rename log_intevals to train_log_freq

Author: grassesi

config/default_config.yml

@@ -151,7 +151,7 @@ data_loader_rng_seed: ???
 run_id: ???
 
 # The period to log in the training loop (in number of batch steps)
-log_intervals:
+train_log_freq:
   terminal: 10
   metrics: 20
   checkpoint: 250

src/weathergen/run_train.py

@@ -70,8 +70,13 @@ def inference_from_args(argl: list[str]):
 
     cf.run_history += [(args.from_run_id, cf.istep)]
 
+<<<<<<< HEAD
     trainer = Trainer(cf.log_intervals)
     trainer.inference(cf, devices, args.from_run_id, args.epoch)
+=======
+    trainer = Trainer(cf.train_log_freq)
+    trainer.inference(cf, args.from_run_id, args.epoch)
+>>>>>>> acf76bd (rename `log_intevals` to `train_log_freq`)
 
 
 ####################################################################################################
@@ -138,7 +143,7 @@ def train_continue_from_args(argl: list[str]):
     # track history of run to ensure traceability of results
     cf.run_history += [(args.from_run_id, cf.istep)]
 
-    trainer = Trainer(cf.log_intervals)
+    trainer = Trainer(cf.train_log_freq)
     trainer.run(cf, devices, args.from_run_id, args.epoch)
 
 
@@ -184,7 +189,7 @@ def train_with_args(argl: list[str], stream_dir: str | None):
     if cf.with_flash_attention:
         assert cf.with_mixed_precision
 
-    trainer = Trainer(cf.log_intervals)
+    trainer = Trainer(cf.train_log_freq)
 
     try:
         trainer.run(cf, devices)

src/weathergen/train/trainer.py

@@ -54,10 +54,10 @@
 
 
 class Trainer(TrainerBase):
-    def __init__(self, log_intervals: Config):
+    def __init__(self, train_log_freq: Config):
         TrainerBase.__init__(self)
 
-        self.log_intervals = log_intervals
+        self.train_log_freq = train_log_freq
 
     def init(self, cf: Config, devices):
         self.cf = OmegaConf.merge(
@@ -589,11 +589,11 @@ def train(self, epoch):
             self.perf_mem = ddp_average(torch.tensor([perf_mem], device=self.device)).item()
 
             self._log_terminal(bidx, epoch, TRAIN)
-            if bidx % self.log_intervals.metrics == 0:
+            if bidx % self.train_log_freq.metrics == 0:
                 self._log(TRAIN)
 
             # save model checkpoint (with designation _latest)
-            if bidx % self.log_intervals.checkpoint == 0 and bidx > 0:
+            if bidx % self.train_log_freq.checkpoint == 0 and bidx > 0:
                 self.save_model(-1)
 
             self.cf.istep += cf.batch_size_per_gpu
@@ -943,7 +943,7 @@ def _log(self, stage: Stage):
         self.loss_unweighted_hist, self.loss_model_hist, self.stdev_unweighted_hist = [], [], []
 
     def _log_terminal(self, bidx: int, epoch: int, stage: Stage):
-        print_freq = self.log_intervals.terminal
+        print_freq = self.train_log_freq.terminal
         if bidx % print_freq == 0 and bidx > 0 or stage == VAL:
             # compute from last iteration
             avg_loss, losses_all, _ = self._prepare_losses_for_logging()