sinapsis-huggingface 0.2.7__tar.gz → 0.2.8__tar.gz

{sinapsis_huggingface-0.2.7 → sinapsis_huggingface-0.2.8}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: sinapsis-huggingface
-Version: 0.2.7
+Version: 0.2.8
 Summary: Package for HuggingFace-based templates
 Author-email: SinapsisAI <dev@sinapsis.tech>
 Project-URL: Homepage, https://sinapsis.tech

{sinapsis_huggingface-0.2.7 → sinapsis_huggingface-0.2.8}/packages/sinapsis_huggingface.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: sinapsis-huggingface
-Version: 0.2.7
+Version: 0.2.8
 Summary: Package for HuggingFace-based templates
 Author-email: SinapsisAI <dev@sinapsis.tech>
 Project-URL: Homepage, https://sinapsis.tech

{sinapsis_huggingface-0.2.7 → sinapsis_huggingface-0.2.8}/packages/sinapsis_huggingface.egg-info/SOURCES.txt

@@ -21,7 +21,6 @@ packages/sinapsis_huggingface_embeddings/src/sinapsis_huggingface_embeddings/hel
 packages/sinapsis_huggingface_embeddings/src/sinapsis_huggingface_embeddings/templates/__init__.py
 packages/sinapsis_huggingface_embeddings/src/sinapsis_huggingface_embeddings/templates/hugging_face_embedding_extractor.py
 packages/sinapsis_huggingface_embeddings/src/sinapsis_huggingface_embeddings/templates/speaker_embedding_from_audio.py
-packages/sinapsis_huggingface_embeddings/src/sinapsis_huggingface_embeddings/templates/speaker_embedding_from_dataset.py
 packages/sinapsis_huggingface_grounding_dino/src/sinapsis_huggingface_grounding_dino/__init__.py
 packages/sinapsis_huggingface_grounding_dino/src/sinapsis_huggingface_grounding_dino/helpers/__init__.py
 packages/sinapsis_huggingface_grounding_dino/src/sinapsis_huggingface_grounding_dino/helpers/grounding_dino_keys.py

{sinapsis_huggingface-0.2.7 → sinapsis_huggingface-0.2.8}/packages/sinapsis_huggingface_diffusers/src/sinapsis_huggingface_diffusers/templates/base_diffusers.py

@@ -193,6 +193,7 @@ class BaseDiffusers(Template, ABC):
         This method performs garbage collection and clears GPU memory (if applicable) to prevent memory leaks
         and ensure efficient resource usage.
         """
+        del self.generator
         if self.attributes.device == "cuda":
             torch.cuda.empty_cache()
             torch.cuda.ipc_collect()

{sinapsis_huggingface-0.2.7 → sinapsis_huggingface-0.2.8}/packages/sinapsis_huggingface_grounding_dino/src/sinapsis_huggingface_grounding_dino/templates/grounding_dino.py

@@ -1,6 +1,5 @@
 # -*- coding: utf-8 -*-
-
-
+import gc
 from typing import Any, Literal
 
 import torch
@@ -357,6 +356,9 @@ class GroundingDINO(Template):
         This method performs garbage collection and clears GPU memory (if applicable) to prevent memory leaks
         and ensure efficient resource usage.
         """
+        for child in self.model.children():
+            child.cpu()
+            gc.collect()
         if self.attributes.device == "cuda":
             torch.cuda.empty_cache()
             torch.cuda.ipc_collect()

{sinapsis_huggingface-0.2.7 → sinapsis_huggingface-0.2.8}/packages/sinapsis_huggingface_transformers/src/sinapsis_huggingface_transformers/templates/pali_gemma/pali_gemma_base.py

@@ -1,4 +1,5 @@
 # -*- coding: utf-8 -*-
+import gc
 from abc import abstractmethod
 from typing import Any, ClassVar, Literal
 
@@ -104,7 +105,16 @@ class PaliGemmaBase(Template):
         """
 
     def reset_state(self, template_name: str | None = None) -> None:
-        if self.attributes.device == "cuda":
+        with torch.no_grad():
+            for child in self.model.children():
+                child.cpu()
+                del child
+                gc.collect()
+            del self.model
+
+            del self.processor
+
             torch.cuda.empty_cache()
             torch.cuda.ipc_collect()
+
         super().reset_state(template_name)

{sinapsis_huggingface-0.2.7 → sinapsis_huggingface-0.2.8}/packages/sinapsis_huggingface_transformers/src/sinapsis_huggingface_transformers/templates/pali_gemma/pali_gemma_inference.py

