Spaces:
Running
on
Zero
Running
on
Zero
added multimodel loading
Browse files
app.py
CHANGED
@@ -17,20 +17,44 @@ from loguru import logger
|
|
17 |
from PIL import Image
|
18 |
|
19 |
dotenv_path = find_dotenv()
|
20 |
-
|
21 |
load_dotenv(dotenv_path)
|
22 |
|
23 |
-
|
24 |
-
|
25 |
-
|
26 |
-
|
27 |
-
|
28 |
-
|
29 |
-
|
30 |
-
|
31 |
-
|
32 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
33 |
|
|
|
|
|
34 |
|
35 |
def get_frames(video_path: str, max_images: int) -> list[tuple[Image.Image, float]]:
|
36 |
frames: list[tuple[Image.Image, float]] = []
|
@@ -123,10 +147,25 @@ def process_history(history: list[dict]) -> list[dict]:
|
|
123 |
return messages
|
124 |
|
125 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
126 |
@spaces.GPU(duration=120)
|
127 |
def run(
|
128 |
message: dict,
|
129 |
history: list[dict],
|
|
|
130 |
system_prompt: str,
|
131 |
max_new_tokens: int,
|
132 |
max_images: int,
|
@@ -135,12 +174,25 @@ def run(
|
|
135 |
top_k: int,
|
136 |
repetition_penalty: float,
|
137 |
) -> Iterator[str]:
|
138 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
139 |
logger.debug(
|
140 |
-
f"\n message: {message} \n history: {history} \n
|
141 |
-
f"max_new_tokens: {max_new_tokens} \n max_images: {max_images}"
|
142 |
)
|
143 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
144 |
messages = []
|
145 |
if system_prompt:
|
146 |
messages.append(
|
@@ -151,16 +203,16 @@ def run(
|
|
151 |
{"role": "user", "content": process_user_input(message, max_images)}
|
152 |
)
|
153 |
|
154 |
-
inputs =
|
155 |
messages,
|
156 |
add_generation_prompt=True,
|
157 |
tokenize=True,
|
158 |
return_dict=True,
|
159 |
return_tensors="pt",
|
160 |
-
).to(device=
|
161 |
|
162 |
streamer = TextIteratorStreamer(
|
163 |
-
|
164 |
)
|
165 |
generate_kwargs = dict(
|
166 |
inputs,
|
@@ -172,7 +224,7 @@ def run(
|
|
172 |
repetition_penalty=repetition_penalty,
|
173 |
do_sample=True,
|
174 |
)
|
175 |
-
t = Thread(target=
|
176 |
t.start()
|
177 |
|
178 |
output = ""
|
@@ -180,36 +232,53 @@ def run(
|
|
180 |
output += delta
|
181 |
yield output
|
182 |
|
183 |
-
|
184 |
-
|
185 |
-
|
186 |
-
|
187 |
-
|
188 |
-
|
189 |
-
|
190 |
-
|
191 |
-
|
192 |
-
|
193 |
-
|
194 |
-
|
195 |
-
|
196 |
-
),
|
197 |
-
gr.Slider(label="Max Images", minimum=1, maximum=4, step=1, value=2),
|
198 |
-
gr.Slider(
|
199 |
-
label="Temperature", minimum=0.1, maximum=2.0, step=0.1, value=0.7
|
200 |
-
),
|
201 |
-
gr.Slider(
|
202 |
-
label="Top P", minimum=0.1, maximum=1.0, step=0.05, value=0.9
|
203 |
-
),
|
204 |
-
gr.Slider(
|
205 |
-
label="Top K", minimum=1, maximum=100, step=1, value=50
|
206 |
),
|
207 |
-
|
208 |
-
|
209 |
-
|
210 |
-
|
211 |
-
|
212 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
213 |
|
214 |
if __name__ == "__main__":
|
215 |
demo.launch()
|
|
|
17 |
from PIL import Image
|
18 |
|
19 |
dotenv_path = find_dotenv()
|
|
|
20 |
load_dotenv(dotenv_path)
|
21 |
|
22 |
+
