cocoindex 0.2.18__cp311-abi3-manylinux_2_28_aarch64.whl → 0.2.20__cp311-abi3-manylinux_2_28_aarch64.whl

cocoindex/llm.py

@@ -14,6 +14,7 @@ class LlmApiType(Enum):
     OPEN_ROUTER = "OpenRouter"
     VOYAGE = "Voyage"
     VLLM = "Vllm"
+    BEDROCK = "Bedrock"
 
 
 @dataclass

cocoindex/sources/_engine_builtin_specs.py

@@ -100,3 +100,6 @@ class Postgres(op.SourceSpec):
 
     # Optional: when set, supports change capture from PostgreSQL notification.
     notification: PostgresNotification | None = None
+
+    # Optional: SQL expression filter for rows (arbitrary SQL boolean expression)
+    filter: str | None = None

{cocoindex-0.2.18.dist-info → cocoindex-0.2.20.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: cocoindex
-Version: 0.2.18
+Version: 0.2.20
 Classifier: Development Status :: 3 - Alpha
 Classifier: License :: OSI Approved :: Apache Software License
 Classifier: Operating System :: OS Independent
@@ -75,7 +75,6 @@ Project-URL: Homepage, https://cocoindex.io/
     <a href="https://trendshift.io/repositories/13939" target="_blank"><img src="https://trendshift.io/api/badge/repositories/13939" alt="cocoindex-io%2Fcocoindex | Trendshift" style="width: 250px; height: 55px;" width="250" height="55"/></a>
 </div>
 
-
 Ultra performant data transformation framework for AI, with core engine written in Rust. Support incremental processing and data lineage out-of-box.  Exceptional developer velocity. Production-ready at day 0.
 
 ⭐ Drop a star to help us grow!
@@ -113,9 +112,8 @@ CocoIndex makes it effortless to transform data with AI, and keep source data an
 
 </br>
 
-
-
 ## Exceptional velocity
+
 Just declare transformation in dataflow with ~100 lines of python
 
 ```python
@@ -139,6 +137,7 @@ CocoIndex follows the idea of [Dataflow](https://en.wikipedia.org/wiki/Dataflow_
 **Particularly**, developers don't explicitly mutate data by creating, updating and deleting. They just need to define transformation/formula for a set of source data.
 
 ## Plug-and-Play Building Blocks
+
 Native builtins for different source, targets and transformations. Standardize interface, make it 1-line code switch between different components - as easy as assembling building blocks.
 
 <p align="center">
@@ -146,6 +145,7 @@ Native builtins for different source, targets and transformations. Standardize i
 </p>
 
 ## Data Freshness
+
 CocoIndex keep source data and target in sync effortlessly.
 
 <p align="center">
@@ -153,11 +153,14 @@ CocoIndex keep source data and target in sync effortlessly.
 </p>
 
 It has out-of-box support for incremental indexing:
+
 - minimal recomputation on source or logic change.
 - (re-)processing necessary portions; reuse cache when possible
 
-## Quick Start:
+## Quick Start
+
 If you're new to CocoIndex, we recommend checking out
+
 - 📖 [Documentation](https://cocoindex.io/docs)
 - ⚡  [Quick Start Guide](https://cocoindex.io/docs/getting_started/quickstart)
 - 🎬 [Quick Start Video Tutorial](https://youtu.be/gv5R8nOXsWU?si=9ioeKYkMEnYevTXT)
@@ -172,7 +175,6 @@ pip install -U cocoindex
 
 2. [Install Postgres](https://cocoindex.io/docs/getting_started/installation#-install-postgres) if you don't have one. CocoIndex uses it for incremental processing.
