fix: keep metric logs cumulative instead of segmental in each checkpoint
This commit is contained in:
@@ -310,7 +310,6 @@ class MetricCallback(TrainCallback):
|
||||
with open(log_file, "w") as f:
|
||||
for log in self.log_cache:
|
||||
f.write(json.dumps(log) + "\n")
|
||||
self.log_cache.clear()
|
||||
|
||||
def on_optimizer_step(self, context):
|
||||
if (
|
||||
|
||||
Reference in New Issue
Block a user