MODEL_CONFIGS = {
|
23 |
+
"Gemma 3 4B IT": {
|
24 |
+
"id": os.getenv("MODEL_ID_27", "google/gemma-3-4b-it"),
|
25 |
+
"supports_video": True,
|
26 |
+
"supports_pdf": False
|
27 |
+
},
|
28 |
+
"Gemma 3 1B IT": {
|
29 |
+
"id": os.getenv("MODEL_ID_12", "google/gemma-3-1b-it"),
|
30 |
+
"supports_video": True,
|
31 |
+
"supports_pdf": False
|
32 |
+
},
|
33 |
+
"Gemma 3N E4B IT": {
|
34 |
+
"id": os.getenv("MODEL_ID_3N", "google/gemma-3n-E4B-it"),
|
35 |
+
"supports_video": False,
|
36 |
+
"supports_pdf": False
|
37 |
+
}
|
38 |
+
}
|
39 |
+
|
40 |
+
# Load all models and processors
|
41 |
+
models = {}
|
42 |
+
processor = Gemma3Processor.from_pretrained("google/gemma-3-4b-it")
|
43 |
+
|
44 |
+
for model_name, config in MODEL_CONFIGS.items():
|
45 |
+
logger.info(f"Loading {model_name}...")
|
46 |
+
|
47 |
+
models[model_name] = Gemma3ForConditionalGeneration.from_pretrained(
|
48 |
+
config["id"],
|
49 |
+
torch_dtype=torch.bfloat16,
|
50 |
+
device_map="auto",
|
51 |
+
attn_implementation="eager",
|
52 |
+
)
|
53 |
+
|
54 |
+
logger.info(f"✓ {model_name} loaded successfully")
|
55 |
|
56 |
+
# Current model selection (default)
|
57 |
+
current_model = "Gemma 3 27B IT"
|
58 |
|
59 |
def get_frames(video_path: str, max_images: int) -> list[tuple[Image.Image, float]]:
|
60 |
frames: list[tuple[Image.Image, float]] = []
|
|
|
147 |
return messages
|
148 |
|
149 |
|
150 |
+
def get_supported_file_types(model_name: str) -> list[str]:
|
151 |
+
"""Get supported file types for the selected model."""
|
152 |
+
config = MODEL_CONFIGS[model_name]
|
153 |
+
|
154 |
+
base_types = [".jpg", ".png", ".jpeg", ".gif", ".bmp", ".webp"]
|
155 |
+
|
156 |
+
if config["supports_video"]:
|
157 |
+
base_types.extend([".mp4", ".mov", ".avi"])
|
158 |
+
|
159 |
+
if config["supports_pdf"]:
|
160 |
+
base_types.append(".pdf")
|
161 |
+
|
162 |
+
return base_types
|
163 |
+
|
164 |
@spaces.GPU(duration=120)
|
165 |
def run(
|
166 |
message: dict,
|
167 |
history: list[dict],
|
168 |
+
model_name: str,
|
169 |
system_prompt: str,
|
170 |
max_new_tokens: int,
|
171 |
max_images: int,
|
|
|
174 |
top_k: int,
|
175 |
repetition_penalty: float,
|
176 |
) -> Iterator[str]:
|
177 |
+
|
178 |
+
global current_model
|
179 |
+
|
180 |
+
if model_name != current_model:
|
181 |
+
current_model = model_name
|
182 |
+
logger.info(f"Switched to model: {model_name}")
|
183 |
+
|
184 |
logger.debug(
|
185 |
+
f"\n message: {message} \n history: {history} \n model: {model_name} \n "
|
186 |
+
f"system_prompt: {system_prompt} \n max_new_tokens: {max_new_tokens} \n max_images: {max_images}"
|
187 |
)
|
188 |
|
189 |
+
config = MODEL_CONFIGS[model_name]
|
190 |
+
if not config["supports_video"] and message.get("files"):
|
191 |
+
for file_path in message["files"]:
|
192 |
+
if file_path.endswith((".mp4", ".mov", ".avi")):
|
193 |
+
yield "Error: Selected model does not support video files. Please choose a video-capable model."