 
-
 ## Define data flow
 
 Follow [Quick Start Guide](https://cocoindex.io/docs/getting_started/quickstart) to define your first indexing flow. An example flow looks like:
@@ -228,6 +230,7 @@ It defines an index flow like this:
 | [Text Embedding](examples/text_embedding) | Index text documents with embeddings for semantic search |
 | [Code Embedding](examples/code_embedding) | Index code embeddings for semantic search |
 | [PDF Embedding](examples/pdf_embedding) | Parse PDF and index text embeddings for semantic search |
+| [PDF Elements Embedding](examples/pdf_elements_embedding) | Extract text and images from PDFs; embed text with SentenceTransformers and images with CLIP; store in Qdrant for multimodal search |
 | [Manuals LLM Extraction](examples/manuals_llm_extraction) | Extract structured information from a manual using LLM |
 | [Amazon S3 Embedding](examples/amazon_s3_embedding) | Index text documents from Amazon S3 |
 | [Azure Blob Storage Embedding](examples/azure_blob_embedding) | Index text documents from Azure Blob Storage |
@@ -244,16 +247,18 @@ It defines an index flow like this:
 | [Custom Output Files](examples/custom_output_files) | Convert markdown files to HTML files and save them to a local directory, using *CocoIndex Custom Targets* |
 | [Patient intake form extraction](examples/patient_intake_extraction) | Use LLM to extract structured data from patient intake forms with different formats |
 
-
 More coming and stay tuned 👀!
 
 ## 📖 Documentation
+
 For detailed documentation, visit [CocoIndex Documentation](https://cocoindex.io/docs), including a [Quickstart guide](https://cocoindex.io/docs/getting_started/quickstart).
 
 ## 🤝 Contributing
+
 We love contributions from our community ❤️. For details on contributing or running the project for development, check out our [contributing guide](https://cocoindex.io/docs/about/contributing).
 
 ## 👥 Community
+
 Welcome with a huge coconut hug 🥥⋆｡˚🤗. We are super excited for community contributions of all kinds - whether it's code improvements, documentation updates, issue reports, feature requests, and discussions in our Discord.
 
 Join our community here:
@@ -263,9 +268,11 @@ Join our community here:
 - ▶️ [Subscribe to our YouTube channel](https://www.youtube.com/@cocoindex-io)
 - 📜 [Read our blog posts](https://cocoindex.io/blogs/)
 
-## Support us:
+## Support us
+
 We are constantly improving, and more features and examples are coming soon. If you love this project, please drop us a star ⭐ at GitHub repo [![GitHub](https://img.shields.io/github/stars/cocoindex-io/cocoindex?color=5B5BD6)](https://github.com/cocoindex-io/cocoindex) to stay tuned and help us grow.
 
 ## License
+
 CocoIndex is Apache 2.0 licensed.
 

{cocoindex-0.2.18.dist-info → cocoindex-0.2.20.dist-info}/RECORD

