104 |
+
return []
|
105 |
except Exception as e:
|
106 |
+
logger.error(f"Error listing tools: {e}")
|
107 |
+
return []
|
108 |
|
109 |
+
def call_tool(self, tool_name: str, args: Dict) -> Dict:
|
110 |
+
"""Call a tool on the MCP server"""
|
111 |
+
if not self.session_id:
|
112 |
+
if not self.connect():
|
113 |
+
return {"error": "Not connected to MCP server"}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
114 |
|
115 |
+
try:
|
116 |
+
response = requests.post(
|
117 |
+
f"{self.server_url}/tools/call",
|
118 |
+
headers={"X-MCP-Session": self.session_id},
|
119 |
+
json={"name": tool_name, "arguments": args}
|
120 |
+
)
|
121 |
+
if response.status_code == 200:
|
122 |
+
result = response.json()
|
123 |
+
logger.info(f"Successfully called tool {tool_name}")
|
124 |
+
return result
|
125 |
+
else:
|
126 |
+
error_msg = f"Failed to call tool {tool_name}: {response.status_code} - {response.text}"
|
127 |
+
logger.error(error_msg)
|
128 |
+
return {"error": error_msg}
|
129 |
+
except Exception as e:
|
130 |
+
error_msg = f"Error calling tool {tool_name}: {e}"
|
131 |
+
logger.error(error_msg)
|
132 |
+
return {"error": error_msg}
|
133 |
+
|
134 |
+
# Text-to-speech client function
|
135 |
+
def text_to_speech(text: str, server_name: str = None) -> Optional[str]:
|
136 |
"""
|
137 |
+
Convert text to speech using an MCP TTS server
|
138 |
+
Returns an audio URL that can be embedded in the chat
|
139 |
+
"""
|
140 |
+
if not server_name or server_name not in MCP_SERVERS:
|
141 |
+
logger.warning(f"TTS server {server_name} not configured")
|
142 |
return None
|
143 |
|
144 |
+
server_url = MCP_SERVERS[server_name].get("url")
|
145 |
+
if not server_url:
|
146 |
+
logger.warning(f"No URL found for TTS server {server_name}")
|
147 |
+
return None
|
148 |
+
|
149 |
+
client = MCPClient(server_url)
|
150 |
+
|
151 |
+
# List available tools to find the TTS tool
|
152 |
+
tools = client.list_tools()
|
153 |
+
tts_tool = next((t for t in tools if "text_to_audio" in t["name"] or "tts" in t["name"]), None)
|
154 |
+
|
155 |
+
if not tts_tool:
|
156 |
+
logger.warning(f"No TTS tool found on server {server_name}")
|
157 |
+
return None
|
158 |
+
|
159 |
+
# Call the TTS tool
|
160 |
+
result = client.call_tool(tts_tool["name"], {"text": text, "speed": 1.0})
|
161 |
+
|
162 |
+
if "error" in result:
|
163 |
+
logger.error(f"TTS error: {result['error']}")
|
164 |
+
return None
|
165 |
+
|
166 |
+
# Process the result - usually a base64 encoded WAV
|
167 |
+
audio_data = result.get("audio") or result.get("content") or result.get("result")
|
168 |
+
|
169 |
+
if isinstance(audio_data, str) and audio_data.startswith("data:audio"):
|
170 |
+
# Already a data URL
|
171 |
+
return audio_data
|
172 |
+
elif isinstance(audio_data, str):
|
173 |
+
# Assume it's base64 encoded
|
174 |
+
return f"data:audio/wav;base64,{audio_data}"
|
175 |
+
else:
|
176 |
+
logger.error(f"Unexpected TTS result format: {type(audio_data)}")
|
177 |
return None
|
178 |
|
179 |
def respond(
|
180 |
message,
|
181 |
+
image_files, # Changed parameter name and structure
|
182 |
history: list[tuple[str, str]],
|
183 |
system_message,
|
184 |
max_tokens,
|
|
|
191 |
custom_model,
|
192 |
model_search_term,
|
193 |
selected_model,
|
|
|
194 |
tts_enabled=False,
|
195 |
+
tts_server=None
|
196 |
):
|
197 |
+
logger.info(f"Received message: {message}")
|
198 |
+
logger.info(f"Received {len(image_files) if image_files else 0} images")
|
199 |
+
logger.info(f"History: {history}")
|
200 |
+
logger.info(f"System message: {system_message}")
|
201 |
+
logger.info(f"Max tokens: {max_tokens}, Temperature: {temperature}, Top-P: {top_p}")
|
202 |
+
logger.info(f"Frequency Penalty: {frequency_penalty}, Seed: {seed}")
|
203 |
+
logger.info(f"Selected provider: {provider}")
|
204 |
+
logger.info(f"Custom API Key provided: {bool(custom_api_key.strip())}")
|
205 |
+
logger.info(f"Selected model (custom_model): {custom_model}")
|
206 |
+
logger.info(f"Model search term: {model_search_term}")
|
207 |
+
logger.info(f"Selected model from radio: {selected_model}")
|
208 |
+
logger.info(f"TTS enabled: {tts_enabled}, TTS server: {tts_server}")
|
|
|
209 |
|
210 |
# Determine which token to use
|
211 |
token_to_use = custom_api_key if custom_api_key.strip() != "" else ACCESS_TOKEN
|
212 |
|
213 |
if custom_api_key.strip() != "":
|
214 |
+
logger.info("USING CUSTOM API KEY: BYOK token provided by user is being used for authentication")
|
215 |
else:
|
216 |
+
logger.info("USING DEFAULT API KEY: Environment variable HF_TOKEN is being used for authentication")
|
217 |
|
218 |
# Initialize the Inference Client with the provider and appropriate token
|
219 |
client = InferenceClient(token=token_to_use, provider=provider)
|
220 |
+
logger.info(f"Hugging Face Inference Client initialized with {provider} provider.")
