v0.9.0

CHANGELOG.md

@@ -1,8 +1,16 @@
+## Release v0.9.0 - TBD
+
+### Added
+
+- Added possibility to train models on custom datasets, without generating synthetic data. 
+- Added Italian `SpamDetection` model.
+- Added the `TextSummarization` model.
+
 ## Release v0.8.0 - March 26, 2026
 
 ### Added
 
-- Added German, Spanish Guardrail model and German, Spanish Spam Detection model.
+- Added German, Spanish `Guardrail` model and German, Spanish `SpamDetection` model.
 
 ### Changed
 

README.md

@@ -28,7 +28,7 @@
 </p>
 
 <p align="center">
-  <strong>Small Language Model Inference, Fine-Tuning and Observability. No GPU, no labeled data needed.</strong>
+    <strong>Small Language Model Inference, Fine-Tuning and Observability. No GPU, no labeled data needed.</strong>
 </p>
 
 ---
@@ -61,21 +61,18 @@ Artifex changes that:
 
 At this time, Artifex supports the following models:
 
-| Task | Language(s) | Description | Default Model | How to use |
+| Task | Available Languages | Description | Default Model | How to use |
 |------|----------|-------------|---------------|------------|
-| Guardrail | English | Flags unsafe, harmful, or off-topic messages. | [tanaos/tanaos-guardrail-v2](https://huggingface.co/tanaos/tanaos-guardrail-v2) | [Examples](https://docs.tanaos.com/artifex/guardrail/code-examples/)
-| Guardrail | Spanish | Flags unsafe, harmful, or off-topic messages. | [tanaos/tanaos-guardrail-spanish](https://huggingface.co/tanaos/tanaos-guardrail-spanish) | [Examples](https://colab.research.google.com/#fileId=https%3A//huggingface.co/tanaos/tanaos-guardrail-spanish.ipynb) |
-| Guardrail | German | Flags unsafe, harmful, or off-topic messages. | [tanaos/tanaos-guardrail-german](https://huggingface.co/tanaos/tanaos-guardrail-german) | [Examples](https://colab.research.google.com/#fileId=https%3A//huggingface.co/tanaos/tanaos-guardrail-german.ipynb) |
+| Guardrail | English, German, Spanish | Flags unsafe, harmful, or off-topic messages. | [tanaos/tanaos-guardrail-v2](https://huggingface.co/tanaos/tanaos-guardrail-v2) (English version, see the page for the other languages) | [Examples](https://docs.tanaos.com/artifex/guardrail/code-examples/)
 | Intent Classification | English | Classifies user messages into predefined intent categories. | [tanaos/tanaos-intent-classifier-v1](https://huggingface.co/tanaos/tanaos-intent-classifier-v1) | [Examples](https://docs.tanaos.com/artifex/intent-classifier/code-examples/)
 | Reranker | English | Ranks a list of items or search results based on relevance to a query. | [cross-encoder/mmarco-mMiniLMv2-L12-H384-v1](https://huggingface.co/cross-encoder/mmarco-mMiniLMv2-L12-H384-v1) | [Examples](https://docs.tanaos.com/artifex/reranker/code-examples/)
 | Sentiment Analysis | English | Determines the sentiment (positive, negative, neutral) of a given text. | [tanaos/tanaos-sentiment-analysis-v1](https://huggingface.co/tanaos/tanaos-sentiment-analysis-v1) | [Examples](https://docs.tanaos.com/artifex/sentiment-analysis/code-examples/)
 | Emotion Detection | English | Identifies the emotion expressed in a given text. | [tanaos/tanaos-emotion-detection-v1](https://huggingface.co/tanaos/tanaos-emotion-detection-v1) | [Examples](https://docs.tanaos.com/artifex/emotion-detection/code-examples/)
 | Named Entity Recognition | English | Detects and classifies named entities in text (e.g., persons, organizations, locations). | [tanaos/tanaos-NER-v1](https://huggingface.co/tanaos/tanaos-NER-v1) | [Examples](https://docs.tanaos.com/artifex/named-entity-recognition/code-examples/)
 | Text Anonymization | English | Removes personally identifiable information (PII) from text. | [tanaos/tanaos-text-anonymizer-v1](https://huggingface.co/tanaos/tanaos-text-anonymizer-v1) | [Examples](https://docs.tanaos.com/artifex/text-anonymization/code-examples/)
-| Spam Detection | English | Identifies whether a message is spam or not. | [tanaos/tanaos-spam-detection-v1](https://huggingface.co/tanaos/tanaos-spam-detection-v1) | [Examples](https://docs.tanaos.com/artifex/spam-detection/code-examples/)
-| Spam Detection | Spanish | Identifies whether a message is spam or not. | [tanaos/tanaos-spam-detection-spanish](https://huggingface.co/tanaos/tanaos-spam-detection-spanish) | [Examples](https://colab.research.google.com/#fileId=https%3A//huggingface.co/tanaos/tanaos-spam-detection-spanish.ipynb) |
-| Spam Detection | German | Identifies whether a message is spam or not. | [tanaos/tanaos-spam-detection-german](https://huggingface.co/tanaos/tanaos-spam-detection-german) | [Examples](https://colab.research.google.com/#fileId=https%3A//huggingface.co/tanaos/tanaos-spam-detection-german.ipynb) |
+| Spam Detection | English, German, Spanish, Italian | Identifies whether a message is spam or not. | [tanaos/tanaos-spam-detection-v1](https://huggingface.co/tanaos/tanaos-spam-detection-v1) (English version, see the page for the other languages) | [Examples](https://docs.tanaos.com/artifex/spam-detection/code-examples/)
 | Topic Classification | English | Classifies text into predefined topics. | [tanaos/tanaos-topic-classification-v1](https://huggingface.co/tanaos/tanaos-topic-classification-v1) | [Examples](https://docs.tanaos.com/artifex/topic-classification/code-examples/)
+| Text Summarization | English | Rewrites text in a more concise way. | [tanaos/tanaos-text-summarization-v1](https://huggingface.co/tanaos/tanaos-text-summarization-v1) | [Examples](https://docs.tanaos.com/artifex/text-summarization/code-examples/)
 
