diff --git a/configs/pwc_hotpot_qa.yaml b/configs/pwc_hotpot_qa.yaml
index 26da9d3..b261aea 100644
--- a/configs/pwc_hotpot_qa.yaml
+++ b/configs/pwc_hotpot_qa.yaml
@@ -45,7 +45,3 @@ train_ds_names:
 val_ds_names:
   - pwc
   - hotpot_qa
-
-test_ds_names:
-  - pwc
-  - hotpot_qa
diff --git a/configs/qa.yaml b/configs/qa.yaml
index a146ade..c699d9b 100644
--- a/configs/qa.yaml
+++ b/configs/qa.yaml
@@ -50,7 +50,7 @@ train_ds_names:
 
 val_ds_names:
   - fw_qa_3_pretrain
-  - self_gen/google/gemma-2-2b-it/fw_qa_3
+  - self_gen/google/gemma-2-2b-it/fw_qa_3_small
   - self_gen/google/gemma-2-2b-it/ctx_qa
   - self_gen/google/gemma-2-2b-it/pwc
   - self_gen/google/gemma-2-2b-it/hotpot_qa
@@ -62,4 +62,4 @@ val_ds_names:
   - squad
 
 load_best_model_at_end: false
-metric_for_best_model: eval_pwc_loss
+metric_for_best_model: eval_fw_qa_3_pretrain_loss
diff --git a/configs/self_gen_3_and_pretrain_small.yaml b/configs/self_gen_3_and_pretrain_small.yaml
index 96fbfcb..7184ea1 100644
--- a/configs/self_gen_3_and_pretrain_small.yaml
+++ b/configs/self_gen_3_and_pretrain_small.yaml
@@ -51,7 +51,7 @@ train_ds_names:
 
 val_ds_names:
   - fw_qa_3_pretrain
-  - self_gen/google/gemma-2-2b-it/fw_qa_3
+  - self_gen/google/gemma-2-2b-it/fw_qa_3_small
   - self_gen/google/gemma-2-2b-it/ctx_qa
   - self_gen/google/gemma-2-2b-it/pwc
   - self_gen/google/gemma-2-2b-it/hotpot_qa
@@ -63,4 +63,4 @@ val_ds_names:
   - squad
 
 load_best_model_at_end: false
-metric_for_best_model: eval_pwc_loss
+metric_for_best_model: eval_fw_qa_3_pretrain_loss 
diff --git a/configs/self_gen_3_small.yaml b/configs/self_gen_3_small.yaml
index 862e0ab..a8580a3 100644
--- a/configs/self_gen_3_small.yaml
+++ b/configs/self_gen_3_small.yaml
@@ -50,7 +50,7 @@ train_ds_names:
 
 val_ds_names:
   - fw_qa_3_pretrain
-  - self_gen/google/gemma-2-2b-it/fw_qa_3
+  - self_gen/google/gemma-2-2b-it/fw_qa_3_small
   - self_gen/google/gemma-2-2b-it/ctx_qa
   - self_gen/google/gemma-2-2b-it/pwc
   - self_gen/google/gemma-2-2b-it/hotpot_qa
@@ -62,4 +62,4 @@ val_ds_names:
   - squad
 
 load_best_model_at_end: false
-metric_for_best_model: eval_pwc_loss
+metric_for_best_model: eval_fw_qa_3_pretrain_loss