ColQwen2

The ColQwen2 is a Vision Language Model (VLM) that uses a novel architecture and training strategy to efficiently index documents from their visual features. RBLN NPUs can accelerate ColQwen2 model inference using Optimum RBLN.

API Reference

Classes

RBLNColQwen2ForRetrieval

Bases: RBLNModel

RBLNColQwen2ForRetrieval is a model for document retrieval using vision-language models. This model inherits from [RBLNModel]. Check the superclass documentation for the generic methods the library implements for all its models.

A class to convert and run pre-trained transformers based ColQwen2ForRetrieval model on RBLN devices. It implements the methods to convert a pre-trained transformers ColQwen2ForRetrieval model into a RBLN transformer model by:

• transferring the checkpoint weights of the original into an optimized RBLN graph,
• compiling the resulting graph using the RBLN compiler.

Examples:

import torch
from PIL import Image
from transformers import ColQwen2Processor

from optimum.rbln import RBLNColQwen2ForRetrieval, RBLNColQwen2ForRetrievalConfig

rbln_config = {
    "vlm": {
        "visual": {
            "max_seq_lens": 6400,
        },
        "tensor_parallel_size": 4,
        "kvcache_partition_len": 16384,
        "max_seq_len": 16384 * 7,
    },
}
model = RBLNColQwen2ForRetrieval.from_pretrained("vidore/colqwen2-v1.0-hf", rbln_config=config)
model.save_pretrained("compiled-colqwen2-v1.0-hf")

# The document page screenshots from your corpus. Below are dummy images.
images = [
    Image.new("RGB", (128, 128), color="white"),
    Image.new("RGB", (64, 32), color="black"),
]
processor = ColQwen2Processor.from_pretrained("vidore/colqwen2-v1.0-hf")

queries = [
    "When was the United States Declaration of Independence proclaimed?",
    "Who printed the edition of Romeo and Juliet?",
]
inputs_images = processor(images=images)
inputs_text = processor(text=queries)

# Forward pass
with torch.no_grad():
    image_embeddings = model(**inputs_images).embeddings
    query_embeddings = model(**inputs_text).embeddings

scores = processor.score_retrieval(query_embeddings, image_embeddings)
print("Retrieval scores (query x image):")
print(scores)

Functions

from_model(model, config=None, rbln_config=None, model_save_dir=None, subfolder='', **kwargs) classmethod

Converts and compiles a pre-trained HuggingFace library model into a RBLN model. This method performs the actual model conversion and compilation process.

Parameters:

Name	Type	Description	Default
model	PreTrainedModel	The PyTorch model to be compiled. The object must be an instance of the HuggingFace transformers PreTrainedModel class.	required
config	Optional[PretrainedConfig]	The configuration object associated with the model.	None
rbln_config	Optional[Union[RBLNModelConfig, Dict]]	Configuration for RBLN model compilation and runtime. This can be provided as a dictionary or an instance of the model's configuration class (e.g., RBLNLlamaForCausalLMConfig for Llama models). For detailed configuration options, see the specific model's configuration class documentation.	None
kwargs	Any	Additional keyword arguments. Arguments with the prefix rbln_ are passed to rbln_config, while the remaining arguments are passed to the HuggingFace library.	{}

The method performs the following steps:

1. Compiles the PyTorch model into an optimized RBLN graph
2. Configures the model for the specified NPU device
3. Creates the necessary runtime objects if requested
4. Saves the compiled model and configurations

Returns:

Type	Description
RBLNModel	A RBLN model instance ready for inference on RBLN NPU devices.

forward(input_ids=None, inputs_embeds=None, attention_mask=None, pixel_values=None, image_grid_thw=None, output_hidden_states=None, return_dict=None, **kwargs)

Runs a ColQwen2 retrieval forward pass on text tokens and optional visual inputs.