@@ -69,6 +69,7 @@ class PaliGemmaInference(PaliGemmaBase):
 
     def __init__(self, attributes: TemplateAttributeType) -> None:
         super().__init__(attributes)
+        self.model = self.model.eval()
         self.prompt = self.attributes.prompt
 
     def _prepare_inputs(self, image_content: np.ndarray) -> dict:
@@ -237,11 +238,12 @@ class PaliGemmaInference(PaliGemmaBase):
         Returns:
             DataContainer: Processed container with added annotations
         """
+        self.logger.debug("EXECUTING TEMPLATE")
         if container.texts:
             self.process_from_text_packet(container)
         else:
             self.process_from_prompt(container)
-
+        self.logger.debug("finished execution")
         return container
 
     @staticmethod

{sinapsis_huggingface-0.2.7 → sinapsis_huggingface-0.2.8}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "sinapsis-huggingface"
-version = "0.2.7"
+version = "0.2.8"
 description = "Package for HuggingFace-based templates"
 authors = [
     {name = "SinapsisAI", email = "dev@sinapsis.tech"},

sinapsis_huggingface-0.2.7/packages/sinapsis_huggingface_embeddings/src/sinapsis_huggingface_embeddings/templates/speaker_embedding_from_dataset.py

@@ -1,93 +0,0 @@
-# -*- coding: utf-8 -*-
-
-from typing import Literal
-
-from datasets import load_dataset
-from sinapsis_core.data_containers.data_packet import DataContainer
-from sinapsis_core.template_base import Template
-from sinapsis_core.template_base.base_models import OutputTypes, TemplateAttributes, UIPropertiesMetadata
-from sinapsis_core.utils.env_var_keys import SINAPSIS_CACHE_DIR
-
-
-class SpeakerEmbeddingFromDatasetAttributes(TemplateAttributes):
-    """Attributes for the SpeakerEmbeddingFromDataset template.
-
-    Attributes:
-        dataset_path (str): Path or name of the Hugging Face dataset containing speaker embeddings.
-            For example, `"Matthijs/cmu-arctic-xvectors"`.
-        data_cache_dir (str): Directory to cache the downloaded dataset. Defaults to the value of
-            the `SINAPSIS_CACHE_DIR` environment variable.
-        split (str): Dataset split to use (e.g., "train", "validation", or "test").
-            Defaults to `"validation"`.
-        sample_idx (int): Index of the dataset sample to extract the embedding from.
-        xvector_key (str): Key in the dataset sample that stores the xvector. Defaults to `"xvector"`.
-        target_packet (Literal["texts", "audios"]): Type of packet in the `DataContainer` to which
-            the embedding will be attached. Must be either `"texts"` or `"audios"`.
-    """
-
-    dataset_path: str
-    data_cache_dir: str = str(SINAPSIS_CACHE_DIR)
-    split: str = "validation"
-    sample_idx: int
-    xvector_key: str = "xvector"
-    target_packet: Literal["texts", "audios"]
-
-
-class SpeakerEmbeddingFromDataset(Template):
-    """
-    Template to retrieve and attach speaker embeddings from a Hugging Face dataset.
-    This template extracts a specified embedding (e.g., xvector) from a dataset and attaches
-    it to the `embedding` attribute of each `TextPacket` in a `DataContainer`.
-
-    Usage example:
-
-    agent:
-      name: my_test_agent
-    templates:
-    - template_name: InputTemplate
-      class_name: InputTemplate
-      attributes: {}
-    - template_name: SpeakerEmbeddingFromDataset
-      class_name: SpeakerEmbeddingFromDataset
-      template_input: InputTemplate
-      attributes:
-        dataset_path: '/path/to/hugging/face/dataset'
-        data_cache_dir: /path/to/cache/dir
-        split: validation
-        sample_idx: '1'
-        xvector_key: xvector
-        target_packet: 'audios'
-
-
-
-    """
-
-    AttributesBaseModel = SpeakerEmbeddingFromDatasetAttributes
-    UIProperties = UIPropertiesMetadata(category="HuggingFace", output_type=OutputTypes.AUDIO)
-
-    def execute(self, container: DataContainer) -> DataContainer:
-        """Retrieve and attach speaker embeddings to specified packets in a DataContainer.
-
-        Args:
-            container (DataContainer): The container holding the packets to which the embedding will be
-                attached.
-
-        Returns:
-            DataContainer: The updated container with embeddings attached to the `embedding`
-                attribute of the specified packet type.
-        """
-        packets = getattr(container, self.attributes.target_packet)
-        embeddings_dataset = load_dataset(
-            self.attributes.dataset_path,
-            split=self.attributes.split,
-            cache_dir=self.attributes.data_cache_dir,
-        )
-        speaker_embedding = embeddings_dataset[self.attributes.sample_idx][self.attributes.xvector_key]
-        self.logger.info(
-            f"Attaching embedding from index {self.attributes.sample_idx} to "
-            f"{len(packets)} {self.attributes.target_packet} packets."
-        )
-        for packet in packets:
-            packet.embedding = speaker_embedding
-
-        return container