Spaces:
Running
on
Zero
Running
on
Zero
Update app.py
Browse files
app.py
CHANGED
|
@@ -12,7 +12,6 @@ import spaces
|
|
| 12 |
import torch
|
| 13 |
import numpy as np
|
| 14 |
from PIL import Image
|
| 15 |
-
import edge_tts
|
| 16 |
import cv2
|
| 17 |
|
| 18 |
from transformers import (
|
|
@@ -24,7 +23,6 @@ from transformers import (
|
|
| 24 |
Gemma3ForConditionalGeneration,
|
| 25 |
)
|
| 26 |
from transformers.image_utils import load_image
|
| 27 |
-
from diffusers import StableDiffusionXLPipeline, EulerAncestralDiscreteScheduler
|
| 28 |
|
| 29 |
# Constants
|
| 30 |
MAX_MAX_NEW_TOKENS = 2048
|
|
@@ -51,7 +49,7 @@ def progress_bar_html(label: str) -> str:
|
|
| 51 |
</style>
|
| 52 |
'''
|
| 53 |
|
| 54 |
-
# TEXT
|
| 55 |
|
| 56 |
model_id = "prithivMLmods/FastThink-0.5B-Tiny"
|
| 57 |
tokenizer = AutoTokenizer.from_pretrained(model_id)
|
|
@@ -62,11 +60,6 @@ model = AutoModelForCausalLM.from_pretrained(
|
|
| 62 |
)
|
| 63 |
model.eval()
|
| 64 |
|
| 65 |
-
TTS_VOICES = [
|
| 66 |
-
"en-US-JennyNeural", # @tts1
|
| 67 |
-
"en-US-GuyNeural", # @tts2
|
| 68 |
-
]
|
| 69 |
-
|
| 70 |
# MULTIMODAL (OCR) MODELS
|
| 71 |
|
| 72 |
MODEL_ID_VL = "prithivMLmods/Qwen2-VL-OCR-2B-Instruct"
|
|
@@ -77,11 +70,6 @@ model_m = Qwen2VLForConditionalGeneration.from_pretrained(
|
|
| 77 |
torch_dtype=torch.float16
|
| 78 |
).to("cuda").eval()
|
| 79 |
|
| 80 |
-
async def text_to_speech(text: str, voice: str, output_file="output.mp3"):
|
| 81 |
-
communicate = edge_tts.Communicate(text, voice)
|
| 82 |
-
await communicate.save(output_file)
|
| 83 |
-
return output_file
|
| 84 |
-
|
| 85 |
def clean_chat_history(chat_history):
|
| 86 |
cleaned = []
|
| 87 |
for msg in chat_history:
|
|
@@ -114,46 +102,9 @@ ENABLE_CPU_OFFLOAD = os.getenv("ENABLE_CPU_OFFLOAD", "0") == "1"
|
|
| 114 |
|
| 115 |
dtype = torch.float16 if device.type == "cuda" else torch.float32
|
| 116 |
|
| 117 |
-
# STABLE DIFFUSION IMAGE GENERATION MODEL (Lightning 5 only)
|
| 118 |
-
|
| 119 |
-
if torch.cuda.is_available():
|
| 120 |
-
pipe = StableDiffusionXLPipeline.from_pretrained(
|
| 121 |
-
"SG161222/RealVisXL_V5.0_Lightning",
|
| 122 |
-
torch_dtype=dtype,
|
| 123 |
-
use_safetensors=True,
|
| 124 |
-
add_watermarker=False
|
| 125 |
-
).to(device)
|
| 126 |
-
pipe.text_encoder = pipe.text_encoder.half()
|
| 127 |
-
if ENABLE_CPU_OFFLOAD:
|
| 128 |
-
pipe.enable_model_cpu_offload()
|
| 129 |
-
else:
|
| 130 |
-
pipe.to(device)
|
| 131 |
-
print("Loaded RealVisXL_V5.0_Lightning on Device!")
|
| 132 |
-
if USE_TORCH_COMPILE:
|
| 133 |
-
pipe.unet = torch.compile(pipe.unet, mode="reduce-overhead", fullgraph=True)
|
| 134 |
-
print("Model RealVisXL_V5.0_Lightning Compiled!")
|
| 135 |
-
else:
|
| 136 |
-
pipe = StableDiffusionXLPipeline.from_pretrained(
|
| 137 |
-
"SG161222/RealVisXL_V5.0_Lightning",
|
| 138 |
-
torch_dtype=dtype,
|
| 139 |
-
use_safetensors=True,
|
| 140 |
-
add_watermarker=False
|
| 141 |
-
).to(device)
|
| 142 |
-
print("Running on CPU; model loaded in float32.")
