nv-ingest-api 2025.11.4.dev20251104__tar.gz → 2025.11.8.dev20251108__tar.gz

{nv_ingest_api-2025.11.4.dev20251104/src/nv_ingest_api.egg-info → nv_ingest_api-2025.11.8.dev20251108}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: nv-ingest-api
-Version: 2025.11.4.dev20251104
+Version: 2025.11.8.dev20251108
 Summary: Python module with core document ingestion functions.
 Author-email: Jeremy Dyer <jdyer@nvidia.com>
 License:                                  Apache License

{nv_ingest_api-2025.11.4.dev20251104 → nv_ingest_api-2025.11.8.dev20251108}/src/nv_ingest_api/internal/primitives/nim/nim_client.py

@@ -27,7 +27,7 @@ logger = logging.getLogger(__name__)
 
 # Regex pattern to detect CUDA-related errors in Triton gRPC responses
 CUDA_ERROR_REGEX = re.compile(
-    r"(illegal memory access|illegal instruction|invalid argument|failed to (copy|load|perform) .*: .*|TritonModelException: failed to copy data: .*)",  # noqa: E501
+    r"(model reload|illegal memory access|illegal instruction|invalid argument|failed to (copy|load|perform) .*: .*|TritonModelException: failed to copy data: .*)",  # noqa: E501
     re.IGNORECASE,
 )
 

{nv_ingest_api-2025.11.4.dev20251104 → nv_ingest_api-2025.11.8.dev20251108}/src/nv_ingest_api/internal/schemas/meta/ingest_job_schema.py

@@ -146,6 +146,7 @@ class IngestTaskEmbedSchema(BaseModelNoExt):
     audio_elements_modality: Optional[str] = None
     custom_content_field: Optional[str] = None
     result_target_field: Optional[str] = None
+    dimensions: Optional[int] = None
 
 
 class IngestTaskVdbUploadSchema(BaseModelNoExt):

{nv_ingest_api-2025.11.4.dev20251104 → nv_ingest_api-2025.11.8.dev20251108}/src/nv_ingest_api/internal/schemas/transform/transform_image_caption_schema.py

@@ -10,7 +10,7 @@ class ImageCaptionExtractionSchema(BaseModel):
     api_key: str = Field(default="", repr=False)
     endpoint_url: str = "https://integrate.api.nvidia.com/v1/chat/completions"
     prompt: str = "Caption the content of this image:"
-    model_name: str = "nvidia/llama-3.1-nemotron-nano-vl-8b-v1"
+    model_name: str = "nvidia/nemotron-nano-12b-v2-vl"
     raise_on_failure: bool = False
     model_config = ConfigDict(extra="forbid")
 

{nv_ingest_api-2025.11.4.dev20251104 → nv_ingest_api-2025.11.8.dev20251108}/src/nv_ingest_api/internal/schemas/transform/transform_text_embedding_schema.py

@@ -30,6 +30,7 @@ class TextEmbeddingSchema(BaseModel):
     audio_elements_modality: str = Field(default="text")
     custom_content_field: Optional[str] = None
     result_target_field: Optional[str] = None
+    dimensions: Optional[int] = None
 
     model_config = ConfigDict(extra="forbid")
 

{nv_ingest_api-2025.11.4.dev20251104 → nv_ingest_api-2025.11.8.dev20251108}/src/nv_ingest_api/internal/transform/embed_text.py

@@ -40,6 +40,7 @@ def _make_async_request(
     truncate: str,
     filter_errors: bool,
     modalities: Optional[List[str]] = None,
+    dimensions: Optional[int] = None,
 ) -> list:
     """
     Interacts directly with the NIM embedding service to calculate embeddings for a batch of prompts.
@@ -96,6 +97,7 @@ def _make_async_request(
             model=embedding_model,
             encoding_format=encoding_format,
             extra_body=extra_body,
+            dimensions=dimensions,
         )
 
         response["embedding"] = resp.data
@@ -124,6 +126,7 @@ def _async_request_handler(
     truncate: str,
     filter_errors: bool,
     modalities: Optional[List[str]] = None,
+    dimensions: Optional[int] = None,
 ) -> List[dict]:
     """
     Gathers calculated embedding results from the NIM embedding service concurrently.
@@ -168,6 +171,7 @@ def _async_request_handler(
                 truncate=truncate,
                 filter_errors=filter_errors,
                 modalities=modality_batch,
+                dimensions=dimensions,
             )
             for prompt_batch, modality_batch in zip(prompts, modalities)
         ]
@@ -186,6 +190,7 @@ def _async_runner(
     truncate: str,
     filter_errors: bool,
     modalities: Optional[List[str]] = None,
+    dimensions: Optional[int] = None,
 ) -> dict:
     """
     Concurrently launches all NIM embedding requests and flattens the results.
@@ -224,6 +229,7 @@ def _async_runner(
         truncate,
         filter_errors,
         modalities=modalities,
+        dimensions=dimensions,
     )
 
     flat_results = {"embeddings": [], "info_msgs": []}
@@ -562,6 +568,7 @@ def transform_create_text_embeddings_internal(
     endpoint_url = task_config.get("endpoint_url") or transform_config.embedding_nim_endpoint
     model_name = task_config.get("model_name") or transform_config.embedding_model
     custom_content_field = task_config.get("custom_content_field") or transform_config.custom_content_field
+    dimensions = task_config.get("dimensions") or transform_config.dimensions
 
     if execution_trace_log is None:
         execution_trace_log = {}
@@ -636,6 +643,7 @@ def transform_create_text_embeddings_internal(
                 transform_config.truncate,
                 False,
                 modalities=modality_batches,
+                dimensions=dimensions,
             )
             # Build a simple row index -> embedding map
             embeddings_dict = dict(
@@ -680,6 +688,7 @@ def transform_create_text_embeddings_internal(
                 transform_config.input_type,
                 transform_config.truncate,
                 False,
+                dimensions=dimensions,
             )
             custom_embeddings_dict = dict(
                 zip(

{nv_ingest_api-2025.11.4.dev20251104 → nv_ingest_api-2025.11.8.dev20251108/src/nv_ingest_api.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: nv-ingest-api
-Version: 2025.11.4.dev20251104
+Version: 2025.11.8.dev20251108
 Summary: Python module with core document ingestion functions.
 Author-email: Jeremy Dyer <jdyer@nvidia.com>
 License:                                  Apache License