shadow-tools 0.7.1__tar.gz → 0.7.2__tar.gz

{shadow_tools-0.7.1 → shadow_tools-0.7.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: shadow-tools
-Version: 0.7.1
+Version: 0.7.2
 Summary: A practical all-in-one Python toolkit for HTTP, WebSockets, files, data, validation, async work, and CLIs.
 Project-URL: Homepage, https://github.com/WilliamLaurento/shadow-tools
 Project-URL: Issues, https://github.com/WilliamLaurento/shadow-tools/issues
@@ -73,6 +73,9 @@ print(say(bot, "The future of technology", max_tokens=80))
 bot.save("my_text_model")
 ```
 
+Training prints the tokenizer status plus each epoch and loss. Use
+`learn("corpus.txt", epochs=20, verbose=False)` when you do not want updates.
+
 Install it with `pip install "shadow-tools[ml,nlp]"`. The package trains the
 tokenizer and Transformer together, then converts predicted token IDs back to
 normal text for you.

{shadow_tools-0.7.1 → shadow_tools-0.7.2}/README.md

@@ -45,6 +45,9 @@ print(say(bot, "The future of technology", max_tokens=80))
 bot.save("my_text_model")
 ```
 
+Training prints the tokenizer status plus each epoch and loss. Use
+`learn("corpus.txt", epochs=20, verbose=False)` when you do not want updates.
+
 Install it with `pip install "shadow-tools[ml,nlp]"`. The package trains the
 tokenizer and Transformer together, then converts predicted token IDs back to
 normal text for you.

{shadow_tools-0.7.1 → shadow_tools-0.7.2}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "shadow-tools"
-version = "0.7.1"
+version = "0.7.2"
 description = "A practical all-in-one Python toolkit for HTTP, WebSockets, files, data, validation, async work, and CLIs."
 readme = "README.md"
 requires-python = ">=3.10"

{shadow_tools-0.7.1 → shadow_tools-0.7.2}/src/shadowtools/__init__.py

@@ -15,7 +15,7 @@ from .text import chunks, slugify, to_camel, to_snake
 from .training import SmartTokenizer, fine_tune, make_loaders, predict_text, train, train_text
 from .validation import is_email, is_url, require_keys
 
-__version__ = "0.7.1"
+__version__ = "0.7.2"
 
 __all__ = [
     "append_text", "audit_headers", "BPETokenizer", "chunks", "ensure_dir", "env", "file_hash", "GeneratorConfig", "generate_random_ips", "generate_test_ips", "generate_text", "get", "hash_text", "is_email", "is_subdomain_of", "is_url", "learn", "make_password", "make_subdomain", "random_token", "read_csv",

{shadow_tools-0.7.1 → shadow_tools-0.7.2}/src/shadowtools/generator.py

@@ -19,6 +19,7 @@ class GeneratorConfig:
     batch_size: int = 16
     learning_rate: float = 3e-4
     device: str = "auto"
+    verbose: bool = True
 
 
 class TextGenerator:
@@ -78,18 +79,24 @@ class TextGenerator:
 
         return CausalTransformer().to(self.device)
 
-    def fit(self, texts: str | Sequence[str], *, epochs: int = 10) -> list[dict[str, float]]:
+    def fit(self, texts: str | Sequence[str], *, epochs: int = 10,
+            verbose: bool | None = None) -> list[dict[str, float]]:
         """Train from a large text string or a list of documents."""
         if epochs < 1:
             raise ValueError("epochs must be at least 1.")
         corpus = texts if isinstance(texts, str) else "\n".join(texts)
         if not isinstance(corpus, str) or len(corpus.strip()) < 100:
             raise ValueError("Give the generator at least 100 characters of training text.")
+        show_progress = self.config.verbose if verbose is None else verbose
+        if show_progress:
+            print("Building tokenizer...")
         self.tokenizer = BPETokenizer(vocab_size=self.config.vocab_size, min_frequency=1).train([corpus])
         token_ids = self.tokenizer.encode(corpus, add_special_tokens=False)["input_ids"]
         if len(token_ids) <= self.config.block_size:
             raise ValueError("Training text needs more tokens than block_size.")
         self.model = self._build_model(self.tokenizer.size)
+        if show_progress:
+            print(f"Tokenizer ready: {self.tokenizer.size} tokens. Training model for {epochs} epoch(s)...")
         torch = self.torch
         block_size = self.config.block_size
 
@@ -119,7 +126,10 @@ class TextGenerator:
                 torch.nn.utils.clip_grad_norm_(self.model.parameters(), 1.0)
                 optimizer.step()
                 total_loss += float(loss.detach().cpu())
-            history.append({"epoch": float(epoch), "loss": total_loss / len(loader)})
+            metrics = {"epoch": float(epoch), "loss": total_loss / len(loader)}
+            history.append(metrics)
+            if show_progress:
+                print(f"Epoch {epoch}/{epochs} | loss: {metrics['loss']:.4f}")
         return history
 
     def generate(self, prompt: str, *, max_tokens: int = 80, temperature: float = 0.8,
@@ -181,11 +191,12 @@ def train_generator(texts: str | Sequence[str], *, epochs: int = 10,
                     vocab_size: int = 4_000, block_size: int = 64,
                     embedding_size: int = 192, heads: int = 6, layers: int = 4,
                     batch_size: int = 16, learning_rate: float = 3e-4,
-                    device: str = "auto") -> tuple[TextGenerator, list[dict[str, float]]]:
+                    device: str = "auto", verbose: bool = True) -> tuple[TextGenerator, list[dict[str, float]]]:
     """Train a text generator in one call, without writing model boilerplate."""
     generator = TextGenerator(GeneratorConfig(vocab_size=vocab_size, block_size=block_size,
                               embedding_size=embedding_size, heads=heads, layers=layers,
-                              batch_size=batch_size, learning_rate=learning_rate, device=device))
+                              batch_size=batch_size, learning_rate=learning_rate, device=device,
+                              verbose=verbose))
     return generator, generator.fit(texts, epochs=epochs)
 
 
@@ -194,7 +205,8 @@ def generate_text(generator: TextGenerator, prompt: str, **options: Any) -> str:
     return generator.generate(prompt, **options)
 
 
-def learn(source: str | Path | Sequence[str], *, epochs: int = 10, **options: Any) -> TextGenerator:
+def learn(source: str | Path | Sequence[str], *, epochs: int = 10,
+          verbose: bool = True, **options: Any) -> TextGenerator:
     """Beginner-friendly training: give it a text file, text, or list of texts."""
     text: str | Sequence[str]
     if isinstance(source, Path):
@@ -207,7 +219,7 @@ def learn(source: str | Path | Sequence[str], *, epochs: int = 10, **options: An
             text = source
     else:
         text = source
-    generator, _ = train_generator(text, epochs=epochs, **options)
+    generator, _ = train_generator(text, epochs=epochs, verbose=verbose, **options)
     return generator