main
[Inference/Kernel] Optimize paged attention: Refactor key cache layout (#5643)
[Inference] Delete duplicated copy_vector (#5716)
[Inference/Refactor] Refactor compilation mechanism and unified multi hw (#5613)
[Inference/Feat] Add kvcache quant support for fused_rotary_embedding_cache_copy (#5680)
[Inference/Feat] Add convert_fp8 op for fp8 test in the future (#5706)
add paged-attetionv2: support seq length split across thread block (#5707)
[Inference] Remove unnecessary float4_ and rename float8_ to float8 (#5679)