[Fix] Fix several bugs on DSA models (#22430)
This commit is contained in:
@@ -1491,8 +1491,10 @@ class ServerArgs:
|
||||
self.nsa_decode_backend = "tilelang"
|
||||
elif kv_cache_dtype == "fp8_e4m3":
|
||||
if major >= 10:
|
||||
self.nsa_prefill_backend = "trtllm"
|
||||
self.nsa_decode_backend = "trtllm"
|
||||
if not user_set_prefill:
|
||||
self.nsa_prefill_backend = "trtllm"
|
||||
if not user_set_decode:
|
||||
self.nsa_decode_backend = "trtllm"
|
||||
else:
|
||||
# flashmla_auto dispatches to flashmla_sparse/flashmla_kv based on hardware and heuristics
|
||||
if not user_set_prefill:
|
||||
|
||||
Reference in New Issue
Block a user