Upd/tutorials (#252)

* rename zero-shot presets

* small fixes

* upd concepts a little bit

* update basic data tutorial

* update advanced data usage guide

* update modules user guide

* add embedder configuration guide

* minor fixes

* upd automl guide

* fix guides runtime errors

* minor changes

* minor fix

* fix f-string

* try to fix typing

Author: voorhs

autointent/_dump_tools/unit_dumpers.py

@@ -215,13 +215,13 @@ def dump(obj: PeftModel, path: Path, exists_ok: bool) -> None:
             ptuning_path = path / "ptuning"
             ptuning_path.mkdir(parents=True, exist_ok=exists_ok)
             obj.save_pretrained(str(ptuning_path / "peft"))
-            obj.base_model.save_pretrained(ptuning_path / "base_model")  # type: ignore[attr-defined]
+            obj.base_model.save_pretrained(ptuning_path / "base_model")
         else:
             # strategy to save lora models: merge adapters and save as usual hugging face model
             lora_path = path / "lora"
             lora_path.mkdir(parents=True, exist_ok=exists_ok)
             merged_model: PreTrainedModel = obj.merge_and_unload()
-            merged_model.save_pretrained(lora_path)  # type: ignore[attr-defined]
+            merged_model.save_pretrained(lora_path)
 
     @staticmethod
     def load(path: Path, **kwargs: Any) -> PeftModel:  # noqa: ANN401, ARG004
@@ -248,7 +248,7 @@ class HFModelDumper(BaseObjectDumper[PreTrainedModel]):
     @staticmethod
     def dump(obj: PreTrainedModel, path: Path, exists_ok: bool) -> None:
         path.mkdir(parents=True, exist_ok=exists_ok)
-        obj.save_pretrained(path)  # type: ignore[attr-defined]
+        obj.save_pretrained(path)
 
     @staticmethod
     def load(path: Path, **kwargs: Any) -> PreTrainedModel:  # noqa: ANN401, ARG004
@@ -265,7 +265,7 @@ class HFTokenizerDumper(BaseObjectDumper[PreTrainedTokenizer | PreTrainedTokeniz
     @staticmethod
     def dump(obj: PreTrainedTokenizer | PreTrainedTokenizerFast, path: Path, exists_ok: bool) -> None:
         path.mkdir(parents=True, exist_ok=exists_ok)
-        obj.save_pretrained(path)  # type: ignore[union-attr]
+        obj.save_pretrained(path)
 
     @staticmethod
     def load(path: Path, **kwargs: Any) -> PreTrainedTokenizer | PreTrainedTokenizerFast:  # noqa: ANN401, ARG004

…ent/_presets/zero-shot-transformers.yaml …ointent/_presets/zero-shot-encoders.yamlautointent/_presets/zero-shot-transformers.yaml renamed to autointent/_presets/zero-shot-encoders.yaml autointent/_presets/zero-shot-transformers.yaml renamed to autointent/_presets/zero-shot-encoders.yaml

@@ -124,7 +124,7 @@ class Split:
     "transformers-heavy",
     "transformers-light",
     "transformers-no-hpo",
-    "zero-shot-openai",
-    "zero-shot-transformers",
+    "zero-shot-llm",
+    "zero-shot-encoders",
 ]
 """Some presets that our library supports."""

autointent/_presets/zero-shot-openai.yaml autointent/_presets/zero-shot-llm.yamlautointent/_presets/zero-shot-openai.yaml renamed to autointent/_presets/zero-shot-llm.yaml

@@ -181,7 +181,7 @@ def _train(self, tokenized_dataset: DatasetDict) -> None:
                 load_best_model_at_end=self.early_stopping_config.metric is not None,
             )
 
-            trainer = Trainer(  # type: ignore[no-untyped-call]
+            trainer = Trainer(
                 model=self._model,
                 args=training_args,
                 train_dataset=tokenized_dataset["train"],
@@ -192,10 +192,10 @@ def _train(self, tokenized_dataset: DatasetDict) -> None:
                 callbacks=self._get_trainer_callbacks(),
             )
             if not self.print_progress:
-                trainer.remove_callback(PrinterCallback)  # type: ignore[attr-defined]
-                trainer.remove_callback(ProgressCallback)  # type: ignore[attr-defined]
+                trainer.remove_callback(PrinterCallback)
+                trainer.remove_callback(ProgressCallback)
 
