update eval_strategy (huggingface#1662)

Author: kashif

tests/test_dpo_trainer.py

@@ -94,6 +94,7 @@ def _init_dummy_dataset(self):
             ["t5", "sppo_hard", True],
             ["gpt2", "nca_pair", False],
             ["t5", "nca_pair", True],
+            ["gpt2", "robust", True],
         ]
     )
     def test_dpo_trainer(self, name, loss_type, pre_compute):
@@ -317,7 +318,7 @@ def test_tr_dpo_trainer(self):
                 remove_unused_columns=False,
                 gradient_accumulation_steps=4,
                 learning_rate=9e-1,
-                evaluation_strategy="steps",
+                eval_strategy="steps",
                 precompute_ref_log_probs=False,
                 sync_ref_model=True,
                 ref_model_mixup_alpha=0.5,
@@ -508,6 +509,10 @@ def test_dpo_lora_bf16_autocast_llama(self):
             ["gpt2", "bco_pair", False, True],
             ["gpt2", "bco_pair", True, False],
             ["gpt2", "bco_pair", True, True],
+            ["gpt2", "robust", False, False],
+            ["gpt2", "robust", False, True],
+            ["gpt2", "robust", True, False],
+            ["gpt2", "robust", True, True],
         ]
     )
     @require_bitsandbytes

tests/test_kto_trainer.py

@@ -271,7 +271,7 @@ def test_kto_trainer_bco_udm(self):
                 remove_unused_columns=False,
                 gradient_accumulation_steps=4,
                 learning_rate=9e-1,
-                evaluation_strategy="steps",
+                eval_strategy="steps",
                 beta=0.1,
                 loss_type="bco",
             )

tests/test_sft_trainer.py

@@ -263,7 +263,7 @@ def test_sft_trainer_uncorrect_data(self):
             training_args = SFTConfig(
                 output_dir=tmp_dir,
                 dataloader_drop_last=True,
-                evaluation_strategy="steps",
+                eval_strategy="steps",
                 max_steps=2,
                 eval_steps=1,
                 save_steps=1,
@@ -281,7 +281,7 @@ def test_sft_trainer_uncorrect_data(self):
             training_args = SFTConfig(
                 output_dir=tmp_dir,
                 dataloader_drop_last=True,
-                evaluation_strategy="steps",
+                eval_strategy="steps",
                 max_steps=2,
                 eval_steps=1,
                 save_steps=1,
@@ -298,7 +298,7 @@ def test_sft_trainer_uncorrect_data(self):
             training_args = SFTConfig(
                 output_dir=tmp_dir,
                 dataloader_drop_last=True,
-                evaluation_strategy="steps",
+                eval_strategy="steps",
                 max_steps=2,
                 eval_steps=1,
                 save_steps=1,
@@ -315,7 +315,7 @@ def test_sft_trainer_uncorrect_data(self):
             training_args = SFTConfig(
                 output_dir=tmp_dir,
                 dataloader_drop_last=True,
-                evaluation_strategy="steps",
+                eval_strategy="steps",
                 max_steps=2,
                 eval_steps=1,
                 save_steps=1,
@@ -331,7 +331,7 @@ def test_sft_trainer_uncorrect_data(self):
             training_args = SFTConfig(
                 output_dir=tmp_dir,
                 dataloader_drop_last=True,
-                evaluation_strategy="steps",
+                eval_strategy="steps",
                 max_steps=2,
                 eval_steps=1,
                 save_steps=1,
@@ -352,7 +352,7 @@ def test_sft_trainer_uncorrect_data(self):
                 training_args = SFTConfig(
                     output_dir=tmp_dir,
                     dataloader_drop_last=True,
-                    evaluation_strategy="steps",
+                    eval_strategy="steps",
                     max_steps=2,
                     eval_steps=1,
                     save_steps=1,
@@ -372,7 +372,7 @@ def test_sft_trainer_uncorrect_data(self):
                 training_args = SFTConfig(
                     output_dir=tmp_dir,
                     dataloader_drop_last=True,
-                    evaluation_strategy="steps",
+                    eval_strategy="steps",
                     max_steps=2,
                     eval_steps=1,
                     save_steps=1,
@@ -390,7 +390,7 @@ def test_sft_trainer_uncorrect_data(self):
             training_args = SFTConfig(
                 output_dir=tmp_dir,
                 dataloader_drop_last=True,
-                evaluation_strategy="steps",
+                eval_strategy="steps",
                 max_steps=2,
                 eval_steps=1,
                 save_steps=1,
@@ -1089,7 +1089,7 @@ def test_sft_trainer_eval_packing(self):
             training_args = SFTConfig(
                 output_dir=tmp_dir,
                 dataloader_drop_last=True,
-                evaluation_strategy="steps",
+                eval_strategy="steps",
                 max_steps=4,
                 eval_steps=2,
                 save_steps=2,
@@ -1111,7 +1111,7 @@ def test_sft_trainer_eval_packing(self):
             training_args = SFTConfig(
                 output_dir=tmp_dir,
                 dataloader_drop_last=True,
-                evaluation_strategy="steps",
+                eval_strategy="steps",
                 max_steps=4,
                 eval_steps=2,
                 save_steps=2,