{"passport":{"slug":"unified-kv-cache-management-for-vision-language-action-models","display_name":"Unified KV Cache Management for Vision-Language-Action Models","model_type":"multimodal","parameter_count":null,"context_window":null,"claimed_developer":"OxyGen","confirmed_developer":null,"developer_confirmed_at":null,"developer_source":null,"first_appeared_at":"2023-03-14T00:00:00+00:00","first_appeared_on":"rss_pipeline","status":"confirmed","availability":null,"deployment_status":"not_deployed","availability_scope":null,"gate":1,"eu_ai_act":"GPAI","regulatory_status":{"eu_ai_act":"GPAI"},"open_source":true,"weights_available":null,"license":null,"provenance_chain":[],"parent_model":null,"model_family":null,"superseded_by":null,"supersedes":[],"recommended_replacement":null,"product_links":null},"events":[{"event_type":"first_appearance","event_date":"2023-03-14T00:00:00+00:00","title":"First recorded in AI news","detail":"OxyGen proposes a unified KV cache management paradigm for Mixture-of-Transformers (MoT) Vision-Language-Action Models (VLAs) to enable efficient multi-task parallelism for on-device deployment. It achieves up to 3.7x speedup over isolated execution, delivering over 200 tokens/s language throughput and 70 Hz action frequency simultaneously without action quality degradation.","source_url":"https://arxiv.org/abs/2603.14371","source_name":"ArXiv AI"}],"delta":{"days_unattributed":1214,"claimed_developer":"OxyGen","confirmed_developer":null,"match":false}}