User argpars utils to show default args on command line

ch04/04_gqa/gpt_with_kv_gqa.py

@@ -290,7 +290,7 @@ def generate_text_simple_cached(model, idx, max_new_tokens,
 
 
 def main():
-    parser = argparse.ArgumentParser(description="Run GPT with grouped-query attention.")
+    parser = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter, description="Run GPT with grouped-query attention.")
     parser.add_argument("--emb_dim", type=int, default=768, help="Model embedding dimension.")
     parser.add_argument("--n_heads", type=int, default=12, help="Number of attention heads.")
     parser.add_argument("--n_layers", type=int, default=12, help="Number of transformer blocks.")

ch04/04_gqa/gpt_with_kv_mha.py

@@ -278,7 +278,7 @@ def generate_text_simple_cached(model, idx, max_new_tokens,
 
 
 def main():
-    parser = argparse.ArgumentParser(description="Run GPT with standard multi-head attention.")
+    parser = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter, description="Run GPT with standard multi-head attention.")
     parser.add_argument("--emb_dim", type=int, default=768, help="Model embedding dimension.")
     parser.add_argument("--n_heads", type=int, default=12, help="Number of attention heads.")
     parser.add_argument("--n_layers", type=int, default=12, help="Number of transformer blocks.")

ch04/04_gqa/memory_estimator_gqa.py

@@ -31,7 +31,7 @@ def calc_kv_bytes_total(batch, context_length, emb_dim, n_heads,
 
 
 def main():
-    p = argparse.ArgumentParser(description="Estimate KV-cache memory for MHA vs GQA")
+    p = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter, description="Estimate KV-cache memory for MHA vs GQA")
     p.add_argument("--context_length", default=1024, type=int)
     p.add_argument("--emb_dim", required=True, type=int)
     p.add_argument("--n_heads", required=True, type=int)

ch04/05_mla/gpt_with_kv_mha.py

@@ -278,7 +278,7 @@ def generate_text_simple_cached(model, idx, max_new_tokens,
 
 
 def main():
-    parser = argparse.ArgumentParser(description="Run GPT with standard multi-head attention.")
+    parser = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter, description="Run GPT with standard multi-head attention.")
     parser.add_argument("--emb_dim", type=int, default=768, help="Model embedding dimension.")
     parser.add_argument("--n_heads", type=int, default=12, help="Number of attention heads.")
     parser.add_argument("--n_layers", type=int, default=12, help="Number of transformer blocks.")

ch04/05_mla/gpt_with_kv_mla.py

@@ -286,13 +286,13 @@ def generate_text_simple_cached(model, idx, max_new_tokens,
 
 
 def main():
-    parser = argparse.ArgumentParser(description="Run GPT with standard multi-head attention.")
+    parser = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter, description="Run GPT with standard multi-head attention.")
     parser.add_argument("--emb_dim", type=int, default=768, help="Model embedding dimension.")
     parser.add_argument("--n_heads", type=int, default=12, help="Number of attention heads.")
     parser.add_argument("--n_layers", type=int, default=12, help="Number of transformer blocks.")
     parser.add_argument("--max_new_tokens", type=int, default=200, help="Number of tokens to generate.")
     parser.add_argument("--latent_dim", type=int, default=None,
-                        help="Latent dim for MLA (default: d_out//8)")
+        help="Latent dim for MLA")
 
     args = parser.parse_args()
 

ch04/05_mla/memory_estimator_mla.py

@@ -37,7 +37,7 @@ def calc_mla_bytes_total(batch, context_length, n_layers, latent_dim, bytes_per_
 
 
 def main():
-    p = argparse.ArgumentParser(description="Estimate KV-cache memory for MHA vs GQA vs MLA")
+    p = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter, description="Estimate KV-cache memory for MHA vs GQA vs MLA")
     p.add_argument("--context_length", default=1024, type=int)
     p.add_argument("--emb_dim", required=True, type=int)
     p.add_argument("--n_heads", required=True, type=int)

ch04/06_swa/gpt_with_kv_mha.py

@@ -278,7 +278,7 @@ def generate_text_simple_cached(model, idx, max_new_tokens,
 
 
 def main():
-    parser = argparse.ArgumentParser(description="Run GPT with standard multi-head attention.")
+    parser = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter, description="Run GPT with standard multi-head attention.")
     parser.add_argument("--emb_dim", type=int, default=768, help="Model embedding dimension.")
     parser.add_argument("--n_heads", type=int, default=12, help="Number of attention heads.")
     parser.add_argument("--n_layers", type=int, default=12, help="Number of transformer blocks.")

