Multimodal LLM Tokenization & Compression
7 repos
Tokenization and vector quantization techniques for multimodal large language models, particularly focused on unified representations that integrate text, images, and other modalities. The cluster centers on efficient token compression and quantization methods (notably through projects like UniToken and unitok_tokenizer) that enable more compact and effective processing of multimodal data in large-scale models, with applications spanning text-to-image generation and multimodal understanding tasks.
Python · 4
multimodal-understanding ·466
vector-quantization ·466
text-to-image-generation ·466
large-language-models ·466
vila_u_llama ·26
safetensors ·26
any-to-any ·26
feature-extraction ·11
text-generation ·0
transformers ·0