|
221 |
|
222 |
# Convert seed to None if -1 (meaning random)
|
223 |
if seed == -1:
|
|
|
249 |
}
|
250 |
})
|
251 |
except Exception as e:
|
252 |
+
logger.error(f"Error encoding image: {e}")
|
253 |
else:
|
254 |
# Text-only message
|
255 |
user_content = message
|
256 |
|
257 |
# Prepare messages in the format expected by the API
|
258 |
messages = [{"role": "system", "content": system_message}]
|
259 |
+
logger.info("Initial messages array constructed.")
|
260 |
|
261 |
# Add conversation history to the context
|
262 |
for val in history:
|
|
|
285 |
}
|
286 |
})
|
287 |
except Exception as e:
|
288 |
+
logger.error(f"Error encoding history image: {e}")
|
289 |
|
290 |
messages.append({"role": "user", "content": history_content})
|
291 |
else:
|
292 |
# Regular text message
|
293 |
messages.append({"role": "user", "content": user_part})
|
294 |
+
logger.info(f"Added user message to context (type: {type(user_part)})")
|
295 |
|
296 |
if assistant_part:
|
297 |
messages.append({"role": "assistant", "content": assistant_part})
|
298 |
+
logger.info(f"Added assistant message to context: {assistant_part}")
|
299 |
|
300 |
# Append the latest user message
|
301 |
messages.append({"role": "user", "content": user_content})
|
302 |
+
logger.info(f"Latest user message appended (content type: {type(user_content)})")
|
303 |
|
304 |
# Determine which model to use, prioritizing custom_model if provided
|
305 |
model_to_use = custom_model.strip() if custom_model.strip() != "" else selected_model
|
306 |
+
logger.info(f"Model selected for inference: {model_to_use}")
|
307 |
|
308 |
# Start with an empty string to build the response as tokens stream in
|
309 |
response = ""
|
310 |
+
logger.info(f"Sending request to {provider} provider.")
|
311 |
|
312 |
# Prepare parameters for the chat completion request
|
313 |
parameters = {
|
|
|
330 |
**parameters
|
331 |
)
|
332 |
|
333 |
+
logger.info("Received tokens: ")
|
334 |
|
335 |
# Process the streaming response
|
336 |
for chunk in stream:
|
|
|
343 |
response += token_text
|
344 |
yield response
|
345 |
|
346 |
+
# If TTS is enabled and we have a response, convert it to speech
|
347 |
+
if tts_enabled and tts_server and response:
|
348 |
+
logger.info(f"Converting response to speech using TTS server: {tts_server}")
|
349 |
+
try:
|
350 |
+
audio_url = text_to_speech(response, tts_server)
|
351 |
+
if audio_url:
|
352 |
+
# Add audio tag to the end of the response
|
353 |
+
response += f"\n\n<audio src='{audio_url}' controls></audio>"
|
354 |
+
yield response
|
355 |
+
else:
|
356 |
+
logger.warning("TTS conversion failed, continuing without audio")
|
357 |
+
except Exception as e:
|
358 |
+
logger.error(f"Error in TTS conversion: {e}")
|
359 |
+
# Continue without TTS if there's an error
|
360 |
+
|
361 |
print()
|
362 |
except Exception as e:
|
363 |
+
logger.error(f"Error during inference: {e}")
|
364 |
response += f"\nError: {str(e)}"
|
365 |
yield response
|
366 |
|
367 |
+
logger.info("Completed response generation.")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
368 |
|
369 |
# Function to validate provider selection based on BYOK
|
370 |
def validate_provider(api_key, provider):
|
|
|
372 |
return gr.update(value="hf-inference")
|
373 |
return gr.update(value=provider)
|
374 |
|
375 |
+
# Function to list available MCP servers
|
376 |
+
def list_mcp_servers():
|
377 |
+
"""List all configured MCP servers"""
|
378 |
+
return list(MCP_SERVERS.keys())
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
379 |
|
380 |
# GRADIO UI
|
381 |
with gr.Blocks(theme="Nymbo/Nymbo_Theme") as demo:
|
|
|
383 |
chatbot = gr.Chatbot(
|
384 |
height=600,
|
385 |
show_copy_button=True,
|
386 |
+
placeholder="Select a model and begin chatting. Now supports multiple inference providers, multimodal inputs, and MCP servers",
|
387 |
layout="panel"
|
388 |
)
|
389 |
+
logger.info("Chatbot interface created.")