 Looking for models in other languages? Our [Enterprise License](#license-paid--enterprise-solutions) includes models in any language. Reach out at [info@tanaos.com](mailto:info@tanaos.com) for more details.
 

artifex/__init__.py

@@ -14,6 +14,7 @@
         SpamDetection, TopicClassification
     from .models.named_entity_recognition import NamedEntityRecognition, TextAnonymization
     from .models.reranker import Reranker
+    from .models.text_summarization import TextSummarization
     from .config import config
 console.print(f"[green]✔ Initializing Artifex[/green]")
 
@@ -56,6 +57,7 @@ def __init__(self, api_key: Optional[str] = None):
         self._text_anonymization = None
         self._spam_detection = None
         self._topic_classification = None
+        self._text_summarization = None
 
     def text_classification(self) -> ClassificationModel:
         """
@@ -160,12 +162,12 @@ def text_anonymization(self) -> TextAnonymization:
         return self._text_anonymization
 
     def spam_detection(
-        self, language: Literal["english", "spanish", "german"] = "english"
+        self, language: Literal["english", "spanish", "german", "italian"] = "english"
     ) -> SpamDetection:
         """
         Lazy loads the SpamDetection instance.
         Args:
-            language (Literal["english", "spanish", "german"]): The language of the text data.
+            language (Literal["english", "spanish", "german", "italian"]): The language of the text data.
         Returns:
             SpamDetection: An instance of the SpamDetection class.
         """
@@ -187,4 +189,16 @@ def topic_classification(self) -> TopicClassification:
         if self._topic_classification is None:
             with console.status("Loading Topic Classification model..."):
                 self._topic_classification = TopicClassification(synthex=self._synthex_client)
-        return self._topic_classification
+        return self._topic_classification
+
+    def text_summarization(self) -> TextSummarization:
+        """
+        Lazy loads the TextSummarization instance.
+        Returns:
+            TextSummarization: An instance of the TextSummarization class.
+        """
+
+        if self._text_summarization is None:
+            with console.status("Loading Text Summarization model..."):
+                self._text_summarization = TextSummarization(synthex=self._synthex_client)
+        return self._text_summarization

