Spaces:
Running
on
Zero
Running
on
Zero
Update app.py
Browse files
app.py
CHANGED
@@ -1,13 +1,9 @@
|
|
1 |
import gradio as gr
|
2 |
import torch
|
3 |
from transformers import (
|
4 |
-
AutoTokenizer,
|
5 |
-
|
6 |
-
|
7 |
-
SpeechT5ForTextToSpeech,
|
8 |
-
SpeechT5HifiGan,
|
9 |
-
WhisperProcessor,
|
10 |
-
WhisperForConditionalGeneration
|
11 |
)
|
12 |
from datasets import load_dataset
|
13 |
import os
|
@@ -17,7 +13,7 @@ import soundfile as sf
|
|
17 |
import librosa
|
18 |
import yaml
|
19 |
|
20 |
-
#
|
21 |
HUGGINGFACE_MODEL_ID = "HuggingFaceH4/Qwen2.5-1.5B-Instruct-gkd"
|
22 |
TORCH_DTYPE = torch.bfloat16
|
23 |
MAX_NEW_TOKENS = 512
|
@@ -30,7 +26,7 @@ TTS_MODEL_ID = "microsoft/speecht5_tts"
|
|
30 |
TTS_VOCODER_ID = "microsoft/speecht5_hifigan"
|
31 |
STT_MODEL_ID = "openai/whisper-small"
|
32 |
|
33 |
-
#
|
34 |
tokenizer = None
|
35 |
llm_model = None
|
36 |
tts_processor = None
|
@@ -41,9 +37,12 @@ whisper_processor = None
|
|
41 |
whisper_model = None
|
42 |
first_load = True
|
43 |
|
|
|
44 |
def generate_pretty_html(data):
|
45 |
html = """
|
46 |
-
<div style="font-family: Arial, sans-serif; max-width: 600px; margin: auto;
|
|
|
|
|
47 |
<h2 style="color: #2c3e50; border-bottom: 2px solid #ddd; padding-bottom: 10px;">Model Info</h2>
|
48 |
"""
|
49 |
for key, value in data.items():
|
@@ -61,14 +60,17 @@ def load_config():
|
|
61 |
return yaml.safe_load(f)
|
62 |
|
63 |
def render_modern_info():
|
64 |
-
|
65 |
-
|
|
|
|
|
|
|
66 |
|
67 |
def load_readme():
|
68 |
with open("README.md", "r", encoding="utf-8") as f:
|
69 |
return f.read()
|
70 |
|
71 |
-
#
|
72 |
def split_text_into_chunks(text, max_chars=400):
|
73 |
sentences = text.replace("...", ".").split(". ")
|
74 |
chunks = []
|
@@ -83,7 +85,7 @@ def split_text_into_chunks(text, max_chars=400):
|
|
83 |
chunks.append(current_chunk)
|
84 |
return [f"{chunk}." for chunk in chunks if chunk.strip()]
|
85 |
|
86 |
-
#
|
87 |
@spaces.GPU
|
88 |
def load_models():
|
89 |
global tokenizer, llm_model, tts_processor, tts_model, tts_vocoder, speaker_embeddings, whisper_processor, whisper_model
|
@@ -132,7 +134,7 @@ def load_models():
|
|
132 |
except Exception as e:
|
133 |
print(f"Error loading Whisper: {e}")
|
134 |
|
135 |
-
#
|
136 |
@spaces.GPU
|
137 |
def generate_response_and_audio(message, history):
|
138 |
global first_load
|
@@ -196,7 +198,6 @@ def generate_response_and_audio(message, history):
|
|
196 |
|
197 |
return history + [{"role": "assistant", "content": generated_text}], audio_path
|
198 |
|
199 |
-
# --- Transcribe Audio ---
|
200 |
@spaces.GPU
|
201 |
def transcribe_audio(filepath):
|
202 |
global first_load
|
@@ -216,7 +217,7 @@ def transcribe_audio(filepath):
|
|
216 |
except Exception as e:
|
217 |
return f"Transcription failed: {e}"
|
218 |
|
219 |
-
#
|
220 |
with gr.Blocks() as demo:
|
221 |
gr.Markdown("# Qwen2.5 Chatbot with Voice Input/Output")
|
222 |
|
@@ -233,9 +234,15 @@ with gr.Blocks() as demo:
|
|
233 |
|
234 |
clear_btn = gr.Button("Clear All")
|
235 |
clear_btn.click(lambda: ([], "", None), None, [chatbot, text_input, audio_output])
|
236 |
-
|
237 |
gr.Markdown(load_readme())
|
238 |
gr.Markdown("---")
|
239 |
-
#html_output = gr.HTML()
|
240 |
|
241 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
import gradio as gr
|
2 |
import torch
|
3 |
from transformers import (
|
4 |
+
AutoTokenizer, AutoModelForCausalLM,
|
5 |
+
SpeechT5Processor, SpeechT5ForTextToSpeech, SpeechT5HifiGan,