-            trainer.train()  # type: ignore[attr-defined]
+            trainer.train()
 
     def _get_trainer_callbacks(self) -> list[TrainerCallback]:
         res: list[TrainerCallback] = []

autointent/custom_types/_types.py

@@ -60,16 +60,16 @@ AutoIntent's architecture centers around transformer-based text embeddings, prov
 
 .. _concepts-multiclass-multilabel:
 
-Classification Paradigms
-========================
+Multi- vs. Single-label classification
+======================================
 
 AutoIntent supports various classification scenarios through its flexible decision module:
 
 **🏷️ Multi-Class Classification**
-   Traditional single-label classification where each input belongs to exactly one class. Uses argmax or threshold-based decisions on predicted probabilities.
+   Each input gets assigned to exactly one category - like sorting emails into "Spam", "Work", or "Personal" folders. Common examples include sentiment analysis (positive/negative/neutral) or determining user intent where each message has a single purpose. The model picks the single best match from all possible categories.
 
 **🔖 Multi-Label Classification** 
-   Each input can belong to multiple classes simultaneously. Employs adaptive thresholding strategies that can be sample-specific or learned globally across the dataset.
+   Each input can belong to multiple categories at once - like tagging a news article as both "Politics" and "Economics". Essential for scenarios like multi-intent messages ("book a flight and check weather"), content tagging, or any situation where multiple labels can apply simultaneously. The model almost independently decides whether each possible category fits or not.
 
 
 .. _concepts-oos:

autointent/modules/scoring/_bert.py

@@ -50,7 +50,7 @@
     "sphinx.ext.intersphinx",
     "sphinx_multiversion",
     "sphinx.ext.napoleon",
-    "sphinx_toolbox.collapse"
+    "sphinx_toolbox.collapse",
 ]
 
 templates_path = ["_templates"]

docs/source/concepts.rst

@@ -65,7 +65,7 @@ Reference
 :doc:`🔧 API Reference <autoapi/autointent/index>`
    Complete technical documentation for all classes, methods, and functions. Essential reference for developers integrating AutoIntent into their applications.
 
-   Key sections: :doc:`Modules <autoapi/autointent/modules/index>` | :doc:`Metrics <autoapi/autointent/metrics/index>`
+   Key section: :doc:`Modules <autoapi/autointent/modules/index>`
 
 
 .. toctree::

docs/source/conf.py

@@ -173,7 +173,7 @@ AutoIntent provides presets that balance different objectives:
    pipeline_heavy = Pipeline.from_preset("classic-heavy")    # Performance-focused
    
    # Different model types  
-   pipeline_zero_shot = Pipeline.from_preset("zero-shot-transformers")  # No training data
+   pipeline_zero_shot = Pipeline.from_preset("zero-shot-encoders")  # No training data
 
 Bayesian Optimization Theory
 -----------------------------

docs/source/index.rst

@@ -25,17 +25,13 @@ This is one of the ways to solve the problem of an overwhelming number of combin
 
 This algorithm checks fewer combinations, which speeds up the process. To implement such an algorithm, it is necessary to be able to evaluate the quality of not only the final prediction of the entire pipeline but also its intermediate predictions. The main drawback of this approach is that the decisions made are optimal only locally, not globally. The metrics for evaluating intermediate predictions are only a proxy signal for the quality of the final prediction.
 
-This approach has been available in our library since release v0.0.1.
-
 Random Search
 -------------
 
 A simpler strategy is to take a random subset of the full search space (random grid search). A straightforward strategy is to iterate through all combinations in random order until a certain time budget is exhausted.
 
 This approach is less intelligent than the greedy strategy because, at any moment during the random combination search, poor embedders or any other bad parameters might keep appearing, despite they have been tested already. The greedy strategy would have eliminated such embedders at the beginning and not revisited them. On the other hand, random search, by its nature, does not rely on any local decisions.
 
-The implementation of this optimization method is planned for release v0.1.0.
-
 Bayesian Optimization
 ---------------------