@@ -1,22 +1,21 @@
-cocoindex-0.2.18.dist-info/METADATA,sha256=W5VPkE2KMAEDvAV-JO66Ca8EAMSbP_w1QykrjRXejuY,13444
-cocoindex-0.2.18.dist-info/WHEEL,sha256=T94Vf-8hBLuJYmQaKIvspCD375-5CHbUeNmaNVtwQwY,108
-cocoindex-0.2.18.dist-info/entry_points.txt,sha256=_NretjYVzBdNTn7dK-zgwr7YfG2afz1u1uSE-5bZXF8,46
-cocoindex-0.2.18.dist-info/licenses/THIRD_PARTY_NOTICES.html,sha256=MpyWY1rkfN94bYXo7IgLpY4F9cc1S4Vn5muesIpG5VM,719620
+cocoindex-0.2.20.dist-info/METADATA,sha256=PMLqa8rFhhAtRQCDWSvUSQbKy3vLYHdHatftA49W0e4,13644
+cocoindex-0.2.20.dist-info/WHEEL,sha256=T94Vf-8hBLuJYmQaKIvspCD375-5CHbUeNmaNVtwQwY,108
+cocoindex-0.2.20.dist-info/entry_points.txt,sha256=_NretjYVzBdNTn7dK-zgwr7YfG2afz1u1uSE-5bZXF8,46
+cocoindex-0.2.20.dist-info/licenses/THIRD_PARTY_NOTICES.html,sha256=SJ-7q0eqT40cFyT1cXqQkxWocFEuLT6PrETn5dhxiX8,719620
 cocoindex/__init__.py,sha256=6qZWVkK4WZ01BIAg3CPh_bRRdA6Clk4d4Q6OnZ2jFa4,2630
-cocoindex/_engine.abi3.so,sha256=Foh8iKEC5LxeHl1EzzUZfAmBqDbbyJ0f1OXG6BVWTK8,74683152
+cocoindex/_engine.abi3.so,sha256=YiQOMxjygiJrhOFTPvtHZ_mpvpfvDRy2KEhrRG_XwBQ,74720848
 cocoindex/auth_registry.py,sha256=g-uLDWLYW5NMbYe7q4Y-sU5dSyrlJXBEciyWtAiP9KE,1340
 cocoindex/cli.py,sha256=19IszBXOzqGn0xOV1SaS-oR9NupTmIm18uzFNET7NTQ,23978
 cocoindex/engine_object.py,sha256=5YTuWoR3WILhyt3PW-d9es3MAas_xD6tZZqvipN-sjg,10050
 cocoindex/engine_value.py,sha256=8M7MbwVG2bfd3kFptGGbQHBAp9pD3TVjrBiBDOAhD5M,23211
 cocoindex/flow.py,sha256=JWPTR2G6TdPJkO5ZlrCcyDyQ8utUS4zZWNR8zsHTeW8,40074
-cocoindex/functions.py,sha256=V4ljBnCprvA25XlCVvNLwK5ergXiEcKU76jkOGC-X3A,12882
 cocoindex/functions/__init__.py,sha256=V2IF4h-Cqq4OD_GN3Oqdry-FArORyRCKmqJ7g5UlJr8,1021
 cocoindex/functions/_engine_builtin_specs.py,sha256=WpCGrjUfJBa8xZP5JiEmA8kLu7fp9Rcs7ynpuJmvSGg,1786
 cocoindex/functions/colpali.py,sha256=oACyG3qG2dquyCJ6bT7FkMkua5rXDLSxnOHcgoz9waU,8865
 cocoindex/functions/sbert.py,sha256=1z5OJT-blXT6tVN5vEvEzvYAzOnzs1RCnu1UbCUP6wM,2162
 cocoindex/index.py,sha256=tz5ilvmOp0BtroGehCQDqWK_pIX9m6ghkhcxsDVU8WE,982
 cocoindex/lib.py,sha256=spfdU4IbzdffHyGdrQPIw_qGo9aX0OAAboqsjj8bTiQ,2290
-cocoindex/llm.py,sha256=Pv_cdnRngTLtuLU9AUmS8izIHhcKVnuBNolC33f9BDI,851
+cocoindex/llm.py,sha256=8ZdJhOmhdb2xEcCxk6rDpnj6hlhCyFBmJdhCNMqAOP4,875
 cocoindex/op.py,sha256=Ycvr6lJf7hcCCjYUqHtXZqzSeDD-FQdP3_jcmZUV_zI,26896
 cocoindex/py.typed,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 cocoindex/query_handler.py,sha256=X-SQT71LHiOOXn6-TJlQcGodJk-iT8p_1TcIMvRLBRI,1344
@@ -24,7 +23,7 @@ cocoindex/runtime.py,sha256=4NxcltaDZvA3RR3Pnt6gH_f99jcWSyMH_1Xi5BjbtwY,1342
 cocoindex/setting.py,sha256=1Dx8ktjwf-8BiXrbsmfn5Mzudb2SQYqFdRnSNGVKaLk,4960
 cocoindex/setup.py,sha256=7uIHKN4FOCuoidPXcKyGTrkqpkl9luL49-6UcnMxYzw,3068
 cocoindex/sources/__init__.py,sha256=Yu9VHNaGlOEE3jpqfIseswsg25Le3HzwDr6XJAn22Ns,78
