Model Overview
ICT-TIME-and-Querit/BOOM_4B_v1 is a 4 billion parameter text embedding model, adapted from the Qwen/Qwen3-4B architecture. It is designed to generate robust general text embeddings, utilizing a 32k context length and employing last token pooling for embedding extraction. This model is particularly optimized for tasks requiring semantic similarity and understanding.
Key Capabilities
- Text Embedding Generation: Creates high-quality vector representations of text for various downstream tasks.
- Long Context Handling: Supports a substantial context length of 32,000 tokens, allowing for the processing of longer documents.
- Semantic Search and Retrieval: Excels in tasks like retrieving relevant passages given a query, as highlighted by its usage examples.
- MTEB Benchmark Performance: While its MTEB score of 63.52 is competitive, it is noted that the Qwen3-Embedding-4B model, from which it is adapted, achieves a higher score of 69.45, suggesting potential for further optimization or specific use-case advantages for BOOM_4B_v1.
Usage and Integration
This model can be easily integrated using sentence-transformers or directly with transformers. It supports flash_attention_2 for improved acceleration and memory efficiency. Queries can be enhanced with specific prompts to guide the embedding process, as shown in the provided code examples.
Good for
- Information Retrieval Systems: Ideal for building search engines or recommendation systems that rely on semantic matching.
- Text Classification and Clustering: Suitable for organizing and categorizing large volumes of text data.
- Semantic Similarity Tasks: Effective for identifying related texts or paraphrases.
- Developers seeking a 4B parameter embedding model: Offers a balance between performance and computational resources for embedding generation.