fix(api): bypass model cache entirely when limit is 0
This commit is contained in:
parent
2b29b099f0
commit
df5fb93bc0
|
@ -5,14 +5,14 @@ formatters:
|
||||||
handlers:
|
handlers:
|
||||||
console:
|
console:
|
||||||
class: logging.StreamHandler
|
class: logging.StreamHandler
|
||||||
level: INFO
|
level: DEBUG
|
||||||
formatter: simple
|
formatter: simple
|
||||||
stream: ext://sys.stdout
|
stream: ext://sys.stdout
|
||||||
loggers:
|
loggers:
|
||||||
'':
|
'':
|
||||||
level: INFO
|
level: DEBUG
|
||||||
handlers: [console]
|
handlers: [console]
|
||||||
propagate: True
|
propagate: True
|
||||||
root:
|
root:
|
||||||
level: INFO
|
level: DEBUG
|
||||||
handlers: [console]
|
handlers: [console]
|
||||||
|
|
|
@ -96,7 +96,7 @@ def load_pipeline(
|
||||||
lpw: bool,
|
lpw: bool,
|
||||||
):
|
):
|
||||||
pipe_key = (pipeline, model, device.device, device.provider, lpw)
|
pipe_key = (pipeline, model, device.device, device.provider, lpw)
|
||||||
scheduler_key = (scheduler_type,)
|
scheduler_key = (scheduler_type, model)
|
||||||
|
|
||||||
cache_pipe = server.cache.get("diffusion", pipe_key)
|
cache_pipe = server.cache.get("diffusion", pipe_key)
|
||||||
|
|
||||||
|
|
|
@ -25,6 +25,10 @@ class ModelCache:
|
||||||
return None
|
return None
|
||||||
|
|
||||||
def set(self, tag: str, key: Any, value: Any) -> None:
|
def set(self, tag: str, key: Any, value: Any) -> None:
|
||||||
|
if self.limit == 0:
|
||||||
|
logger.debug("Cache limit set to 0, not caching model: %s", tag)
|
||||||
|
return
|
||||||
|
|
||||||
for i in range(len(self.cache)):
|
for i in range(len(self.cache)):
|
||||||
t, k, v = self.cache[i]
|
t, k, v = self.cache[i]
|
||||||
if tag == t:
|
if tag == t:
|
||||||
|
|
Loading…
Reference in New Issue