|
6 |
+
WhisperProcessor, WhisperForConditionalGeneration
|
|
|
|
|
|
|
|
|
7 |
)
|
8 |
from datasets import load_dataset
|
9 |
import os
|
|
|
13 |
import librosa
|
14 |
import yaml
|
15 |
|
16 |
+
# ================== Configuration ==================
|
17 |
HUGGINGFACE_MODEL_ID = "HuggingFaceH4/Qwen2.5-1.5B-Instruct-gkd"
|
18 |
TORCH_DTYPE = torch.bfloat16
|
19 |
MAX_NEW_TOKENS = 512
|
|
|
26 |
TTS_VOCODER_ID = "microsoft/speecht5_hifigan"
|
27 |
STT_MODEL_ID = "openai/whisper-small"
|
28 |
|
29 |
+
# ================== Global Variables ==================
|
30 |
tokenizer = None
|
31 |
llm_model = None
|
32 |
tts_processor = None
|
|
|
37 |
whisper_model = None
|
38 |
first_load = True
|
39 |
|
40 |
+
# ================== UI Helpers ==================
|
41 |
def generate_pretty_html(data):
|
42 |
html = """
|
43 |
+
<div style="font-family: Arial, sans-serif; max-width: 600px; margin: auto;
|
44 |
+
background-color: #f9f9f9; border-radius: 10px; padding: 20px;
|
45 |
+
box-shadow: 0 4px 12px rgba(0,0,0,0.1);">
|
46 |
<h2 style="color: #2c3e50; border-bottom: 2px solid #ddd; padding-bottom: 10px;">Model Info</h2>
|
47 |
"""
|
48 |
for key, value in data.items():
|
|
|
60 |
return yaml.safe_load(f)
|
61 |
|
62 |
def render_modern_info():
|
63 |
+
try:
|
64 |
+
config = load_config()
|
65 |
+
return generate_pretty_html(config)
|
66 |
+
except Exception as e:
|
67 |
+
return f"<div style='color: red;'>Error loading config: {str(e)}</div>"
|
68 |
|
69 |
def load_readme():
|
70 |
with open("README.md", "r", encoding="utf-8") as f:
|
71 |
return f.read()
|
72 |
|
73 |
+
# ================== Helper Functions ==================
|
74 |
def split_text_into_chunks(text, max_chars=400):
|
75 |
sentences = text.replace("...", ".").split(". ")
|
76 |
chunks = []
|
|
|
85 |
chunks.append(current_chunk)
|
86 |
return [f"{chunk}." for chunk in chunks if chunk.strip()]
|
87 |
|
88 |
+
# ================== Model Loading ==================
|
89 |
@spaces.GPU
|
90 |
def load_models():
|
91 |
global tokenizer, llm_model, tts_processor, tts_model, tts_vocoder, speaker_embeddings, whisper_processor, whisper_model
|
|
|
134 |
except Exception as e:
|
135 |
print(f"Error loading Whisper: {e}")
|
136 |
|
137 |
+
# ================== Chat & Audio Functions ==================
|
138 |
@spaces.GPU
|
139 |
def generate_response_and_audio(message, history):
|
140 |
global first_load
|
|
|
198 |
|
199 |
return history + [{"role": "assistant", "content": generated_text}], audio_path
|
200 |
|
|
|
201 |
@spaces.GPU
|
202 |
def transcribe_audio(filepath):
|
203 |
global first_load
|
|
|
217 |
except Exception as e:
|
218 |
return f"Transcription failed: {e}"
|
219 |
|
220 |
+
# ================== Gradio UI ==================
|
221 |
with gr.Blocks() as demo:
|
222 |
gr.Markdown("# Qwen2.5 Chatbot with Voice Input/Output")
|
223 |
|
|
|
234 |
|
235 |
clear_btn = gr.Button("Clear All")
|
236 |
clear_btn.click(lambda: ([], "", None), None, [chatbot, text_input, audio_output])
|
237 |
+
|
238 |
gr.Markdown(load_readme())
|
239 |
gr.Markdown("---")
|
|
|
240 |
|
241 |
+
# ✅ Define html_output BEFORE using it
|
242 |
+
html_output = gr.HTML("<div style='text-align:center; padding: 20px;'>Loading model info...</div>")
|
243 |
+
|
244 |
+
# ✅ Now this works!
|
245 |
+
demo.load(fn=render_modern_info, outputs=html_output)
|
246 |
+
|
247 |
+
# ================== Launch App ==================
|
248 |
+
demo.queue().launch()
|