[MLX] Fix startup after logical token capacity change (#41314)
This commit is contained in:
@@ -107,6 +107,13 @@ class MlxModelRunnerStub(ModelRunner):
|
||||
# that path working instead of raising AttributeError.
|
||||
prefill_aware_swa = False
|
||||
|
||||
@property
|
||||
def logical_max_total_num_tokens(self) -> int:
|
||||
"""MLX pool slots already count logical tokens; no DCP expansion applies."""
|
||||
return self.req_to_token_pool.schedulable_token_capacity(
|
||||
self.max_total_num_tokens
|
||||
)
|
||||
|
||||
@property
|
||||
def preloaded_weights_bytes(self) -> int:
|
||||
"""Return zero for the base Torch loader accounting hook.
|
||||
|
||||
Reference in New Issue
Block a user