|
194 |
+
return
|
195 |
+
|
196 |
messages = []
|
197 |
if system_prompt:
|
198 |
messages.append(
|
|
|
203 |
{"role": "user", "content": process_user_input(message, max_images)}
|
204 |
)
|
205 |
|
206 |
+
inputs = processor.apply_chat_template(
|
207 |
messages,
|
208 |
add_generation_prompt=True,
|
209 |
tokenize=True,
|
210 |
return_dict=True,
|
211 |
return_tensors="pt",
|
212 |
+
).to(device=models[current_model].device, dtype=torch.bfloat16)
|
213 |
|
214 |
streamer = TextIteratorStreamer(
|
215 |
+
processor, timeout=60.0, skip_prompt=True, skip_special_tokens=True
|
216 |
)
|
217 |
generate_kwargs = dict(
|
218 |
inputs,
|
|
|
224 |
repetition_penalty=repetition_penalty,
|
225 |
do_sample=True,
|
226 |
)
|
227 |
+
t = Thread(target=models[current_model].generate, kwargs=generate_kwargs)
|
228 |
t.start()
|
229 |
|
230 |
output = ""
|
|
|
232 |
output += delta
|
233 |
yield output
|
234 |
|
235 |
+
def create_interface():
|
236 |
+
"""Create interface with model selector."""
|
237 |
+
|
238 |
+
initial_file_types = get_supported_file_types(current_model)
|
239 |
+
|
240 |
+
demo = gr.ChatInterface(
|
241 |
+
fn=run,
|
242 |
+
type="messages",
|
243 |
+
chatbot=gr.Chatbot(type="messages", scale=1, allow_tags=["image"]),
|
244 |
+
textbox=gr.MultimodalTextbox(
|
245 |
+
file_types=initial_file_types,
|
246 |
+
file_count="multiple",
|
247 |
+
autofocus=True
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
248 |
),
|
249 |
+
multimodal=True,
|
250 |
+
additional_inputs=[
|
251 |
+
gr.Dropdown(
|
252 |
+
label="Model",
|
253 |
+
choices=list(MODEL_CONFIGS.keys()),
|
254 |
+
value=current_model,
|
255 |
+
info="Select which model to use for generation"
|
256 |
+
),
|
257 |
+
gr.Textbox(label="System Prompt", value="You are a helpful assistant."),
|
258 |
+
gr.Slider(
|
259 |
+
label="Max New Tokens", minimum=100, maximum=2000, step=10, value=700
|
260 |
+
),
|
261 |
+
gr.Slider(label="Max Images", minimum=1, maximum=8, step=1, value=2),
|
262 |
+
gr.Slider(
|
263 |
+
label="Temperature", minimum=0.1, maximum=2.0, step=0.1, value=0.7
|
264 |
+
),
|
265 |
+
gr.Slider(
|
266 |
+
label="Top P", minimum=0.1, maximum=1.0, step=0.05, value=0.9
|
267 |
+
),
|
268 |
+
gr.Slider(
|
269 |
+
label="Top K", minimum=1, maximum=100, step=1, value=50
|
270 |
+
),
|
271 |
+
gr.Slider(
|
272 |
+
label="Repetition Penalty", minimum=1.0, maximum=2.0, step=0.05, value=1.1
|
273 |
+
),
|
274 |
+
],
|
275 |
+
stop_btn=False,
|
276 |
+
title="Multi-Model Gemma Chat"
|
277 |
+
)
|
278 |
+
|
279 |
+
return demo
|
280 |
+
|
281 |
+
demo = create_interface()
|
282 |
|
283 |
if __name__ == "__main__":
|
284 |
demo.launch()
|