-cocoindex/sources/_engine_builtin_specs.py,sha256=NI4Uq6ffX5nfsgzneDErSBK4YH8hccaYj-InEBihJpo,3191
+cocoindex/sources/_engine_builtin_specs.py,sha256=s4AxMLi2j3ZHmzACVEGAdVe05gY8PRZ_mYMxWR7scDY,3304
 cocoindex/subprocess_exec.py,sha256=r1xO84uek4VP4I6i87JMwsH5xFm3vKW0ABvgn0jskt4,10088
 cocoindex/targets/__init__.py,sha256=HQG7I4U0xQhHiYctiUvwEBLxT2727oHP3xwrqotjmhk,78
 cocoindex/targets/_engine_builtin_specs.py,sha256=glXUN5bj11Jxky1VPvmGnWnMHXTQWEh08INcbldo3F4,3375
@@ -40,4 +39,4 @@ cocoindex/typing.py,sha256=so_RusbhBmg_uLoZTY7W_pqU0aIJwFarkTF5NQufl4o,23944
 cocoindex/user_app_loader.py,sha256=bc3Af-gYRxJ9GpObtpjegZY855oQBCv5FGkrkWV2yGY,1873
 cocoindex/utils.py,sha256=hUhX-XV6XGCtJSEIpBOuDv6VvqImwPlgBxztBTw7u0U,598
 cocoindex/validation.py,sha256=PZnJoby4sLbsmPv9fOjOQXuefjfZ7gmtsiTGU8SH-tc,3090
-cocoindex-0.2.18.dist-info/RECORD,,
+cocoindex-0.2.20.dist-info/RECORD,,

{cocoindex-0.2.18.dist-info → cocoindex-0.2.20.dist-info}/licenses/THIRD_PARTY_NOTICES.html

@@ -2428,7 +2428,7 @@ Software.
                 <h3 id="Apache-2.0">Apache License 2.0</h3>
                 <h4>Used by:</h4>
                 <ul class="license-used-by">
-                    <li><a href=" https://crates.io/crates/cocoindex ">cocoindex 0.2.18</a></li>
+                    <li><a href=" https://crates.io/crates/cocoindex ">cocoindex 0.2.20</a></li>
                     <li><a href=" https://github.com/awesomized/crc-fast-rust ">crc-fast 1.3.0</a></li>
                     <li><a href=" https://github.com/qdrant/rust-client ">qdrant-client 1.15.0</a></li>
                 </ul>

cocoindex/functions.py

@@ -1,375 +0,0 @@
-"""All builtin functions."""
-
-import dataclasses
-import functools
-from typing import Any, Literal
-
-import numpy as np
-from numpy.typing import NDArray
-
-from . import llm, op
-from .typing import Vector
-
-
-class ParseJson(op.FunctionSpec):
-    """Parse a text into a JSON object."""
-
-
-@dataclasses.dataclass
-class CustomLanguageSpec:
-    """Custom language specification."""
-
-    language_name: str
-    separators_regex: list[str]
-    aliases: list[str] = dataclasses.field(default_factory=list)
-
-
-@dataclasses.dataclass
-class ColPaliModelInfo:
-    """Data structure for ColPali model and processor."""
-
-    model: Any
-    processor: Any
-    dimension: int
-    device: Any
-
-
-class SplitRecursively(op.FunctionSpec):
-    """Split a document (in string) recursively."""
-
-    custom_languages: list[CustomLanguageSpec] = dataclasses.field(default_factory=list)
-
-
-class SplitBySeparators(op.FunctionSpec):
-    """
-    Split text by specified regex separators only.
-    Output schema matches SplitRecursively for drop-in compatibility:
-        KTable rows with fields: location (Range), text (Str), start, end.
