集中式支持fa3 (#3112)

This commit is contained in:
yangjianfengo1
2025-08-01 18:03:36 +08:00
committed by GitHub
parent bdb83e007d
commit 64d7a3194d
4 changed files with 257 additions and 25 deletions

View File

@@ -294,6 +294,7 @@ elif paddle.is_compiled_with_cuda():
"gpu_ops/fused_rotary_position_encoding.cu",
"gpu_ops/noaux_tc.cu",
"gpu_ops/custom_all_reduce/all_reduce.cu",
"gpu_ops/merge_prefill_decode_output.cu",
]
# pd_disaggregation