|
| 143 |
-
|
| 144 |
-
DEFAULT_MODEL = "Lightning 5"
|
| 145 |
-
models = {
|
| 146 |
-
"Lightning 5": pipe
|
| 147 |
-
}
|
| 148 |
-
|
| 149 |
-
def save_image(img: Image.Image) -> str:
|
| 150 |
-
unique_name = str(uuid.uuid4()) + ".png"
|
| 151 |
-
img.save(unique_name)
|
| 152 |
-
return unique_name
|
| 153 |
-
|
| 154 |
# GEMMA3-4B MULTIMODAL MODEL
|
| 155 |
|
| 156 |
-
gemma3_model_id = "google/gemma-3-4b-it"
|
| 157 |
gemma3_model = Gemma3ForConditionalGeneration.from_pretrained(
|
| 158 |
gemma3_model_id, device_map="auto"
|
| 159 |
).eval()
|
|
@@ -196,91 +147,51 @@ def generate(
|
|
| 196 |
|
| 197 |
lower_text = text.lower().strip()
|
| 198 |
|
| 199 |
-
# IMAGE GENERATION BRANCH (Stable Diffusion model using @lightningv5)
|
| 200 |
-
if lower_text.startswith("@lightningv5"):
|
| 201 |
-
# Remove the model flag from the prompt.
|
| 202 |
-
prompt_clean = re.sub(r"@lightningv5", "", text, flags=re.IGNORECASE).strip().strip('"')
|
| 203 |
-
|
| 204 |
-
# Default parameters for single image generation.
|
| 205 |
-
width = 1024
|
| 206 |
-
height = 1024
|
| 207 |
-
guidance_scale = 6.0
|
| 208 |
-
seed_val = 0
|
| 209 |
-
randomize_seed_flag = True
|
| 210 |
-
|
| 211 |
-
seed_val = int(randomize_seed_fn(seed_val, randomize_seed_flag))
|
| 212 |
-
generator = torch.Generator(device=device).manual_seed(seed_val)
|
| 213 |
-
|
| 214 |
-
options = {
|
| 215 |
-
"prompt": prompt_clean,
|
| 216 |
-
"negative_prompt": default_negative,
|
| 217 |
-
"width": width,
|
| 218 |
-
"height": height,
|
| 219 |
-
"guidance_scale": guidance_scale,
|
| 220 |
-
"num_inference_steps": 30,
|
| 221 |
-
"generator": generator,
|
| 222 |
-
"num_images_per_prompt": 1,
|
| 223 |
-
"use_resolution_binning": True,
|
| 224 |
-
"output_type": "pil",
|
| 225 |
-
}
|
| 226 |
-
if device.type == "cuda":
|
| 227 |
-
torch.cuda.empty_cache()
|
| 228 |
-
|
| 229 |
-
yield progress_bar_html("Processing Image Generation")
|
| 230 |
-
images = models["Lightning 5"](**options).images
|
| 231 |
-
image_path = save_image(images[0])
|
| 232 |
-
yield gr.Image(image_path)
|
| 233 |
-
return
|
| 234 |
-
|
| 235 |
# GEMMA3-4B TEXT & MULTIMODAL (image) Branch
|
| 236 |
if lower_text.startswith("@gemma3"):
|
| 237 |
-
#
|
| 238 |
-
|
| 239 |
-
|
| 240 |
-
|
| 241 |
-
|
| 242 |
-
|
| 243 |
-
|
| 244 |
-
|
| 245 |
-
|
| 246 |
-
|
| 247 |
-
"role": "user",
|
| 248 |
-
"content": [
|
| 249 |
-
*[{"type": "image", "image": image} for image in images],
|
| 250 |
-
{"type": "text", "text": prompt_clean},
|
| 251 |
-
]
|
| 252 |
-
}]
|
| 253 |
-
else:
|
| 254 |
-
messages = [
|
| 255 |
-
{"role": "system", "content": [{"type": "text", "text": "You are a helpful assistant."}]},
|
| 256 |
-
{"role": "user", "content": [{"type": "text", "text": prompt_clean}]}
|
| 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 |
# GEMMA3-4B VIDEO Branch
|
| 286 |
if lower_text.startswith("@video-infer"):
|
|
@@ -333,20 +244,9 @@ def generate(
|
|
| 333 |
yield buffer
|
| 334 |
return
|
| 335 |
|
| 336 |
-
# Otherwise, handle text/chat
|
| 337 |
-
|
| 338 |
-
|
| 339 |
-
voice_index = next((i for i in range(1, 3) if text.strip().lower().startswith(f"{tts_prefix}{i}")), None)
|
| 340 |
-
|
| 341 |
-
if is_tts and voice_index:
|
| 342 |
-
voice = TTS_VOICES[voice_index - 1]
|
| 343 |
-
text = text.replace(f"{tts_prefix}{voice_index}", "").strip()
|
| 344 |
-
conversation = [{"role": "user", "content": text}]