-    Args:
-        separators_regex: list[str]  # e.g., [r"\\n\\n+"]
-        keep_separator: Literal["NONE", "LEFT", "RIGHT"] = "NONE"
-        include_empty: bool = False
-        trim: bool = True
-    """
-
-    separators_regex: list[str] = dataclasses.field(default_factory=list)
-    keep_separator: Literal["NONE", "LEFT", "RIGHT"] = "NONE"
-    include_empty: bool = False
-    trim: bool = True
-
-
-class EmbedText(op.FunctionSpec):
-    """Embed a text into a vector space."""
-
-    api_type: llm.LlmApiType
-    model: str
-    address: str | None = None
-    output_dimension: int | None = None
-    task_type: str | None = None
-    api_config: llm.VertexAiConfig | None = None
-
-
-class ExtractByLlm(op.FunctionSpec):
-    """Extract information from a text using a LLM."""
-
-    llm_spec: llm.LlmSpec
-    output_type: type
-    instruction: str | None = None
-
-
-class SentenceTransformerEmbed(op.FunctionSpec):
-    """
-    `SentenceTransformerEmbed` embeds a text into a vector space using the [SentenceTransformer](https://huggingface.co/sentence-transformers) library.
-
-    Args:
-
-        model: The name of the SentenceTransformer model to use.
-        args: Additional arguments to pass to the SentenceTransformer constructor. e.g. {"trust_remote_code": True}
-
-    Note:
-        This function requires the optional sentence-transformers dependency.
-        Install it with: pip install 'cocoindex[embeddings]'
-    """
-
-    model: str
-    args: dict[str, Any] | None = None
-
-
-@op.executor_class(
-    gpu=True,
-    cache=True,
-    behavior_version=1,
-    arg_relationship=(op.ArgRelationship.EMBEDDING_ORIGIN_TEXT, "text"),
-)
-class SentenceTransformerEmbedExecutor:
-    """Executor for SentenceTransformerEmbed."""
-
-    spec: SentenceTransformerEmbed
-    _model: Any | None = None
-
-    def analyze(self) -> type:
-        try:
-            # Only import sentence_transformers locally when it's needed, as its import is very slow.
-            import sentence_transformers  # pylint: disable=import-outside-toplevel
-        except ImportError as e:
-            raise ImportError(
-                "sentence_transformers is required for SentenceTransformerEmbed function. "
-                "Install it with one of these commands:\n"
-                "  pip install 'cocoindex[embeddings]'\n"
-                "  pip install sentence-transformers"
-            ) from e
-
-        args = self.spec.args or {}
-        self._model = sentence_transformers.SentenceTransformer(self.spec.model, **args)
-        dim = self._model.get_sentence_embedding_dimension()
-        return Vector[np.float32, Literal[dim]]  # type: ignore
-
-    def __call__(self, text: str) -> NDArray[np.float32]:
-        assert self._model is not None
-        result: NDArray[np.float32] = self._model.encode(text, convert_to_numpy=True)
-        return result
-
-
-@functools.cache
-def _get_colpali_model_and_processor(model_name: str) -> ColPaliModelInfo:
-    """Get or load ColPali model and processor, with caching."""
-    try:
-        from colpali_engine.models import (  # type: ignore[import-untyped]
-            ColPali,
-            ColPaliProcessor,
-            ColQwen2,
-            ColQwen2Processor,
-            ColQwen2_5,
-            ColQwen2_5_Processor,
-            ColIdefics3,
-            ColIdefics3Processor,
-        )
-        from colpali_engine.utils.torch_utils import get_torch_device  # type: ignore[import-untyped]
-        import torch
-    except ImportError as e:
-        raise ImportError(
-            "ColVision models are not available. Make sure cocoindex is installed with ColPali support."
