Spaces:
Running
on
Zero
Running
on
Zero
Update app.py
Browse files
app.py
CHANGED
@@ -9,6 +9,7 @@ from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStream
|
|
9 |
from transformers import Qwen2VLForConditionalGeneration, AutoProcessor, TextIteratorStreamer
|
10 |
from transformers.image_utils import load_image
|
11 |
import time
|
|
|
12 |
|
13 |
DESCRIPTION = """
|
14 |
# QwQ Edge 💬
|
@@ -58,12 +59,20 @@ model_m = Qwen2VLForConditionalGeneration.from_pretrained(
|
|
58 |
torch_dtype=torch.float16
|
59 |
).to("cuda").eval()
|
60 |
|
|
|
|
|
|
|
61 |
async def text_to_speech(text: str, voice: str, output_file="output.mp3"):
|
62 |
"""Convert text to speech using Edge TTS and save as MP3"""
|
63 |
communicate = edge_tts.Communicate(text, voice)
|
64 |
await communicate.save(output_file)
|
65 |
return output_file
|
66 |
|
|
|
|
|
|
|
|
|
|
|
67 |
def clean_chat_history(chat_history):
|
68 |
"""
|
69 |
Filter out any chat entries whose "content" is not a string.
|
@@ -86,8 +95,8 @@ def generate(
|
|
86 |
repetition_penalty: float = 1.2,
|
87 |
):
|
88 |
"""
|
89 |
-
Generates chatbot responses with support for multimodal input and
|
90 |
-
If the query starts with an @tts
|
91 |
"""
|
92 |
text = input_dict["text"]
|
93 |
files = input_dict.get("files", [])
|
@@ -100,22 +109,36 @@ def generate(
|
|
100 |
else:
|
101 |
images = []
|
102 |
|
|
|
103 |
tts_prefix = "@tts"
|
|
|
104 |
is_tts = any(text.strip().lower().startswith(f"{tts_prefix}{i}") for i in range(1, 3))
|
105 |
-
|
106 |
|
107 |
-
if is_tts
|
108 |
-
|
|
|
109 |
text = text.replace(f"{tts_prefix}{voice_index}", "").strip()
|
110 |
# Clear any previous chat history to avoid concatenation issues
|
111 |
conversation = [{"role": "user", "content": text}]
|
|
|
|
|
|
|
112 |
else:
|
113 |
voice = None
|
114 |
text = text.replace(tts_prefix, "").strip()
|
115 |
conversation = clean_chat_history(chat_history)
|
116 |
conversation.append({"role": "user", "content": text})
|
117 |
|
118 |
-
if
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
119 |
# Multimodal branch using the OCR model
|
120 |
messages = [{
|
121 |
"role": "user",
|
@@ -183,6 +206,7 @@ demo = gr.ChatInterface(
|
|
183 |
],
|
184 |
examples=[
|
185 |
["@tts1 Who is Nikola Tesla, and why did he die?"],
|
|
|
186 |
[{"text": "Extract JSON from the image", "files": ["examples/document.jpg"]}],
|
187 |
[{"text": "summarize the letter", "files": ["examples/1.png"]}],
|
188 |
["A train travels 60 kilometers per hour. If it travels for 5 hours, how far will it travel in total?"],
|
|
|
9 |
from transformers import Qwen2VLForConditionalGeneration, AutoProcessor, TextIteratorStreamer
|
10 |
from transformers.image_utils import load_image
|
11 |
import time
|
12 |
+
from gradio_client import Client # For image generation API
|
13 |
|
14 |
DESCRIPTION = """
|
15 |
# QwQ Edge 💬
|
|
|
59 |
torch_dtype=torch.float16
|
60 |
).to("cuda").eval()
|
61 |
|
62 |
+
# Image generation client
|
63 |
+
image_gen_client = Client("prithivMLmods/STABLE-HAMSTER")
|
64 |
+
|
65 |
async def text_to_speech(text: str, voice: str, output_file="output.mp3"):
|
66 |
"""Convert text to speech using Edge TTS and save as MP3"""
|
67 |
communicate = edge_tts.Communicate(text, voice)
|
68 |
await communicate.save(output_file)
|
69 |
return output_file
|
70 |
|
71 |
+
def image_gen(prompt: str):
|
72 |
+
"""Generate an image using the Stable Hamster API"""
|
73 |
+
result = image_gen_client.predict("Image Generation", None, prompt, api_name="/stable_hamster")
|
74 |
+
return result[1] # Return the generated image
|
75 |
+
|
76 |
def clean_chat_history(chat_history):
|
77 |
"""
|
78 |
Filter out any chat entries whose "content" is not a string.
|
|
|
95 |
repetition_penalty: float = 1.2,
|
96 |
):
|
97 |
"""
|
98 |
+
Generates chatbot responses with support for multimodal input, TTS, and image generation.
|
99 |
+
If the query starts with an @tts or @image command, previous chat history is cleared.
|
100 |
"""
|
101 |
text = input_dict["text"]
|
102 |
files = input_dict.get("files", [])
|
|
|
109 |
else:
|
110 |
images = []
|
111 |
|
112 |
+
# Check for TTS or Image Generation commands
|
113 |
tts_prefix = "@tts"
|
114 |
+
image_prefix = "@image"
|
115 |
is_tts = any(text.strip().lower().startswith(f"{tts_prefix}{i}") for i in range(1, 3))
|
116 |
+
is_image = text.strip().lower().startswith(image_prefix)
|
117 |
|
118 |
+
if is_tts:
|
119 |
+
voice_index = next((i for i in range(1, 3) if text.strip().lower().startswith(f"{tts_prefix}{i}")), None)
|
120 |
+
voice = TTS_VOICES[voice_index - 1] if voice_index else None
|
121 |
text = text.replace(f"{tts_prefix}{voice_index}", "").strip()
|
122 |
# Clear any previous chat history to avoid concatenation issues
|
123 |
conversation = [{"role": "user", "content": text}]
|
124 |
+
elif is_image:
|
125 |
+
text = text.replace(image_prefix, "").strip()
|
126 |
+
conversation = [{"role": "user", "content": text}]
|
127 |
else:
|
128 |
voice = None
|
129 |
text = text.replace(tts_prefix, "").strip()
|
130 |
conversation = clean_chat_history(chat_history)
|
131 |
conversation.append({"role": "user", "content": text})
|
132 |
|
133 |
+
if is_image:
|
134 |
+
# Image generation branch
|
135 |
+
yield "Generating image, please wait..."
|
136 |
+
try:
|
137 |
+
image = image_gen(text)
|
138 |
+
yield gr.Image(image)
|
139 |
+
except Exception as e:
|
140 |
+
yield f"Failed to generate image: {str(e)}"
|
141 |
+
elif images:
|
142 |
# Multimodal branch using the OCR model
|
143 |
messages = [{
|
144 |
"role": "user",
|
|
|
206 |
],
|
207 |
examples=[
|
208 |
["@tts1 Who is Nikola Tesla, and why did he die?"],
|
209 |
+
["@image A futuristic cityscape at sunset"],
|
210 |
[{"text": "Extract JSON from the image", "files": ["examples/document.jpg"]}],
|
211 |
[{"text": "summarize the letter", "files": ["examples/1.png"]}],
|
212 |
["A train travels 60 kilometers per hour. If it travels for 5 hours, how far will it travel in total?"],
|