|
390 |
|
391 |
# Multimodal textbox for messages (combines text and file uploads)
|
392 |
msg = gr.MultimodalTextbox(
|
|
|
531 |
)
|
532 |
|
533 |
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)")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
534 |
|
535 |
+
# MCP TTS integration
|
536 |
+
with gr.Accordion("MCP Integration", open=False):
|
537 |
+
gr.Markdown("## Model Context Protocol (MCP) Integration")
|
538 |
+
gr.Markdown("Connect to MCP servers to extend functionality.")
|
539 |
+
|
540 |
+
tts_enabled = gr.Checkbox(
|
541 |
+
label="Enable Text-to-Speech",
|
542 |
+
value=False,
|
543 |
+
info="When enabled, responses will be converted to speech using the selected MCP TTS server"
|
544 |
+
)
|
545 |
+
|
546 |
+
# Create dropdown for available MCP servers
|
547 |
+
available_servers = list_mcp_servers()
|
548 |
+
tts_server = gr.Dropdown(
|
549 |
+
label="TTS Server",
|
550 |
+
choices=available_servers,
|
551 |
+
value=available_servers[0] if available_servers else None,
|
552 |
+
interactive=True,
|
553 |
+
visible=len(available_servers) > 0
|
554 |
+
)
|
555 |
+
|
556 |
+
# If no servers configured, show a message
|
557 |
+
if not available_servers:
|
558 |
+
gr.Markdown("""
|
559 |
+
No MCP servers configured. Add them using the MCP_CONFIG environment variable:
|
560 |
+
```json
|
561 |
+
{
|
562 |
+
"kokoroTTS": {
|
563 |
+
"url": "https://your-kokoro-tts-server/gradio_api/mcp/sse"
|
564 |
+
}
|
565 |
+
}
|
566 |
+
```
|
567 |
+
""")
|
|
|
|
|
568 |
|
569 |
# Chat history state
|
570 |
chat_history = gr.State([])
|
571 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
572 |
# Function to filter models
|
573 |
def filter_models(search_term):
|
574 |
+
logger.info(f"Filtering models with search term: {search_term}")
|
575 |
filtered = [m for m in models_list if search_term.lower() in m.lower()]
|
576 |
+
logger.info(f"Filtered models: {filtered}")
|
577 |
return gr.update(choices=filtered)
|
578 |
|
579 |
# Function to set custom model from radio
|
580 |
def set_custom_model_from_radio(selected):
|
581 |
+
logger.info(f"Featured model selected: {selected}")
|
582 |
return selected
|
583 |
|
584 |
# Function for the chat interface
|
585 |
def user(user_message, history):
|
586 |
# Debug logging for troubleshooting
|
587 |
+
logger.info(f"User message received: {user_message}")
|
588 |
|
589 |
# Skip if message is empty (no text and no files)
|
590 |
if not user_message or (not user_message.get("text") and not user_message.get("files")):
|
591 |
+
logger.info("Empty message, skipping")
|
592 |
return history
|
593 |
|
594 |
# Prepare multimodal message format
|
595 |
text_content = user_message.get("text", "").strip()
|
596 |
files = user_message.get("files", [])
|
597 |
|
598 |
+
logger.info(f"Text content: {text_content}")
|
599 |
+
logger.info(f"Files: {files}")
|
600 |
|
601 |
# If both text and files are empty, skip
|
602 |
if not text_content and not files:
|
603 |
+
logger.info("No content to display")
|
604 |
return history
|
605 |
|
606 |
# Add message with images to history
|
|
|
608 |
# Add text message first if it exists
|
609 |
if text_content:
|
610 |
# Add a separate text message
|
611 |
+
logger.info(f"Adding text message: {text_content}")
|
612 |
history.append([text_content, None])
|
613 |
|
614 |
# Then add each image file separately
|
615 |
for file_path in files:
|
616 |
if file_path and isinstance(file_path, str):
|
617 |
+
logger.info(f"Adding image: {file_path}")
|
618 |
# Add image as a separate message with no text
|
619 |
history.append([f"", None])
|
620 |
|
621 |
return history
|
622 |
else:
