[DSA] Hopper FP8 FlashMLA KV padding (#22372)
This commit is contained in:
@@ -1534,9 +1534,9 @@ class ServerArgs:
|
||||
if not user_set_decode:
|
||||
self.nsa_decode_backend = "trtllm"
|
||||
else:
|
||||
# flashmla_auto dispatches to flashmla_sparse/flashmla_kv based on hardware and heuristics
|
||||
# Hopper FP8 defaults to flashmla_kv for both prefill and decode.
|
||||
if not user_set_prefill:
|
||||
self.nsa_prefill_backend = "flashmla_auto"
|
||||
self.nsa_prefill_backend = "flashmla_kv"
|
||||
if not user_set_decode:
|
||||
self.nsa_decode_backend = "flashmla_kv"
|
||||
else:
|
||||
|
||||
Reference in New Issue
Block a user