Spaces:
Sleeping
Sleeping
Upload folder using huggingface_hub
Browse files- app.py +89 -180
- app_moe.py +13 -3
app.py
CHANGED
@@ -34,11 +34,10 @@ class WatermelonMoEModel(torch.nn.Module):
|
|
34 |
weights: Optional list of weights for each model (None for equal weighting)
|
35 |
"""
|
36 |
super(WatermelonMoEModel, self).__init__()
|
37 |
-
self.models =
|
38 |
self.model_configs = model_configs
|
39 |
|
40 |
# Load each model
|
41 |
-
loaded_count = 0
|
42 |
for config in model_configs:
|
43 |
img_backbone = config["image_backbone"]
|
44 |
audio_backbone = config["audio_backbone"]
|
@@ -50,31 +49,22 @@ class WatermelonMoEModel(torch.nn.Module):
|
|
50 |
model_path = os.path.join(model_dir, f"{img_backbone}_{audio_backbone}_model.pt")
|
51 |
if os.path.exists(model_path):
|
52 |
print(f"\033[92mINFO\033[0m: Loading model {img_backbone}_{audio_backbone} from {model_path}")
|
53 |
-
|
54 |
-
model.load_state_dict(torch.load(model_path, map_location='cpu'))
|
55 |
-
model.eval() # Set to evaluation mode
|
56 |
-
self.models.append(model)
|
57 |
-
loaded_count += 1
|
58 |
-
except Exception as e:
|
59 |
-
print(f"\033[91mERR!\033[0m: Failed to load model from {model_path}: {e}")
|
60 |
-
continue
|
61 |
else:
|
62 |
print(f"\033[91mERR!\033[0m: Model checkpoint not found at {model_path}")
|
63 |
continue
|
|
|
|
|
|
|
64 |
|
65 |
-
# Add a dummy parameter if no models were loaded to prevent StopIteration
|
66 |
-
if loaded_count == 0:
|
67 |
-
print(f"\033[91mERR!\033[0m: No models were successfully loaded!")
|
68 |
-
self.dummy_param = torch.nn.Parameter(torch.zeros(1))
|
69 |
-
|
70 |
# Set model weights (uniform by default)
|
71 |
-
if weights
|
72 |
assert len(weights) == len(self.models), "Number of weights must match number of models"
|
73 |
self.weights = weights
|
74 |
else:
|
75 |
-
self.weights = [1.0 /
|
76 |
|
77 |
-
print(f"\033[92mINFO\033[0m: Loaded {
|
78 |
print(f"\033[92mINFO\033[0m: Model weights: {self.weights}")
|
79 |
|
80 |
def to(self, device):
|
@@ -90,10 +80,9 @@ class WatermelonMoEModel(torch.nn.Module):
|
|
90 |
Forward pass through the MoE model.
|
91 |
Returns the weighted average of all model outputs.
|
92 |
"""
|
93 |
-
# Check if we have models loaded
|
94 |
if not self.models:
|
95 |
print(f"\033[91mERR!\033[0m: No models available for inference!")
