Generic Triple-Latent Compression with Gated Associative Retrieval

Arxiv CS.CL (NLP/language models)

Proposes a triple-latent sequence model with gated associative retrieval to capture higher-order token interactions. Outperforms small Transformers without requiring benchmark-specific parsing.

transformer

Arxiv CS.CL (NLP/language models)·Liu Xiao

6 days ago