[Feature] block sparse attention (#3668)

* 支持稀疏attn

* fix bug

* code style

* fix moba attn get kv shape

* 修复a100编译

* codestyle

* code style

* code style

* code style

* fix conflict

* 增加单侧

* code style

* 增加eblite 加载时间

* fix bug

* for ci

* for ci

* for ci

* for ci

* 支持mlp block size 128

* 增加小算子单测

* fix 单测 mlp

* 将环境变量加入到config里面

* fix rollout config

* 修复显存

* add test server

* add test server

* fix mlp  最后一层使用full attn

fastdeploy/platforms/cuda.py

@@ -64,6 +64,9 @@ class CUDAPlatform(Platform):
         elif selected_backend == _Backend.FLASH_ATTN:
             logger.info("Using FLASH ATTN backend.")
             return "fastdeploy.model_executor.layers.attention.FlashAttentionBackend"
+        elif selected_backend == _Backend.MOBA_ATTN:
+            logger.info("Using MOBA ATTN backend.")
+            return "fastdeploy.model_executor.layers.attention.MobaAttentionBackend"
         else:
             raise ValueError(
                 "Invalid attention backend you specified.\n"