Cleaning up more.

[mygpt.git] / main.py

diff --git a/main.py b/main.py
index 76aeebd..339d185 100755 (executable)
--- a/main.py
+++ b/main.py
@@ -18,7 +18,6 @@ import mygpt
 device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
 
 ######################################################################
-
 parser = argparse.ArgumentParser(description = 'My own GPT.')
 
 parser.add_argument('--log_filename',
@@ -148,7 +147,7 @@ class Task:
     def vocabulary_size(self):
         pass
 
-    def produce_results(self, n_epoch, model, nb_tokens = 50):
+    def produce_results(self, n_epoch, model):
         pass
 
 ######################################################################
@@ -202,31 +201,27 @@ class TaskPicoCLVR(Task):
 
     def batches(self, split = 'train'):
         assert split in { 'train', 'test' }
-        if split == 'train':
-            for batch in tqdm.tqdm(self.train_input.split(self.batch_size), desc = f'epoch-{split}'):
-                yield batch
-        else:
-            for batch in tqdm.tqdm(self.test_input.split(self.batch_size), desc = f'epoch-{split}'):
-                yield batch
+        input = self.train_input if split == 'train' else self.test_input
+        for batch in tqdm.tqdm(input.split(self.batch_size), desc = f'epoch-{split}'):
+            yield batch
 
     def vocabulary_size(self):
         return len(self.token2id)
 
-    def generate(self, descr_primer, model, nb_tokens):
+    def generate(self, primer_descr, model, nb_tokens):
         results = autoregression(
             model, self.batch_size,
-            1, nb_tokens, primer = descr2tensor(descr_primer),
+            nb_samples = 1, nb_tokens = nb_tokens, primer = descr2tensor(primer_descr),
             device = self.device
         )
         return ' '.join([ self.id2token[t.item()] for t in results.flatten() ])
 
-    def produce_results(self, n_epoch, model, nb_tokens = None):
-        if nb_tokens is None:
-            nb_tokens = self.height * self.width + 3
+    def produce_results(self, n_epoch, model):
+        nb_tokens = self.height * self.width + 3
         result_descr = [ ]
         nb_per_primer = 8
 
-        for descr_primer in [
+        for primer_descr in [
                 'red above green <sep> green top <sep> blue right of red <img>',
                 'there is red <sep> there is yellow <sep> there is blue <img>',
                 'red below yellow <sep> yellow below green <sep> green below blue <sep> red right <sep> yellow left <sep> green right <sep> blue left <img>',
@@ -234,7 +229,7 @@ class TaskPicoCLVR(Task):
         ]:
 
             for k in range(nb_per_primer):
-                result_descr.append(self.generate(descr_primer, model, nb_tokens))
+                result_descr.append(self.generate(primer_descr, model, nb_tokens))
 
         img = [ picoclvr.descr2img(d, height = self.height, width = self.width)
                 for d in result_descr ]
@@ -316,7 +311,8 @@ class TaskWiki103(Task):
     def vocabulary_size(self):
         return len(self.vocab)
 
-    def produce_results(self, n_epoch, model, nb_tokens = 50):
+    def produce_results(self, n_epoch, model):
+        nb_tokens = 50
         file_name = f'result_wiki103_{n_epoch:04d}.txt'
 
         with open(file_name, 'w') as outfile:
@@ -377,7 +373,8 @@ class TaskMNIST(Task):
     def vocabulary_size(self):
         return 256
 
-    def produce_results(self, n_epoch, model, nb_samples = 64):
+    def produce_results(self, n_epoch, model):
+        nb_samples = 64
         results = autoregression(model, self.batch_size, nb_samples, 28 * 28, device = self.device)
         image_name = f'result_mnist_{n_epoch:04d}.png'
         torchvision.utils.save_image(1 - results.reshape(-1, 1, 28, 28) / 255.,