working version

log/eval_results_t1v-n-7b23714e-w-0.jsonl

@@ -0,0 +1,5 @@
+{"model": "gs://nb-t5x/eval_norwegian_NCC_2_000_000/checkpoint_2005000", "eval_date": "08-04-2022 12:06:19", "split": "validation", "result": {"accuracy": 86.33333333333333}}
+{"model": "gs://nb-t5x/eval_norwegian_NCC_2_000_000/checkpoint_2005000", "eval_date": "08-04-2022 15:52:57", "task": "categorise", "feature_length": null, "split": "validation", "result": {"accuracy": 86.33333333333333, "f1_macro": 86.33090327169275}}
+{"model": "gs://nb-t5x/eval_norwegian_NCC_2_000_000/checkpoint_2005000", "eval_date": "08-04-2022 17:40:27", "task": "categorise", "feature_length": null, "split": "validation", "result": {"accuracy": 86.33333333333333, "f1_macro": 86.33090327169275}}
+{"model": "gs://nb-t5x/eval_norwegian_NCC_2_000_000/checkpoint_2005000", "eval_date": "08-04-2022 18:07:14", "task": "categorise", "feature_length": null, "split": "validation", "result": {"accuracy": 86.33333333333333, "f1_macro": 86.33090327169275}}
+{"model": "gs://nb-t5x/eval_norwegian_NCC_2_000_000/checkpoint_2005000", "eval_date": "08-04-2022 18:31:25", "task": "categorise", "feature_length": {"inputs": 512, "targets": 2}, "split": "validation", "result": {"accuracy": 84.83333333333334, "f1_macro": 84.82911919977771}}

my_metrics.py

@@ -0,0 +1,7 @@
+import sklearn.metrics
+import numpy as np
+
+def f1_macro(targets, predictions):
+      targets, predictions = np.asarray(targets).astype(int), np.asarray(predictions).astype(int)
+      return {"f1_macro": 100*sklearn.metrics.f1_score(targets, predictions, average='macro')}
+

train_base.sh

@@ -1,7 +1,6 @@
 PROJECT_DIR=${HOME}"/models/t5-parliament-categorisation"
 T5X_DIR="../../t5x"  # directory where the t5x is cloned.
-#Needs to be updated when moving to tpu-v4  it should then be in another zone
-MODEL_DIR="gs://nb-t5x/eval_norwegian_NCC_2_000_000"
+MODEL_DIR="gs://nb-t5x-us-central2/eval2_norwegian_NCC_2_000_000"
 export PYTHONPATH=${PROJECT_DIR}
 
 python3 ${T5X_DIR}/t5x/train.py \