Fix Loss in Gutenberg bonus section (#109)

ch05/01_main-chapter-code/ch05.ipynb

@@ -1081,10 +1081,8 @@
    "source": [
     "def calc_loss_batch(input_batch, target_batch, model, device):\n",
     "    input_batch, target_batch = input_batch.to(device), target_batch.to(device)\n",
-    "\n",
     "    logits = model(input_batch)\n",
-    "    logits = logits.flatten(0, 1)\n",
+    "    loss = torch.nn.functional.cross_entropy(logits.flatten(0, 1), target_batch.flatten())\n",
-    "    loss = torch.nn.functional.cross_entropy(logits, target_batch.flatten())\n",
     "    return loss\n",
     "\n",
     "\n",
@@ -2403,7 +2401,7 @@
    "name": "python",
    "nbconvert_exporter": "python",
    "pygments_lexer": "ipython3",
-   "version": "3.10.6"
+   "version": "3.11.4"
   }
  },
  "nbformat": 4,

ch05/03_bonus_pretraining_on_gutenberg/pretraining_simple.py

@@ -164,18 +164,32 @@ if __name__ == "__main__":
                         help='Learning rate for the optimizer')
     parser.add_argument('--batch_size', type=int, default=4,
                         help='Batch size for training')
+    parser.add_argument('--debug', type=bool, default=False,
+                        help='Uses a very small model for debugging purposes')
 
     args = parser.parse_args()
 
-    GPT_CONFIG_124M = {
+    if args.debug:
-        "vocab_size": 50257,     # Vocabulary size
+        GPT_CONFIG_124M = {
-        "context_length": 1024,  # Context length
+            "vocab_size": 50257,     # Vocabulary size
-        "emb_dim": 768,          # Embedding dimension
+            "context_length": 10,    # Context length
-        "n_heads": 12,           # Number of attention heads
+            "emb_dim": 12,           # Embedding dimension
-        "n_layers": 12,          # Number of layers
+            "n_heads": 2,            # Number of attention heads
-        "drop_rate": 0.1,        # Dropout rate
+            "n_layers": 2,           # Number of layers
-        "qkv_bias": False        # Query-key-value bias
+            "drop_rate": 0.0,        # Dropout rate
-    }
+            "qkv_bias": False        # Query-key-value bias
+        }
+
+    else:
+        GPT_CONFIG_124M = {
+            "vocab_size": 50257,     # Vocabulary size
+            "context_length": 1024,  # Context length
+            "emb_dim": 768,          # Embedding dimension
+            "n_heads": 12,           # Number of attention heads
+            "n_layers": 12,          # Number of layers
+            "drop_rate": 0.1,        # Dropout rate
+            "qkv_bias": False        # Query-key-value bias
+        }
 
     device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
     torch.manual_seed(123)
@@ -210,8 +224,6 @@ if __name__ == "__main__":
     )
 
     epochs_tensor = torch.linspace(0, args.n_epochs, len(train_losses))
-
-    print("debug", epochs_tensor, tokens_seen, train_losses, val_losses, output_dir)
     plot_losses(epochs_tensor, tokens_seen, train_losses, val_losses, output_dir)
 
     torch.save(model.state_dict(), output_dir / "model_pg_final.pth")

ch05/03_bonus_pretraining_on_gutenberg/previous_chapters.py

@@ -244,7 +244,7 @@ def generate_text_simple(model, idx, max_new_tokens, context_size):
 def calc_loss_batch(input_batch, target_batch, model, device):
     input_batch, target_batch = input_batch.to(device), target_batch.to(device)
     logits = model(input_batch)
-    loss = torch.nn.functional.cross_entropy(logits.flatten(0, -1), target_batch.flatten())
+    loss = torch.nn.functional.cross_entropy(logits.flatten(0, 1), target_batch.flatten())
     return loss
 
 

ch05/03_bonus_pretraining_on_gutenberg/tests.py

@@ -0,0 +1,32 @@
+# Copyright (c) Sebastian Raschka under Apache License 2.0 (see LICENSE.txt).
+# Source for "Build a Large Language Model From Scratch"
+#   - https://www.manning.com/books/build-a-large-language-model-from-scratch
+# Code: https://github.com/rasbt/LLMs-from-scratch
+
+# File for internal use (unit tests)
+
+from pathlib import Path
+import os
+import subprocess
+
+
+def test_pretraining():
+
+    sequence = "a b c d"
+    repetitions = 1000
+    content = sequence * repetitions
+
+    folder_path = Path("gutenberg") / "data"
+    file_name = "repeated_sequence.txt"
+
+    os.makedirs(folder_path, exist_ok=True)
+
+    with open(folder_path/file_name, "w") as file:
+        file.write(content)
+
+    result = subprocess.run(
+        ["python", "pretraining_simple.py", "--debug", "true"],
+        capture_output=True, text=True
+    )
+    print(result.stdout)
+    assert "Maximum GPU memory allocated" in result.stdout