|
| 345 |
-
else:
|
| 346 |
-
voice = None
|
| 347 |
-
text = text.replace(tts_prefix, "").strip()
|
| 348 |
-
conversation = clean_chat_history(chat_history)
|
| 349 |
-
conversation.append({"role": "user", "content": text})
|
| 350 |
|
| 351 |
if files:
|
| 352 |
images = [load_image(image) for image in files] if len(files) > 1 else [load_image(files[0])]
|
|
@@ -400,10 +300,6 @@ def generate(
|
|
| 400 |
final_response = "".join(outputs)
|
| 401 |
yield final_response
|
| 402 |
|
| 403 |
-
if is_tts and voice:
|
| 404 |
-
output_file = asyncio.run(text_to_speech(final_response, voice))
|
| 405 |
-
yield gr.Audio(output_file, autoplay=True)
|
| 406 |
-
|
| 407 |
demo = gr.ChatInterface(
|
| 408 |
fn=generate,
|
| 409 |
additional_inputs=[
|
|
@@ -422,16 +318,13 @@ demo = gr.ChatInterface(
|
|
| 422 |
[{"text": "@video-infer Explain what is happening in this video ?", "files": ["examples/oreo.mp4"]}],
|
| 423 |
[{"text": "@video-infer Summarize the events in this video", "files": ["examples/sky.mp4"]}],
|
| 424 |
[{"text": "@video-infer What is in the video ?", "files": ["examples/redlight.mp4"]}],
|
| 425 |
-
['@lightningv5 Chocolate dripping from a donut'],
|
| 426 |
["Python Program for Array Rotation"],
|
| 427 |
-
["@tts1 Who is Nikola Tesla, and why did he die?"],
|
| 428 |
-
["@tts2 What causes rainbows to form?"],
|
| 429 |
],
|
| 430 |
cache_examples=False,
|
| 431 |
type="messages",
|
| 432 |
description="# **Gemma 3 `@gemma3, @video-infer for video understanding`**",
|
| 433 |
fill_height=True,
|
| 434 |
-
textbox=gr.MultimodalTextbox(label="Query Input", file_types=["image", "video"], file_count="multiple", placeholder="@gemma3 for multimodal, @video-infer for video
|
| 435 |
stop_btn="Stop Generation",
|
| 436 |
multimodal=True,
|
| 437 |
)
|
|
|
|
| 12 |
import torch
|
| 13 |
import numpy as np
|
| 14 |
from PIL import Image
|
|
|
|
| 15 |
import cv2
|
| 16 |
|
| 17 |
from transformers import (
|
|
|
|
| 23 |
Gemma3ForConditionalGeneration,
|
| 24 |
)
|
| 25 |
from transformers.image_utils import load_image
|
|
|
|
| 26 |
|
| 27 |
# Constants
|
| 28 |
MAX_MAX_NEW_TOKENS = 2048
|
|
|
|
| 49 |
</style>
|
| 50 |
'''
|
| 51 |
|
| 52 |
+
# TEXT MODEL
|
| 53 |
|
| 54 |
model_id = "prithivMLmods/FastThink-0.5B-Tiny"
|
| 55 |
tokenizer = AutoTokenizer.from_pretrained(model_id)
|
|
|
|
| 60 |
)
|
| 61 |
model.eval()
|
| 62 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 63 |
# MULTIMODAL (OCR) MODELS
|
| 64 |
|
| 65 |
MODEL_ID_VL = "prithivMLmods/Qwen2-VL-OCR-2B-Instruct"
|
|
|
|
| 70 |
torch_dtype=torch.float16
|
| 71 |
).to("cuda").eval()
|
| 72 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 73 |
def clean_chat_history(chat_history):
|
| 74 |
cleaned = []
|
| 75 |
for msg in chat_history:
|
|
|
|
| 102 |
|
| 103 |
dtype = torch.float16 if device.type == "cuda" else torch.float32
|
| 104 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 105 |
# GEMMA3-4B MULTIMODAL MODEL
|
| 106 |
|
| 107 |
+
gemma3_model_id = "google/gemma-3-4b-it" # alternative: google/gemma-3-12b-it
|
| 108 |
gemma3_model = Gemma3ForConditionalGeneration.from_pretrained(
|
| 109 |
gemma3_model_id, device_map="auto"
|
| 110 |
).eval()
|
|
|
|
| 147 |
|
| 148 |
lower_text = text.lower().strip()
|
| 149 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 150 |
# GEMMA3-4B TEXT & MULTIMODAL (image) Branch
|
| 151 |
if lower_text.startswith("@gemma3"):
