update

Author: gauravsh0812

README.md

@@ -1,4 +1,4 @@
-# CleverMath_LLM
+# ClevrMath_LLM
 
 Keep preprocess in the config "True" if running it for the first time. 
 Once you have run it, there is no need to re-process the dataset. 
@@ -8,6 +8,9 @@ re-do these steps. This can avoided by setting the preprocess param to False in
 
 ### Requirements 
 ```
+conda create -n clevrmath python=3.10 -y
+source activate clevrmath
+
 pip install torch==2.0.0 torchvision==0.15.1 torchaudio==2.0.1 --index-url https://download.pytorch.org/whl/cu118
 pip install -r requirements.txt
 ```

config/config.yaml

@@ -19,7 +19,7 @@ dataset:
 training:
   general:
     clip: 1 
-    batch_size: 32
+    batch_size: 4
     epochs: 1
     dropout: 0.1
     learning_rate: 0.0001
@@ -30,4 +30,4 @@ training:
     scheduler_gamma: 0.5
 
   unet_encoder: 
-    input_channels: 3
+    input_channels: 4

preprocessing/__pycache__/create_dataloaders.cpython-39.pyc

@@ -72,10 +72,8 @@ def data_loaders():
     t = open("data/templates.lst").readlines()
 
     assert len(q) == len(l) == len(t)
-
-    max_len = max([len(i.split()) for i in q])
-
-    image_num = range(0, len(q))
+    
+    image_num = range(0, 20)#len(q))
 
     # split the image_num into train, test, validate
     train_val_images, test_images = train_test_split(

preprocessing/create_dataloaders.py

@@ -3,7 +3,7 @@ wandb==0.16.0
 torchtext==0.6.0
 pyyaml==6.0.1
 Pillow==10.1.0
-numpy==1.19.3
+numpy==1.21.0
 pandas==1.4.4
 opencv-python==4.8.1.78
 opencv-contrib-python==4.8.1.78

requirements.txt

@@ -32,7 +32,6 @@ def train(
         _imgs = list()
         for im in imgs:
             tnsr = torch.load(f"data/image_tensors/{int(im.item())}.pt")
-            print("imgs shape: ", tnsr.shape)
             _imgs.append(tnsr)