●builderThis framework presents a more efficient alternative to autoregressive LLM-based compression for high-throughput needs.
●researcherYou can leverage the trade-off between decoding throughput and compression rate using hypernetwork-driven dLLMs.