-        ) from e
-
-    device = get_torch_device("auto")
-
-    # Manual model detection based on model name
-    model_name_lower = model_name.lower()
-
-    try:
-        if "qwen2.5" in model_name_lower:
-            model = ColQwen2_5.from_pretrained(
-                model_name,
-                torch_dtype=torch.bfloat16,
-                device_map=device,
-            ).eval()
-            processor = ColQwen2_5_Processor.from_pretrained(model_name)
-        elif "qwen2" in model_name_lower:
-            model = ColQwen2.from_pretrained(
-                model_name,
-                torch_dtype=torch.bfloat16,
-                device_map=device,
-            ).eval()
-            processor = ColQwen2Processor.from_pretrained(model_name)
-        elif "colsmol" in model_name_lower or "smol" in model_name_lower:
-            # ColSmol models use Idefics3 architecture
-            model = ColIdefics3.from_pretrained(
-                model_name,
-                torch_dtype=torch.bfloat16,
-                device_map=device,
-            ).eval()
-            processor = ColIdefics3Processor.from_pretrained(model_name)
-        else:
-            # Default to ColPali
-            model = ColPali.from_pretrained(
-                model_name,
-                torch_dtype=torch.bfloat16,
-                device_map=device,
-            ).eval()
-            processor = ColPaliProcessor.from_pretrained(model_name)
-
-    except Exception as e:
-        raise RuntimeError(f"Failed to load model {model_name}: {e}")
-
-    # Get dimension from the actual model
-    dimension = _detect_colpali_dimension(model, processor, device)
-
-    return ColPaliModelInfo(
-        model=model,
-        processor=processor,
-        dimension=dimension,
-        device=device,
-    )
-
-
-def _detect_colpali_dimension(model: Any, processor: Any, device: Any) -> int:
-    """Detect ColPali embedding dimension from the actual model config."""
-    # Try to access embedding dimension
-    if hasattr(model.config, "embedding_dim"):
-        dim = model.config.embedding_dim
-    else:
-        # Fallback: infer from output shape with dummy data
-        from PIL import Image
-        import numpy as np
-        import torch
-
-        dummy_img = Image.fromarray(np.zeros((224, 224, 3), np.uint8))
-        # Use the processor to process the dummy image
-        processed = processor.process_images([dummy_img]).to(device)
-        with torch.no_grad():
-            output = model(**processed)
-        dim = int(output.shape[-1])
-    if isinstance(dim, int):
-        return dim
-    else:
-        raise ValueError(f"Expected integer dimension, got {type(dim)}: {dim}")
-    return dim
-
-
-class ColPaliEmbedImage(op.FunctionSpec):
-    """
-    `ColPaliEmbedImage` embeds images using ColVision multimodal models.
-
-    Supports ALL models available in the colpali-engine library, including:
-    - ColPali models (colpali-*): PaliGemma-based, best for general document retrieval
-    - ColQwen2 models (colqwen-*): Qwen2-VL-based, excellent for multilingual text (29+ languages) and general vision
-    - ColSmol models (colsmol-*): Lightweight, good for resource-constrained environments
-    - Any future ColVision models supported by colpali-engine
-
-    These models use late interaction between image patch embeddings and text token
-    embeddings for retrieval.
-
-    Args:
-        model: Any ColVision model name supported by colpali-engine
-               (e.g., "vidore/colpali-v1.2", "vidore/colqwen2.5-v0.2", "vidore/colsmol-v1.0")
-               See https://github.com/illuin-tech/colpali for the complete list of supported models.
-
-    Note:
-        This function requires the optional colpali-engine dependency.
-        Install it with: pip install 'cocoindex[colpali]'
-    """
-
-    model: str
-
-
-@op.executor_class(
-    gpu=True,
-    cache=True,
-    behavior_version=1,
-)
-class ColPaliEmbedImageExecutor:
-    """Executor for ColVision image embedding (ColPali, ColQwen2, ColSmol, etc.)."""
-
-    spec: ColPaliEmbedImage
-    _model_info: ColPaliModelInfo
-
-    def analyze(self) -> type:
-        # Get shared model and dimension
-        self._model_info = _get_colpali_model_and_processor(self.spec.model)
-
-        # Return multi-vector type: Variable patches x Fixed hidden dimension
-        dimension = self._model_info.dimension
-        return Vector[Vector[np.float32, Literal[dimension]]]  # type: ignore
-
-    def __call__(self, img_bytes: bytes) -> Any:
-        try:
-            from PIL import Image
-            import torch
-            import io
-        except ImportError as e:
-            raise ImportError(
-                "Required dependencies (PIL, torch) are missing for ColVision image embedding."
