config

Author: Felipe Mello

apps/sft/llama3_8b.yaml

@@ -61,7 +61,6 @@ metric_logging:
     project: sft-training
     group: sft_exp_${oc.env:USER}
     logging_mode: global_reduce # global_reduce, per_rank_reduce, per_rank_no_reduce
-    per_rank_share_run: False
 
 # profiling:
 #   enable_profiling: false

apps/sft/qwen3_8b.yaml

@@ -55,6 +55,12 @@ activation_checkpoint:
   mode: selective
   selective_ac_option: op
 
+metric_logging:
+  wandb:
+    project: sft-training
+    group: sft_exp_${oc.env:USER}
+    logging_mode: global_reduce # global_reduce, per_rank_reduce, per_rank_no_reduce
+
 # profiling:
 #   enable_profiling: false
 

tests/unit_tests/datasets/test_iterable_utils.py

@@ -142,7 +142,6 @@ def generate_ckpt(
 
 
 def aggregate_metrics(metrics_list: list) -> dict[str, Any]:
-    """Aggregate metrics according to their reduction types (SUM, MEAN, MAX, MIN, STD)."""
     if not metrics_list:
         return {}