docsense.models

Model implementations for DocSense.

class EmbeddingModel(model_name='Qwen/Qwen2-7B', device='cuda', max_length=512, normalize_embeddings=True)

Text embedding model using Qwen.

Parameters:

• model_name (str)

• device (str)

• max_length (int)

• normalize_embeddings (bool)

__init__(model_name='Qwen/Qwen2-7B', device='cuda', max_length=512, normalize_embeddings=True)

Initialize the embedding model.

Parameters:

• model_name (str) – Name or path of the Qwen model

• device (str) – Device to run the model on (‘cuda’ or ‘cpu’)

• max_length (int) – Maximum sequence length for tokenization

• normalize_embeddings (bool) – Whether to L2-normalize the embeddings

encode(texts, batch_size=8)

Generate embeddings for the given texts.

Parameters:

• texts (Union[str, List[str]]) – Single text or list of texts to encode

• batch_size (int) – Number of texts to process at once

Return type:

ndarray

Returns:

numpy array of embeddings with shape (num_texts, embedding_dim)

get_embedding_dim()

Get the dimension of the embeddings.

Return type:

int

Returns:

Embedding dimension

class LLMModel(model_name='Qwen/Qwen2-7B', device='cuda', max_length=2048, temperature=0.0, top_p=1.0, repetition_penalty=1.1)

Wrapper for Qwen language model.

Parameters:

• model_name (str)

• device (str)

• max_length (int)

• temperature (float)

• top_p (float)

• repetition_penalty (float)

__init__(model_name='Qwen/Qwen2-7B', device='cuda', max_length=2048, temperature=0.0, top_p=1.0, repetition_penalty=1.1)

Initialize the LLM model.

Parameters:

• model_name (str) – Name of the Qwen model to use

• device (str) – Device to run the model on (‘cuda’ or ‘cpu’)

• max_length (int) – Maximum sequence length for generation

• temperature (float) – Sampling temperature (0.0 for deterministic output)

• top_p (float) – Nucleus sampling parameter (1.0 for no filtering)

• repetition_penalty (float) – Penalty for repeating tokens

generate(question, context)

Generate answer based on context.

Parameters:

• question (str)

• context (List[str])

Return type:

Dict[str, Any]

get_max_length()

Get the maximum sequence length.

Return type:

int

Returns:

Maximum sequence length

Modules

embeddings	Text embedding model implementation using Qwen.
llm	LLM (Large Language Model) wrapper implementation.