Parameters:

Name	Type	Description	Default
input_ids	LongTensor	Indices of the textual tokens. Mutually exclusive with inputs_embeds.	None
inputs_embeds	FloatTensor	Pre-computed embeddings fed directly into the language model.	None
attention_mask	Tensor	Mask that selects which token positions contribute to the loss/embeddings.	None
pixel_values	Tensor	Flattened image patches produced by ColQwen2Processor for document pages.	None
image_grid_thw	LongTensor	Per-image (t, h, w) grid metadata that allows unpadding of pixel_values.	None
output_hidden_states	bool	If True, expose intermediate decoder hidden states.	None
return_dict	bool	If True, return a ColQwen2ForRetrievalOutput; otherwise return a tuple.	None
**kwargs	dict[str, Any]	Extra multimodal args forwarded to the wrapped VLM (e.g. pixel_values_videos, video_grid_thw, second_per_grid_ts).	{}

Returns:

Type	Description
Union[Tuple, ColQwen2ForRetrievalOutput]	Dataclass containing the embeddings and hidden states of the VLM model.

from_pretrained(model_id, export=None, rbln_config=None, **kwargs) classmethod

The from_pretrained() function is utilized in its standard form as in the HuggingFace transformers library. User can use this function to load a pre-trained model from the HuggingFace library and convert it to a RBLN model to be run on RBLN NPUs.

Parameters:

Name	Type	Description	Default
model_id	Union[str, Path]	The model id of the pre-trained model to be loaded. It can be downloaded from the HuggingFace model hub or a local path, or a model id of a compiled model using the RBLN Compiler.	required
export	Optional[bool]	A boolean flag to indicate whether the model should be compiled. If None, it will be determined based on the existence of the compiled model files in the model_id.	None
rbln_config	Optional[Union[Dict, RBLNModelConfig]]	Configuration for RBLN model compilation and runtime. This can be provided as a dictionary or an instance of the model's configuration class (e.g., RBLNLlamaForCausalLMConfig for Llama models). For detailed configuration options, see the specific model's configuration class documentation.	None
kwargs	Any	Additional keyword arguments. Arguments with the prefix rbln_ are passed to rbln_config, while the remaining arguments are passed to the HuggingFace library.	{}

Returns:

Type	Description
RBLNModel	A RBLN model instance ready for inference on RBLN NPU devices.

save_pretrained(save_directory, push_to_hub=False, **kwargs)

Saves a model and its configuration file to a directory, so that it can be re-loaded using the [~optimum.rbln.modeling_base.RBLNBaseModel.from_pretrained] class method.

Parameters:

Name	Type	Description	Default
save_directory	Union[str, Path]	Directory where to save the model file.	required
push_to_hub	bool	Whether or not to push your model to the HuggingFace model hub after saving it.	False

Classes

RBLNColQwen2ForRetrievalConfig

Bases: RBLNDecoderOnlyModelConfig

Configuration class for RBLN ColQwen2 models for document retrieval.

This class extends RBLNModelConfig with specific configurations for ColQwen2 models, including vision tower settings and multi-sequence length support.

Example usage

from optimum.rbln import RBLNColQwen2ForRetrievalConfig, RBLNColQwen2ForRetrievalConfig

# Create a configuration object
config = RBLNColQwen2ForRetrievalConfig(
    vlm = {
        "visual": {
            "max_seq_lens": 6400,
            "device": 0,
        },
        "max_seq_len": 32_768,
        "tensor_parallel_size": 4,
        "device": [0, 1, 2, 3],
        "output_hidden_states": False,
        }
)

# Use the configuration with from_pretrained
model = RBLNColQwen2ForRetrieval.from_pretrained(
    "vidore/colqwen2-v1.0-hf",
    export=True,
    rbln_config=config
)

Functions

__init__(batch_size=None, output_hidden_states=None, vlm=None, **kwargs)

Parameters:

Name	Type	Description	Default
batch_size	Optional[int]	The batch size for the model.	None
output_hidden_states	Optional[bool]	Whether to output the hidden states of the VLM model.	None
vlm	Optional[RBLNModelConfig]	Configuration for the VLM component.	None
kwargs		Additional arguments passed to the parent RBLNModelConfig.	{}

Raises: ValueError: If batch_size is not a positive integer.

load(path, **kwargs) classmethod

Load a RBLNModelConfig from a path.

Parameters:

Name	Type	Description	Default
path	str	Path to the RBLNModelConfig file or directory containing the config file.	required
kwargs	Any	Additional keyword arguments to override configuration values. Keys starting with 'rbln_' will have the prefix removed and be used to update the configuration.	{}

Returns:

Name	Type	Description
RBLNModelConfig	RBLNModelConfig	The loaded configuration instance.

Note

This method loads the configuration from the specified path and applies any provided overrides. If the loaded configuration class doesn't match the expected class, a warning will be logged.