|
623 |
# For text-only messages
|
624 |
+
logger.info(f"Adding text-only message: {text_content}")
|
625 |
history.append([text_content, None])
|
626 |
return history
|
627 |
|
628 |
# Define bot response function
|
629 |
+
def bot(history, system_msg, max_tokens, temperature, top_p, freq_penalty, seed, provider, api_key, custom_model, search_term, selected_model, tts_enabled, tts_server):
|
630 |
# Check if history is valid
|
631 |
if not history or len(history) == 0:
|
632 |
+
logger.info("No history to process")
|
633 |
return history
|
634 |
|
635 |
# Get the most recent message and detect if it's an image
|
636 |
user_message = history[-1][0]
|
637 |
+
logger.info(f"Processing user message: {user_message}")
|
638 |
|
639 |
is_image = False
|
640 |
image_path = None
|
|
|
645 |
is_image = True
|
646 |
# Extract image path from markdown format 
|
647 |
image_path = user_message.replace(".replace(")", "")
|
648 |
+
logger.info(f"Image detected: {image_path}")
|
649 |
text_content = "" # No text for image-only messages
|
650 |
|
651 |
# Look back for text context if this is an image
|
|
|
655 |
prev_message = history[-2][0]
|
656 |
if isinstance(prev_message, str) and not prev_message.startswith(":
|
657 |
text_context = prev_message
|
658 |
+
logger.info(f"Using text context from previous message: {text_context}")
|
659 |
|
660 |
# Process message through respond function
|
661 |
history[-1][1] = ""
|
|
|
678 |
custom_model,
|
679 |
search_term,
|
680 |
selected_model,
|
681 |
+
tts_enabled,
|
682 |
+
tts_server
|
|
|
683 |
):
|
684 |
history[-1][1] = response
|
685 |
yield history
|
|
|
700 |
custom_model,
|
701 |
search_term,
|
702 |
selected_model,
|
703 |
+
tts_enabled,
|
704 |
+
tts_server
|
|
|
705 |
):
|
706 |
history[-1][1] = response
|
707 |
yield history
|
|
|
716 |
bot,
|
717 |
[chatbot, system_message_box, max_tokens_slider, temperature_slider, top_p_slider,
|
718 |
frequency_penalty_slider, seed_slider, provider_radio, byok_textbox, custom_model_box,
|
719 |
+
model_search_box, featured_model_radio, tts_enabled, tts_server],
|
720 |
[chatbot]
|
721 |
).then(
|
722 |
lambda: {"text": "", "files": []}, # Clear inputs after submission
|
|
|
730 |
inputs=model_search_box,
|
731 |
outputs=featured_model_radio
|
732 |
)
|
733 |
+
logger.info("Model search box change event linked.")
|
734 |
|
735 |
# Connect the featured model radio to update the custom model box
|
736 |
featured_model_radio.change(
|
|
|
738 |
inputs=featured_model_radio,
|
739 |
outputs=custom_model_box
|
740 |
)
|
741 |
+
logger.info("Featured model radio button change event linked.")
|
742 |
|
743 |
# Connect the BYOK textbox to validate provider selection
|
744 |
byok_textbox.change(
|
|
|
746 |
inputs=[byok_textbox, provider_radio],
|
747 |
outputs=provider_radio
|
748 |
)
|
749 |
+
logger.info("BYOK textbox change event linked.")
|
750 |
|
751 |
# Also validate provider when the radio changes to ensure consistency
|
752 |
provider_radio.change(
|
|
|
754 |
inputs=[byok_textbox, provider_radio],
|
755 |
outputs=provider_radio
|
756 |
)
|
757 |
+
logger.info("Provider radio button change event linked.")
|
758 |
+
|
759 |
+
# Update TTS server dropdown visibility based on the TTS toggle
|
760 |
+
tts_enabled.change(
|
761 |
+
lambda enabled: gr.update(visible=enabled and len(list_mcp_servers()) > 0),
|
762 |
+
inputs=tts_enabled,
|
763 |
+
outputs=tts_server
|
764 |
+
)
|
765 |
|
766 |
+
logger.info("Gradio interface initialized.")
|
767 |
|
768 |
if __name__ == "__main__":
|
769 |
+
logger.info("Launching the demo application.")
|
770 |
demo.launch(show_api=True)
|