|
96 |
-
return torch.tensor([0.0], device=mfcc.device)
|
97 |
|
98 |
outputs = []
|
99 |
|
@@ -101,6 +90,8 @@ class WatermelonMoEModel(torch.nn.Module):
|
|
101 |
with torch.no_grad():
|
102 |
for i, model in enumerate(self.models):
|
103 |
output = model(mfcc, image)
|
|
|
|
|
104 |
outputs.append(output * self.weights[i])
|
105 |
|
106 |
# Return weighted average
|
@@ -166,196 +157,114 @@ def predict_sugar_content(audio, image, model_dir="models", weights=None):
|
|
166 |
"""Function with GPU acceleration to predict watermelon sugar content in Brix using MoE model"""
|
167 |
try:
|
168 |
# Check CUDA availability inside the GPU-decorated function
|
169 |
-
if torch.cuda.is_available()
|
170 |
-
|
171 |
-
print(f"\033[92mINFO\033[0m: CUDA is available. Using device: {device}")
|
172 |
-
else:
|
173 |
-
device = torch.device("cpu")
|
174 |
-
print(f"\033[92mINFO\033[0m: CUDA is not available. Using device: {device}")
|
175 |
|
176 |
# Load MoE model
|
177 |
moe_model = WatermelonMoEModel(TOP_MODELS, model_dir, weights)
|
178 |
-
|
179 |
-
moe_model = moe_model.to(device)
|
180 |
moe_model.eval()
|
181 |
print(f"\033[92mINFO\033[0m: Loaded MoE model with {len(moe_model.models)} backbone models")
|
182 |
|
183 |
-
# Debug information about input types
|
184 |
-
print(f"\033[92mDEBUG\033[0m: Audio input type: {type(audio)}")
|
185 |
-
print(f"\033[92mDEBUG\033[0m: Audio input shape/length: {len(audio)}")
|
186 |
-
print(f"\033[92mDEBUG\033[0m: Image input type: {type(image)}")
|
187 |
-
if isinstance(image, np.ndarray):
|
188 |
-
print(f"\033[92mDEBUG\033[0m: Image input shape: {image.shape}")
|
189 |
-
|
190 |
# Handle different audio input formats
|
191 |
-
if isinstance(audio, tuple) and len(audio)
|
192 |
-
|
193 |
-
sample_rate, audio_data = audio
|
194 |
-
print(f"\033[92mDEBUG\033[0m: Audio sample rate: {sample_rate}")
|
195 |
-
print(f"\033[92mDEBUG\033[0m: Audio data shape: {audio_data.shape}")
|
196 |
-
elif isinstance(audio, tuple) and len(audio) > 2:
|
197 |
-
# Sometimes Gradio returns (sample_rate, audio_data, other_info...)
|
198 |
-
sample_rate, audio_data = audio[0], audio[-1]
|
199 |
-
print(f"\033[92mDEBUG\033[0m: Audio sample rate: {sample_rate}")
|
200 |
-
print(f"\033[92mDEBUG\033[0m: Audio data shape: {audio_data.shape}")
|
201 |
elif isinstance(audio, str):
|
202 |
-
# Direct path to audio file
|
203 |
audio_data, sample_rate = torchaudio.load(audio)
|
204 |
-
print(f"\033[92mDEBUG\033[0m: Loaded audio from path with shape: {audio_data.shape}")
|
205 |
else:
|
206 |
return f"Error: Unsupported audio format. Got {type(audio)}"
|
207 |
|
208 |
-
#
|
209 |
-
temp_dir = "temp"
|
210 |
-
os.makedirs(temp_dir, exist_ok=True)
|
211 |
-
|
212 |
-
temp_audio_path = os.path.join(temp_dir, "temp_audio.wav")
|
213 |
-
temp_image_path = os.path.join(temp_dir, "temp_image.jpg")
|
214 |
-
|
215 |
-
# Import necessary libraries
|
216 |
-
from PIL import Image
|
217 |
-
|
218 |
-
# Audio handling - direct processing from the data in memory
|
219 |
if isinstance(audio_data, np.ndarray):
|
220 |
-
# Convert numpy array to tensor
|
221 |
-
print(f"\033[92mDEBUG\033[0m: Converting numpy audio with shape {audio_data.shape} to tensor")
|
222 |
audio_tensor = torch.tensor(audio_data).float()
|
223 |
-
|
224 |
-
# Handle different audio dimensions
|