-            ) from e
-
-        model = self._model_info.model
-        processor = self._model_info.processor
-        device = self._model_info.device
-
-        pil_image = Image.open(io.BytesIO(img_bytes)).convert("RGB")
-        inputs = processor.process_images([pil_image]).to(device)
-        with torch.no_grad():
-            embeddings = model(**inputs)
-
-        # Return multi-vector format: [patches, hidden_dim]
-        if len(embeddings.shape) != 3:
-            raise ValueError(
-                f"Expected 3D tensor [batch, patches, hidden_dim], got shape {embeddings.shape}"
-            )
-
-        # Keep patch-level embeddings: [batch, patches, hidden_dim] -> [patches, hidden_dim]
-        patch_embeddings = embeddings[0]  # Remove batch dimension
-
-        return patch_embeddings.cpu().to(torch.float32).numpy()
-
-
-class ColPaliEmbedQuery(op.FunctionSpec):
-    """
-    `ColPaliEmbedQuery` embeds text queries using ColVision multimodal models.
-
-    Supports ALL models available in the colpali-engine library, including:
-    - ColPali models (colpali-*): PaliGemma-based, best for general document retrieval
-    - ColQwen2 models (colqwen-*): Qwen2-VL-based, excellent for multilingual text (29+ languages) and general vision
-    - ColSmol models (colsmol-*): Lightweight, good for resource-constrained environments
-    - Any future ColVision models supported by colpali-engine
-
-    This produces query embeddings compatible with ColVision image embeddings
-    for late interaction scoring (MaxSim).
-
-    Args:
-        model: Any ColVision model name supported by colpali-engine
-               (e.g., "vidore/colpali-v1.2", "vidore/colqwen2.5-v0.2", "vidore/colsmol-v1.0")
-               See https://github.com/illuin-tech/colpali for the complete list of supported models.
-
-    Note:
-        This function requires the optional colpali-engine dependency.
-        Install it with: pip install 'cocoindex[colpali]'
-    """
-
-    model: str
-
-
-@op.executor_class(
-    gpu=True,
-    cache=True,
-    behavior_version=1,
-)
-class ColPaliEmbedQueryExecutor:
-    """Executor for ColVision query embedding (ColPali, ColQwen2, ColSmol, etc.)."""
-
-    spec: ColPaliEmbedQuery
-    _model_info: ColPaliModelInfo
-
-    def analyze(self) -> type:
-        # Get shared model and dimension
-        self._model_info = _get_colpali_model_and_processor(self.spec.model)
-
-        # Return multi-vector type: Variable tokens x Fixed hidden dimension
-        dimension = self._model_info.dimension
-        return Vector[Vector[np.float32, Literal[dimension]]]  # type: ignore
-
-    def __call__(self, query: str) -> Any:
-        try:
-            import torch
-        except ImportError as e:
-            raise ImportError(
-                "Required dependencies (torch) are missing for ColVision query embedding."
-            ) from e
-
-        model = self._model_info.model
-        processor = self._model_info.processor
-        device = self._model_info.device
-
-        inputs = processor.process_queries([query]).to(device)
-        with torch.no_grad():
-            embeddings = model(**inputs)
-
-        # Return multi-vector format: [tokens, hidden_dim]
-        if len(embeddings.shape) != 3:
-            raise ValueError(
-                f"Expected 3D tensor [batch, tokens, hidden_dim], got shape {embeddings.shape}"
-            )
-
-        # Keep token-level embeddings: [batch, tokens, hidden_dim] -> [tokens, hidden_dim]
-        token_embeddings = embeddings[0]  # Remove batch dimension
-
-        return token_embeddings.cpu().to(torch.float32).numpy()