ch04/06_swa/gpt_with_kv_swa.py

@@ -311,7 +311,7 @@ def generate_text_simple_cached(model, idx, max_new_tokens,
 
 
 def main():
-    parser = argparse.ArgumentParser(description="Run GPT with standard multi-head attention.")
+    parser = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter, description="Run GPT with standard multi-head attention.")
     parser.add_argument("--emb_dim", type=int, default=768, help="Model embedding dimension.")
     parser.add_argument("--n_heads", type=int, default=12, help="Number of attention heads.")
     parser.add_argument("--n_layers", type=int, default=12, help="Number of transformer blocks.")

ch04/06_swa/memory_estimator_swa.py

@@ -90,7 +90,7 @@ def estimate_totals(context_length, sliding_window_size, emb_dim, n_heads, n_lay
 
 
 def main():
-    p = argparse.ArgumentParser(description="Estimate KV-cache memory for MHA/GQA with SWA layer ratio")
+    p = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter, description="Estimate KV-cache memory for MHA/GQA with SWA layer ratio")
     p.add_argument("--context_length", default=1024, type=int)
     p.add_argument("--sliding_window_size", required=True, type=int,
                    help="SWA window size W per SWA layer.")

ch04/06_swa/plot_memory_estimates_swa.py

@@ -102,7 +102,7 @@ def calc_kv_bytes_total_gqa_swa(
 
 
 def main():
-    p = argparse.ArgumentParser(
+    p = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter,
         description="KV-cache vs Context Length — MHA vs GQA with SWA overlays"
     )
     p.add_argument("--emb_dim", type=int, required=True)

ch04/07_moe/gpt_with_kv_ffn.py

@@ -341,7 +341,7 @@ def generate_text_simple_cached(model, idx, max_new_tokens,
 
 
 def main():
-    parser = argparse.ArgumentParser()
+    parser = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter)
     parser.add_argument("--emb_dim", type=int, default=768, help="Model embedding dimension.")
     parser.add_argument("--hidden_dim", type=int, default=768*4, help="Intermediate FFN size.")
     parser.add_argument("--n_heads", type=int, default=12, help="Number of attention heads.")

ch04/07_moe/gpt_with_kv_moe.py

@@ -401,7 +401,7 @@ def generate_text_simple_cached(model, idx, max_new_tokens,
 
 
 def main():
-    parser = argparse.ArgumentParser()
+    parser = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter)
     parser.add_argument("--emb_dim", type=int, default=768, help="Model embedding dimension.")
     parser.add_argument("--hidden_dim", type=int, default=768*4, help="Intermediate FFN or MoE size.")
     parser.add_argument("--n_heads", type=int, default=12, help="Number of attention heads.")

ch04/07_moe/memory_estimator_moe.py

@@ -65,7 +65,7 @@ def estimate_params_and_hidden(
 
 
 def main():
-    p = argparse.ArgumentParser(
+    p = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter,
         description="Estimate FFN vs MoE parameter memory"
     )
     p.add_argument("--emb_dim", type=int, required=True,

ch04/07_moe/plot_memory_estimates_moe.py

@@ -91,7 +91,7 @@ def plot_active_params_vs_experts(
 
 
 def main():
-    p = argparse.ArgumentParser(description="Plot Dense vs MoE active parameters.")
+    p = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter, description="Plot Dense vs MoE active parameters.")
     p.add_argument("--emb_dim", type=int, required=True, help="Embedding dimension")
     p.add_argument("--hidden_dim", type=int, required=True, help="Dense FFN hidden size")
     p.add_argument("--ffn_type", choices=["gelu", "swiglu"], default="swiglu")

ch04/08_deltanet/plot_memory_estimates_gated_deltanet.py

@@ -36,7 +36,7 @@ def convert_to_gb(x):
 
 
 def main():
-    p = argparse.ArgumentParser(description="Memory vs. Context Length: MHA vs. DeltaNet (3:1 mix)")
+    p = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter, description="Memory vs. Context Length: MHA vs. DeltaNet (3:1 mix)")
     p.add_argument("--batch", type=int, default=1)
     p.add_argument("--emb_dim", type=int, default=2048)
     p.add_argument("--n_heads", type=int, default=16)

ch05/01_main-chapter-code/gpt_generate.py

@@ -253,16 +253,16 @@ def main(gpt_config, input_prompt, model_size, device):
 
 if __name__ == "__main__":
 
