log average metrics via wandb

gfun/vgfs/commons.py

@@ -38,6 +38,13 @@ def format_langkey_wandb(lang_dict):
     return log_dict
 
 
+def format_average_wandb(avg_dict):
+    log_dict = {}
+    for metric, value in avg_dict.items():
+        log_dict[f"average metric/{metric}"] = value
+    return log_dict
+
+
 def XdotM(X, M, sif):
     E = X.dot(M)
     if sif:
@@ -221,7 +228,7 @@ class Trainer:
                     {
                         "loss/val": eval_loss,
                         **format_langkey_wandb(lang_metrics),
-                        "average/metrics": avg_metrics,
+                        **format_average_wandb(avg_metrics),
                     },
                     commit=False,
                 )

main.py

@@ -1,6 +1,6 @@
 import os
 
-os.environ["CUDA_VISIBLE_DEVICES"] = "0"
+os.environ["CUDA_VISIBLE_DEVICES"] = "3"
 
 from argparse import ArgumentParser
 from time import time
@@ -11,7 +11,7 @@ from gfun.generalizedFunnelling import GeneralizedFunnelling
 
 """
 TODO:
-    - [!] LR scheduler
+    - [!] add support for mT5
     - [!] CLS dataset is loading only "books" domain data
     - [!] documents should be trimmed to the same length (?)
     - [!] overall gfun results logger