225 |
-
if audio_data.ndim == 1:
|
226 |
-
# Single channel audio
|
227 |
-
audio_tensor = audio_tensor.unsqueeze(0)
|
228 |
-
elif audio_data.ndim == 2:
|
229 |
-
# Ensure channels are first dimension
|
230 |
-
if audio_data.shape[0] > audio_data.shape[1]:
|
231 |
-
# More rows than columns, probably (samples, channels)
|
232 |
-
audio_tensor = torch.tensor(audio_data.T).float()
|
233 |
else:
|
234 |
-
# Already a tensor
|
235 |
audio_tensor = audio_data.float()
|
236 |
|
237 |
-
|
238 |
-
|
239 |
-
# Skip saving/loading and process directly
|
240 |
mfcc = app_process_audio_data(audio_tensor, sample_rate)
|
241 |
-
|
|
|
242 |
|
243 |
-
#
|
244 |
if isinstance(image, np.ndarray):
|
245 |
-
|
246 |
-
pil_image = Image.fromarray(image)
|
247 |
-
pil_image.save(temp_image_path)
|
248 |
-
print(f"\033[92mDEBUG\033[0m: Saved image to {temp_image_path}")
|
249 |
elif isinstance(image, str):
|
250 |
-
|
251 |
-
temp_image_path = image
|
252 |
-
print(f"\033[92mDEBUG\033[0m: Using provided image path: {temp_image_path}")
|
253 |
else:
|
254 |
return f"Error: Unsupported image format. Got {type(image)}"
|
255 |
|
256 |
-
# Process image
|
257 |
-
print(f"\033[92mDEBUG\033[0m: Loading and preprocessing image from {temp_image_path}")
|
258 |
-
image_tensor = torchvision.io.read_image(temp_image_path)
|
259 |
-
print(f"\033[92mDEBUG\033[0m: Loaded image shape: {image_tensor.shape}")
|
260 |
image_tensor = image_tensor.float()
|
261 |
processed_image = process_image_data(image_tensor)
|
262 |
-
|
263 |
-
|
264 |
-
# Add batch dimension for inference and move to device
|
265 |
-
if mfcc is not None:
|
266 |
-
# Ensure mfcc is on the same device as the model
|
267 |
-
mfcc = mfcc.unsqueeze(0).to(device)
|
268 |
-
print(f"\033[92mDEBUG\033[0m: Final MFCC shape with batch dimension: {mfcc.shape}, device: {mfcc.device}")
|
269 |
-
|
270 |
-
if processed_image is not None:
|
271 |
-
# Ensure processed_image is on the same device as the model
|
272 |
-
processed_image = processed_image.unsqueeze(0).to(device)
|
273 |
-
print(f"\033[92mDEBUG\033[0m: Final image shape with batch dimension: {processed_image.shape}, device: {processed_image.device}")
|
274 |
-
|
275 |
-
# Double-check model is on the correct device
|
276 |
-
try:
|
277 |
-
param = next(moe_model.parameters())
|
278 |
-
print(f"\033[92mDEBUG\033[0m: MoE model device: {param.device}")
|
279 |
-
|
280 |
-
# Check individual models
|
281 |
-
for i, model in enumerate(moe_model.models):
|
282 |
-
try:
|
283 |
-
model_param = next(model.parameters())
|
284 |
-
print(f"\033[92mDEBUG\033[0m: Model {i} device: {model_param.device}")
|
285 |
-
except StopIteration:
|
286 |
-
print(f"\033[91mERR!\033[0m: Model {i} has no parameters!")
|
287 |
-
except StopIteration:
|
288 |
-
print(f"\033[91mERR!\033[0m: MoE model has no parameters!")
|
289 |
-
|
290 |
-
# Run inference with MoE model
|
291 |
-
print(f"\033[92mDEBUG\033[0m: Running inference with MoE model on device: {device}")
|
292 |
-
if mfcc is not None and processed_image is not None:
|
293 |
-
with torch.no_grad():
|
294 |
-
brix_value = moe_model(mfcc, processed_image)
|
295 |
-
print(f"\033[92mDEBUG\033[0m: Prediction successful: {brix_value.item()}")
|
296 |
-
else:
|
297 |
-
return "Error: Failed to process inputs. Please check the debug logs."
