WUSH-KV: KV Cache Quantization with Data-Adaptive Transforms Paper • 2609.38121 • Published 7 days ago • 24
Swift 1.5 27B Collection Swift 1.5 on Qwen3.8-27B: stronger than Swift 1.0 on agentic and coding tasks, with fewer thinking tokens. BF16 weights and every quant. • 11 items • Updated 11 days ago • 22