artifex/config.py

@@ -70,11 +70,17 @@ def DEFAULT_SYNTHEX_DATASET_NAME(self) -> str:
     NER_HF_BASE_MODEL: str = "tanaos/tanaos-NER-v1"
     NER_TOKENIZER_MAX_LENGTH: int = 256
     NER_TAGNAME_MAX_LENGTH: int = 25
+
+    # Text Summarization Model
+    TEXT_SUMMARIZATION_HF_BASE_MODEL: str = "tanaos/tanaos-text-summarization-v1"
+    TEXT_SUMMARIZATION_MAX_INPUT_LENGTH: int = 512
+    TEXT_SUMMARIZATION_MAX_TARGET_LENGTH: int = 128
 
     # Spam Detection Models
     SPAM_DETECTION_ENGLISH_HF_BASE_MODEL: str = "tanaos/tanaos-spam-detection-v1"
     SPAM_DETECTION_SPANISH_HF_BASE_MODEL: str = "tanaos/tanaos-spam-detection-spanish"
     SPAM_DETECTION_GERMAN_HF_BASE_MODEL: str = "tanaos/tanaos-spam-detection-german"
+    SPAM_DETECTION_ITALIAN_HF_BASE_MODEL: str = "tanaos/tanaos-spam-detection-italian"
     SPAM_DETECTION_EVAL_DATASET: str = "hf://datasets/Deysi/spam-detection-dataset/data/train-00000-of-00001-daf190ce720b3dbb.parquet"
 
     # Topic Classification Model

artifex/core/_hf_patches.py

@@ -1,4 +1,5 @@
-from transformers import Trainer, TrainerState, TrainingArguments, TrainerCallback, TrainerControl
+from transformers import Trainer, TrainerState, TrainingArguments, TrainerCallback, TrainerControl, \
+    Seq2SeqTrainer
 from transformers.trainer_utils import TrainOutput
 from typing import Any, Dict
 from rich.progress import Progress, BarColumn, TextColumn, TimeElapsedColumn, TaskID
@@ -35,6 +36,32 @@ def silent_print(*a: Any, **k: Any) -> None:
             return super().train(*args, **kwargs)
         finally:
             builtins.print = orig_print
+
+
+class SilentSeq2SeqTrainer(Seq2SeqTrainer):
+    """
+    A regular transformers.Seq2SeqTrainer which prevents the tedious final training summary
+    dictionary from being printed to the console.
+    """
+
+    def train(self, *args: Any, **kwargs: Any) -> TrainOutput:
+        import builtins
+        orig_print = builtins.print
+
+        def silent_print(*a: Any, **k: Any) -> None:
+            if (
+                len(a) == 1
+                and isinstance(a[0], dict)
+                and "train_runtime" in a[0]
+            ):
+                return
+            return orig_print(*a, **k)
+
+        builtins.print = silent_print
+        try:
+            return super().train(*args, **kwargs)
+        finally:
+            builtins.print = orig_print
 
 
 class RichProgressCallback(TrainerCallback):

artifex/models/__init__.py

@@ -7,6 +7,8 @@
 
 from .reranker import Reranker
 
+from .text_summarization import TextSummarization
+
 __all__ = [
     "ClassificationModel",
     "EmotionDetection",
@@ -18,5 +20,6 @@
     "Reranker",
     "SpamDetection",
     "TopicClassification",
-    "Guardrail"
+    "Guardrail",
+    "TextSummarization"
 ]