|
298 |
|
299 |
-
#
|
300 |
-
|
301 |
-
|
302 |
-
|
303 |
-
|
304 |
-
|
305 |
-
|
306 |
-
|
307 |
-
|
308 |
-
result += "- EfficientNet-B3 + Transformer\n"
|
309 |
-
result += "- EfficientNet-B0 + Transformer\n"
|
310 |
-
result += "- ResNet-50 + Transformer\n\n"
|
311 |
-
|
312 |
-
# Add Brix scale visualization
|
313 |
-
result += "Sugar Content Scale (in Β°Brix):\n"
|
314 |
-
result += "ββββββββββββββββββββββββββββββββββ\n"
|
315 |
-
|
316 |
-
# Create the scale display with Brix ranges
|
317 |
-
scale_ranges = [
|
318 |
-
(0, 8, "Low Sugar (< 8Β° Brix)"),
|
319 |
-
(8, 9, "Mild Sweetness (8-9Β° Brix)"),
|
320 |
-
(9, 10, "Medium Sweetness (9-10Β° Brix)"),
|
321 |
-
(10, 11, "Sweet (10-11Β° Brix)"),
|
322 |
-
(11, 13, "Very Sweet (11-13Β° Brix)")
|
323 |
-
]
|
324 |
-
|
325 |
-
# Find which category the prediction falls into
|
326 |
-
user_category = None
|
327 |
-
for min_val, max_val, category_name in scale_ranges:
|
328 |
-
if min_val <= brix_score < max_val:
|
329 |
-
user_category = category_name
|
330 |
-
break
|
331 |
-
if brix_score >= scale_ranges[-1][0]: # Handle edge case
|
332 |
-
user_category = scale_ranges[-1][2]
|
333 |
-
|
334 |
-
# Display the scale with the user's result highlighted
|
335 |
-
for min_val, max_val, category_name in scale_ranges:
|
336 |
-
if category_name == user_category:
|
337 |
-
result += f"βΆ {min_val}-{max_val}: {category_name} β (YOUR WATERMELON)\n"
|
338 |
-
else:
|
339 |
-
result += f" {min_val}-{max_val}: {category_name}\n"
|
340 |
-
|
341 |
-
result += "ββββββββββββββββββββββββββββββββββ\n\n"
|
342 |
|
343 |
-
#
|
344 |
-
|
345 |
-
|
346 |
-
|
347 |
-
|
348 |
-
|
349 |
-
|
350 |
-
|
351 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
352 |
else:
|
353 |
-
result += "
|
354 |
-
|
355 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
356 |
else:
|
357 |
-
|
358 |
-
|
|
|
359 |
except Exception as e:
|
360 |
import traceback
|
361 |
error_msg = f"Error: {str(e)}\n\n"
|
|
|
34 |
weights: Optional list of weights for each model (None for equal weighting)
|
35 |
"""
|
36 |
super(WatermelonMoEModel, self).__init__()
|
37 |
+
self.models = []
|
38 |
self.model_configs = model_configs
|
39 |
|
40 |
# Load each model
|
|
|
41 |
for config in model_configs:
|
42 |
img_backbone = config["image_backbone"]
|
43 |
audio_backbone = config["audio_backbone"]
|
|
|
49 |
model_path = os.path.join(model_dir, f"{img_backbone}_{audio_backbone}_model.pt")
|
50 |
if os.path.exists(model_path):
|
51 |
print(f"\033[92mINFO\033[0m: Loading model {img_backbone}_{audio_backbone} from {model_path}")
|
52 |
+
model.load_state_dict(torch.load(model_path, map_location='cpu'))
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
53 |
else:
|
54 |
print(f"\033[91mERR!\033[0m: Model checkpoint not found at {model_path}")
|
55 |
continue
|
56 |
+
|
57 |
+
model.eval() # Set to evaluation mode
|
58 |
+
self.models.append(model)
|
59 |
|
|
|
|
|
|
|
|
|
|
|
60 |
# Set model weights (uniform by default)
|
61 |
+
if weights:
|
62 |
assert len(weights) == len(self.models), "Number of weights must match number of models"
|
63 |
self.weights = weights
|
64 |
else:
|
65 |
+
self.weights = [1.0 / len(self.models)] * len(self.models) if self.models else [1.0]
|
66 |
|
67 |
+
print(f"\033[92mINFO\033[0m: Loaded {len(self.models)} models for MoE ensemble")
|
68 |
print(f"\033[92mINFO\033[0m: Model weights: {self.weights}")
|
69 |
|
70 |
def to(self, device):
|
|
|
80 |
Forward pass through the MoE model.
|
81 |
Returns the weighted average of all model outputs.
|
82 |
"""
|
|
|
83 |
if not self.models:
|
84 |
print(f"\033[91mERR!\033[0m: No models available for inference!")
