registering
Browse files
app.py
CHANGED
@@ -11,7 +11,7 @@ import sys
|
|
11 |
# import timm
|
12 |
from types import SimpleNamespace
|
13 |
# from transformers import AutoModel, pipeline
|
14 |
-
from transformers import AutoModelForImageClassification, AutoModel
|
15 |
import torch
|
16 |
|
17 |
sys.path.insert(1, "../")
|
@@ -254,6 +254,17 @@ def predict_and_analyze(model_name, num_channels, dim, input_channel, image):
|
|
254 |
EfficientNetConfig.model_type = hparams.model_type
|
255 |
|
256 |
config.save_pretrained(save_directory=model_loading_name)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
257 |
# config = EfficientNetConfig.from_pretrained(model_loading_name)
|
258 |
|
259 |
# model = EfficientNetPreTrained.from_pretrained(model_loading_name)
|
|
|
11 |
# import timm
|
12 |
from types import SimpleNamespace
|
13 |
# from transformers import AutoModel, pipeline
|
14 |
+
from transformers import AutoModelForImageClassification, AutoModel, AutoConfig
|
15 |
import torch
|
16 |
|
17 |
sys.path.insert(1, "../")
|
|
|
254 |
EfficientNetConfig.model_type = hparams.model_type
|
255 |
|
256 |
config.save_pretrained(save_directory=model_loading_name)
|
257 |
+
|
258 |
+
model = EfficientNet(dropout=hparams.dropout,
|
259 |
+
num_channels=hparams.num_channels,
|
260 |
+
num_classes=hparams.num_classes,
|
261 |
+
size=hparams.size,
|
262 |
+
stochastic_depth_prob=hparams.stochastic_depth_prob,
|
263 |
+
width_mult=hparams.width_mult,
|
264 |
+
depth_mult=hparams.depth_mult,)
|
265 |
+
|
266 |
+
AutoConfig.register(model_loading_name, config)
|
267 |
+
AutoModel.register(config, model)
|
268 |
# config = EfficientNetConfig.from_pretrained(model_loading_name)
|
269 |
|
270 |
# model = EfficientNetPreTrained.from_pretrained(model_loading_name)
|