|
| 152 |
+
# Remove the gemma3 flag from the prompt.
|
| 153 |
+
prompt_clean = re.sub(r"@gemma3", "", text, flags=re.IGNORECASE).strip().strip('"')
|
| 154 |
+
if files:
|
| 155 |
+
# If image files are provided, load them.
|
| 156 |
+
images = [load_image(f) for f in files]
|
| 157 |
+
messages = [{
|
| 158 |
+
"role": "user",
|
| 159 |
+
"content": [
|
| 160 |
+
*[{"type": "image", "image": image} for image in images],
|
| 161 |
+
{"type": "text", "text": prompt_clean},
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 162 |
]
|
| 163 |
+
}]
|
| 164 |
+
else:
|
| 165 |
+
messages = [
|
| 166 |
+
{"role": "system", "content": [{"type": "text", "text": "You are a helpful assistant."}]},
|
| 167 |
+
{"role": "user", "content": [{"type": "text", "text": prompt_clean}]}
|
| 168 |
+
]
|
| 169 |
+
inputs = gemma3_processor.apply_chat_template(
|
| 170 |
+
messages, add_generation_prompt=True, tokenize=True,
|
| 171 |
+
return_dict=True, return_tensors="pt"
|
| 172 |
+
).to(gemma3_model.device, dtype=torch.bfloat16)
|
| 173 |
+
streamer = TextIteratorStreamer(
|
| 174 |
+
gemma3_processor.tokenizer, timeout=20.0, skip_prompt=True, skip_special_tokens=True
|
| 175 |
+
)
|
| 176 |
+
generation_kwargs = {
|
| 177 |
+
**inputs,
|
| 178 |
+
"streamer": streamer,
|
| 179 |
+
"max_new_tokens": max_new_tokens,
|
| 180 |
+
"do_sample": True,
|
| 181 |
+
"temperature": temperature,
|
| 182 |
+
"top_p": top_p,
|
| 183 |
+
"top_k": top_k,
|
| 184 |
+
"repetition_penalty": repetition_penalty,
|
| 185 |
+
}
|
| 186 |
+
thread = Thread(target=gemma3_model.generate, kwargs=generation_kwargs)
|
| 187 |
+
thread.start()
|
| 188 |
+
buffer = ""
|
| 189 |
+
yield progress_bar_html("Processing with Gemma3")
|
| 190 |
+
for new_text in streamer:
|
| 191 |
+
buffer += new_text
|
| 192 |
+
time.sleep(0.01)
|
| 193 |
+
yield buffer
|
| 194 |
+
return
|
| 195 |
|
| 196 |
# GEMMA3-4B VIDEO Branch
|
| 197 |
if lower_text.startswith("@video-infer"):
|
|
|
|
| 244 |
yield buffer
|
| 245 |
return
|
| 246 |
|
| 247 |
+
# Otherwise, handle text/chat generation.
|
| 248 |
+
conversation = clean_chat_history(chat_history)
|
| 249 |
+
conversation.append({"role": "user", "content": text})
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 250 |
|
| 251 |
if files:
|
| 252 |
images = [load_image(image) for image in files] if len(files) > 1 else [load_image(files[0])]
|
|
|
|
| 300 |
final_response = "".join(outputs)
|
| 301 |
yield final_response
|
| 302 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 303 |
demo = gr.ChatInterface(
|
| 304 |
fn=generate,
|
| 305 |
additional_inputs=[
|
|
|
|
| 318 |
[{"text": "@video-infer Explain what is happening in this video ?", "files": ["examples/oreo.mp4"]}],
|
| 319 |
[{"text": "@video-infer Summarize the events in this video", "files": ["examples/sky.mp4"]}],
|
| 320 |
[{"text": "@video-infer What is in the video ?", "files": ["examples/redlight.mp4"]}],
|
|
|
|
| 321 |
["Python Program for Array Rotation"],
|
|
|
|
|
|
|
| 322 |
],
|
| 323 |
cache_examples=False,
|
| 324 |
type="messages",
|
| 325 |
description="# **Gemma 3 `@gemma3, @video-infer for video understanding`**",
|
| 326 |
fill_height=True,
|
| 327 |
+
textbox=gr.MultimodalTextbox(label="Query Input", file_types=["image", "video"], file_count="multiple", placeholder="@gemma3 for multimodal, @video-infer for video !"),
|
| 328 |
stop_btn="Stop Generation",
|
| 329 |
multimodal=True,
|
| 330 |
)
|