|
85 |
+
return torch.tensor([0.0], device=mfcc.device)
|
86 |
|
87 |
outputs = []
|
88 |
|
|
|
90 |
with torch.no_grad():
|
91 |
for i, model in enumerate(self.models):
|
92 |
output = model(mfcc, image)
|
93 |
+
# print the output value
|
94 |
+
print(f"\033[92mDEBUG\033[0m: Model {i} output: {output}")
|
95 |
outputs.append(output * self.weights[i])
|
96 |
|
97 |
# Return weighted average
|
|
|
157 |
"""Function with GPU acceleration to predict watermelon sugar content in Brix using MoE model"""
|
158 |
try:
|
159 |
# Check CUDA availability inside the GPU-decorated function
|
160 |
+
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
161 |
+
print(f"\033[92mINFO\033[0m: Using device: {device}")
|
|
|
|
|
|
|
|
|
162 |
|
163 |
# Load MoE model
|
164 |
moe_model = WatermelonMoEModel(TOP_MODELS, model_dir, weights)
|
165 |
+
moe_model = moe_model.to(device) # Move entire model to device
|
|
|
166 |
moe_model.eval()
|
167 |
print(f"\033[92mINFO\033[0m: Loaded MoE model with {len(moe_model.models)} backbone models")
|
168 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
169 |
# Handle different audio input formats
|
170 |
+
if isinstance(audio, tuple) and len(audio) >= 2:
|
171 |
+
sample_rate, audio_data = audio[0], audio[1] if len(audio) == 2 else audio[-1]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
172 |
elif isinstance(audio, str):
|
|
|
173 |
audio_data, sample_rate = torchaudio.load(audio)
|
|
|
174 |
else:
|
175 |
return f"Error: Unsupported audio format. Got {type(audio)}"
|
176 |
|
177 |
+
# Convert audio to tensor if needed
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
178 |
if isinstance(audio_data, np.ndarray):
|
|
|
|
|
179 |
audio_tensor = torch.tensor(audio_data).float()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
180 |
else:
|
|
|
181 |
audio_tensor = audio_data.float()
|
182 |
|
183 |
+
# Process audio
|
|
|
|
|
184 |
mfcc = app_process_audio_data(audio_tensor, sample_rate)
|
185 |
+
if mfcc is None:
|
186 |
+
return "Error: Failed to process audio input"
|
187 |
|
188 |
+
# Process image
|
189 |
if isinstance(image, np.ndarray):
|
190 |
+
image_tensor = torch.from_numpy(image).permute(2, 0, 1) # Convert to CxHxW format
|
|
|
|
|
|
|
191 |
elif isinstance(image, str):
|
192 |
+
image_tensor = torchvision.io.read_image(image)
|
|
|
|
|
193 |
else:
|
194 |
return f"Error: Unsupported image format. Got {type(image)}"
|
195 |
|
|
|
|
|
|
|
|
|
196 |
image_tensor = image_tensor.float()
|
197 |
processed_image = process_image_data(image_tensor)
|
198 |
+
if processed_image is None:
|
199 |
+
return "Error: Failed to process image input"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
200 |
|
201 |
+
# Add batch dimension and move to device
|
202 |
+
mfcc = mfcc.unsqueeze(0).to(device)
|
203 |
+
processed_image = processed_image.unsqueeze(0).to(device)
|
204 |
+
|
205 |
+
# Run inference
|
206 |
+
with torch.no_grad():
|
207 |
+
brix_value = moe_model(mfcc, processed_image)
|
208 |
+
prediction = brix_value.item()
|
209 |
+
print(f"\033[92mDEBUG\033[0m: Raw prediction: {prediction}")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
210 |
|
211 |
+
# Ensure prediction is within reasonable bounds (e.g., 6-13 Brix)
|
212 |
+
prediction = max(6.0, min(13.0, prediction))
|
213 |
+
print(f"\033[92mDEBUG\033[0m: Bounded prediction: {prediction}")
|
214 |
+
|
215 |
+
# Format the result
|
216 |
+
result = f"π Predicted Sugar Content: {prediction:.1f}Β° Brix π\n\n"
|
217 |
+
|
218 |
+
# Add extra info about the MoE model
|
219 |
+
result += "Using Ensemble of Top-3 Models:\n"
|
220 |
+
result += "- EfficientNet-B3 + Transformer\n"
|
221 |
+
result += "- EfficientNet-B0 + Transformer\n"
|
222 |
+
result += "- ResNet-50 + Transformer\n\n"