-    parser = argparse.ArgumentParser(description="Generate text with a pretrained GPT-2 model.")
+    parser = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter, description="Generate text with a pretrained GPT-2 model.")
     parser.add_argument(
         "--prompt",
         default="Every effort moves you",
-        help="Prompt text used to seed the generation (default matches the script's built-in prompt)."
+        help="Prompt text used to seed the generation."
     )
     parser.add_argument(
         "--device",
         default="cpu",
-        help="Device for running inference, e.g., cpu, cuda, mps, or auto. Defaults to cpu."
+        help="Device for running inference, e.g., cpu, cuda, mps, or auto."
     )
 
     args = parser.parse_args()

ch05/03_bonus_pretraining_on_gutenberg/prepare_dataset.py

@@ -66,7 +66,7 @@ def combine_files(file_paths, target_dir, max_size_mb=500, separator="<|endoftex
 
 if __name__ == "__main__":
 
-    parser = argparse.ArgumentParser(description="Preprocess and combine text files for pretraining")
+    parser = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter, description="Preprocess and combine text files for pretraining")
 
     parser.add_argument("--data_dir", type=str, default="gutenberg/data/raw",
                         help="Directory containing the downloaded raw training data")

ch05/03_bonus_pretraining_on_gutenberg/pretraining_simple.py

@@ -148,7 +148,7 @@ def train_model_simple(model, optimizer, device, n_epochs,
 
 if __name__ == "__main__":
 
-    parser = argparse.ArgumentParser(description="GPT Model Training Configuration")
+    parser = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter, description="GPT Model Training Configuration")
 
     parser.add_argument("--data_dir", type=str, default="gutenberg/data",
                         help="Directory containing the training data")

ch06/01_main-chapter-code/gpt_class_finetune.py

@@ -239,7 +239,7 @@ if __name__ == "__main__":
 
     import argparse
 
-    parser = argparse.ArgumentParser(
+    parser = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter,
         description="Finetune a GPT model for classification"
     )
     parser.add_argument(

ch06/02_bonus_additional-experiments/additional_experiments.py

@@ -410,7 +410,7 @@ def replace_linear_with_lora(model, rank, alpha, alternative=False):
 
 if __name__ == "__main__":
 
-    parser = argparse.ArgumentParser()
+    parser = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter)
     parser.add_argument(
         "--model_size",
         type=str,

ch06/03_bonus_imdb-classification/train_bert_hf.py

@@ -175,7 +175,7 @@ def train_classifier_simple(model, train_loader, val_loader, optimizer, device,
 
 if __name__ == "__main__":
 
-    parser = argparse.ArgumentParser()
+    parser = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter)
     parser.add_argument(
         "--trainable_layers",
         type=str,

ch06/03_bonus_imdb-classification/train_bert_hf_spam.py

@@ -272,7 +272,7 @@ def train_classifier_simple(model, train_loader, val_loader, optimizer, device,
 
 if __name__ == "__main__":
 
-    parser = argparse.ArgumentParser()
+    parser = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter)
     parser.add_argument(
         "--trainable_layers",
         type=str,

ch06/03_bonus_imdb-classification/train_gpt.py

@@ -225,7 +225,7 @@ def train_classifier_simple(model, train_loader, val_loader, optimizer, device,
 
 if __name__ == "__main__":
 
-    parser = argparse.ArgumentParser()
+    parser = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter)
     parser.add_argument(
         "--model_size",
         type=str,

ch07/01_main-chapter-code/exercise_experiments.py

@@ -534,7 +534,7 @@ if __name__ == "__main__":
 
     import argparse
 
-    parser = argparse.ArgumentParser(
+    parser = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter,
         description="Instruction finetune a GPT model"
     )
     options = {"baseline", "mask_instructions", "alpaca_52k", "phi3_prompt", "lora"}

ch07/01_main-chapter-code/gpt_instruction_finetuning.py

@@ -333,7 +333,7 @@ if __name__ == "__main__":
 
     import argparse
 
-    parser = argparse.ArgumentParser(
+    parser = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter,
         description="Finetune a GPT model for classification"
     )
     parser.add_argument(

ch07/01_main-chapter-code/ollama_evaluate.py

@@ -103,7 +103,7 @@ if __name__ == "__main__":
 
     import argparse
 
-    parser = argparse.ArgumentParser(
+    parser = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter,
         description="Evaluate model responses with ollama"
     )
     parser.add_argument(

ch07/02_dataset-utilities/find-near-duplicates.py

@@ -100,7 +100,7 @@ def find_print_and_remove_near_duplicates(json_data, remove_duplicates=False, th
 if __name__ == "__main__":
     print("scikit-learn version:", sklearn_version)
 
-    parser = argparse.ArgumentParser()
+    parser = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter)
     parser.add_argument(
         "--json_file",
         type=str,