tweak hyperparams

hyperparams.py

@@ -2,4 +2,4 @@ FILTER_COUNT = 32
 KERNEL_SIZE = 2
 CROP_SIZE = 200
 BATCH_SIZE = 4
-EPOCHS = 15
+EPOCHS = 20

inference.py

@@ -1,27 +1,12 @@
-import modal
 import torch
 import numpy as np
 from PIL import Image
 from model import mai
 from augmentation import preprocess_validation
 
-MODEL_NAME = "mai_20240424_180855_4"
 
-image = modal.Image.debian_slim().pip_install(
-    "datasets==2.19.0",
-    "albumentations==1.4.4",
-    "numpy==1.26.4",
-    "torch==2.2.2",
-)
-app = modal.App("multilayer-authenticity-identifier", image=image)
-volume = modal.Volume.from_name("model-store")
-model_store_path = "/vol/models"
-
-
-@app.function(timeout=5000, gpu="T4", volumes={model_store_path: volume})
 def load_model_and_run_inference(img):
-    print(f"REMOTE: {img.shape}")
-    mai.load_state_dict(torch.load(f"{model_store_path}/{MODEL_NAME}"))
+    mai.load_state_dict(torch.load("mai"))
     mai.eval()
     img_batch = np.expand_dims(img, axis=0)
     img_batch = torch.tensor(img_batch)
@@ -30,9 +15,7 @@ def load_model_and_run_inference(img):
     print(prediction)
 
 
-@app.local_entrypoint()
 def main():
     img = Image.open("test_images/dog.jpg")
     img = preprocess_validation(image=np.array(img))["image"]
-    print(img.shape)
-    load_model_and_run_inference.remote(img)
+    load_model_and_run_inference(img)