|
223 |
+
|
224 |
+
# Add Brix scale visualization
|
225 |
+
result += "Sugar Content Scale (in Β°Brix):\n"
|
226 |
+
result += "ββββββββββββββββββββββββββββββββββ\n"
|
227 |
+
|
228 |
+
# Create the scale display with Brix ranges
|
229 |
+
scale_ranges = [
|
230 |
+
(0, 8, "Low Sugar (< 8Β° Brix)"),
|
231 |
+
(8, 9, "Mild Sweetness (8-9Β° Brix)"),
|
232 |
+
(9, 10, "Medium Sweetness (9-10Β° Brix)"),
|
233 |
+
(10, 11, "Sweet (10-11Β° Brix)"),
|
234 |
+
(11, 13, "Very Sweet (11-13Β° Brix)")
|
235 |
+
]
|
236 |
+
|
237 |
+
# Find which category the prediction falls into
|
238 |
+
user_category = None
|
239 |
+
for min_val, max_val, category_name in scale_ranges:
|
240 |
+
if min_val <= prediction < max_val:
|
241 |
+
user_category = category_name
|
242 |
+
break
|
243 |
+
if prediction >= scale_ranges[-1][0]: # Handle edge case
|
244 |
+
user_category = scale_ranges[-1][2]
|
245 |
+
|
246 |
+
# Display the scale with the user's result highlighted
|
247 |
+
for min_val, max_val, category_name in scale_ranges:
|
248 |
+
if category_name == user_category:
|
249 |
+
result += f"βΆ {min_val}-{max_val}: {category_name} β (YOUR WATERMELON)\n"
|
250 |
else:
|
251 |
+
result += f" {min_val}-{max_val}: {category_name}\n"
|
252 |
+
|
253 |
+
result += "ββββββββββββββββββββββββββββββββββ\n\n"
|
254 |
+
|
255 |
+
# Add assessment of the watermelon's sugar content
|
256 |
+
if prediction < 8:
|
257 |
+
result += "Assessment: This watermelon has low sugar content. It may taste bland or slightly bitter."
|
258 |
+
elif prediction < 9:
|
259 |
+
result += "Assessment: This watermelon has mild sweetness. Acceptable flavor but not very sweet."
|
260 |
+
elif prediction < 10:
|
261 |
+
result += "Assessment: This watermelon has moderate sugar content. It should have pleasant sweetness."
|
262 |
+
elif prediction < 11:
|
263 |
+
result += "Assessment: This watermelon has good sugar content! It should be sweet and juicy."
|
264 |
else:
|
265 |
+
result += "Assessment: This watermelon has excellent sugar content! Perfect choice for maximum sweetness and flavor."
|
266 |
+
|
267 |
+
return result
|
268 |
except Exception as e:
|
269 |
import traceback
|
270 |
error_msg = f"Error: {str(e)}\n\n"
|
app_moe.py
CHANGED
@@ -273,9 +273,19 @@ def predict_sugar_content(audio, image, model_dir="models", weights=None):
|
|
273 |
print(f"\033[92mDEBUG\033[0m: Final image shape with batch dimension: {processed_image.shape}, device: {processed_image.device}")
|
274 |
|
275 |
# Double-check model is on the correct device
|
276 |
-
|
277 |
-
|
278 |
-
print(f"\033[92mDEBUG\033[0m:
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
279 |
|
280 |
# Run inference with MoE model
|
281 |
print(f"\033[92mDEBUG\033[0m: Running inference with MoE model on device: {device}")
|
|
|
273 |
print(f"\033[92mDEBUG\033[0m: Final image shape with batch dimension: {processed_image.shape}, device: {processed_image.device}")
|
274 |
|
275 |
# Double-check model is on the correct device
|
276 |
+
try:
|
277 |
+
param = next(moe_model.parameters())
|
278 |
+
print(f"\033[92mDEBUG\033[0m: MoE model device: {param.device}")
|
279 |
+
|
280 |
+
# Check individual models
|
281 |
+
for i, model in enumerate(moe_model.models):
|
282 |
+
try:
|
283 |
+
model_param = next(model.parameters())
|
284 |
+
print(f"\033[92mDEBUG\033[0m: Model {i} device: {model_param.device}")
|
285 |
+
except StopIteration:
|
286 |
+
print(f"\033[91mERR!\033[0m: Model {i} has no parameters!")
|
287 |
+
except StopIteration:
|
288 |
+
print(f"\033[91mERR!\033[0m: MoE model has no parameters!")
|
289 |
|
290 |
# Run inference with MoE model
|
291 |
print(f"\033[92mDEBUG\033[0m: Running inference with MoE model on device: {device}")
|