get_batch_sample -> generate_from_model[_and_ref]

trl/trainer/bco_trainer.py

@@ -1290,7 +1290,7 @@ def _get_train_sampler(self) -> Optional[torch.utils.data.Sampler]:
             return None
         return SequentialSampler(self.train_dataset)
 
-    def get_batch_samples(self, model, batch: Dict[str, torch.LongTensor]) -> Tuple[str, str]:
+    def generate_from_model_and_ref(self, model, batch: Dict[str, torch.LongTensor]) -> Tuple[str, str]:
         """Generate samples from the model and reference model for the given batch of inputs."""
 
         # If one uses `generate_during_eval` with peft + bf16, we need to explicitly call generate with
@@ -1407,7 +1407,7 @@ def evaluation_loop(
                 "prompt_attention_mask": itemgetter(*target_indicies)(random_batch["prompt_attention_mask"]),
                 "prompt": itemgetter(*target_indicies)(random_batch["prompt"]),
             }
-            policy_output_decoded, ref_output_decoded = self.get_batch_samples(self.model, target_batch)
+            policy_output_decoded, ref_output_decoded = self.generate_from_model_and_ref(self.model, target_batch)
 
             self.log(
                 {

trl/trainer/cpo_trainer.py

@@ -847,7 +847,7 @@ def compute_loss(
             return (loss, metrics)
         return loss
 
-    def get_batch_samples(self, model, batch: Dict[str, torch.LongTensor]) -> Tuple[str, str]:
+    def generate_from_model(self, model, batch: Dict[str, torch.LongTensor]) -> Tuple[str, str]:
         """Generate samples from the model and reference model for the given batch of inputs."""
 
         # If one uses `generate_during_eval` with peft + bf16, we need to explicitly call generate with
@@ -938,7 +938,7 @@ def evaluation_loop(
             random_batch = self.data_collator(random_batch_dataset)
             random_batch = self._prepare_inputs(random_batch)
 
-            policy_output_decoded = self.get_batch_samples(self.model, random_batch)
+            policy_output_decoded = self.generate_from_model(self.model, random_batch)
 
             self.log(
                 {

trl/trainer/dpo_trainer.py

@@ -1561,7 +1561,7 @@ def compute_loss(
             return (loss, metrics)
         return loss
 
-    def get_batch_samples(self, model, batch: Dict[str, torch.LongTensor]) -> Tuple[str, str]:
+    def generate_from_model_and_ref(self, model, batch: Dict[str, torch.LongTensor]) -> Tuple[str, str]:
         """Generate samples from the model and reference model for the given batch of inputs."""
 
         # If one uses `generate_during_eval` with peft + bf16, we need to explicitly call generate with
@@ -1672,7 +1672,7 @@ def evaluation_loop(
             random_batch = self.data_collator(random_batch_dataset)
             random_batch = self._prepare_inputs(random_batch)
 
-            policy_output_decoded, ref_output_decoded = self.get_batch_samples(self.model, random_batch)
+            policy_output_decoded, ref_output_decoded = self.generate_from_model_and_ref(self.model, random_batch)
 
             self.log(
                 {

trl/trainer/kto_trainer.py

@@ -1264,7 +1264,7 @@ def _get_train_sampler(self) -> Optional[torch.utils.data.Sampler]:
             return None
         return SequentialSampler(self.train_dataset)
 
-    def get_batch_samples(self, model, batch: Dict[str, torch.LongTensor]) -> Tuple[str, str]:
+    def generate_from_model_and_ref(self, model, batch: Dict[str, torch.LongTensor]) -> Tuple[str, str]:
         """Generate samples from the model and reference model for the given batch of inputs."""
 
         # If one uses `generate_during_eval` with peft + bf16, we need to explicitly call generate with
@@ -1383,7 +1383,7 @@ def evaluation_loop(
                 "prompt_attention_mask": random_batch["prompt_attention_mask"][target_indicies],
                 "prompt": itemgetter(*target_indicies)(random_batch["prompt"]),
             }
-            policy_output_decoded, ref_output_decoded = self.get_batch_samples(self.model, target_batch)
+            policy_output_decoded, ref_output_decoded = self.generate_from_model_and_ref(self.model, target_batch)
 
             self.log(
                 {

trl/trainer/orpo_trainer.py

@@ -866,7 +866,7 @@ def compute_loss(
             return (loss, metrics)
         return loss
 
-    def get_batch_samples(self, model, batch: Dict[str, torch.LongTensor]) -> Tuple[str, str]:
+    def generate_from_model(self, model, batch: Dict[str, torch.LongTensor]) -> Tuple[str, str]:
         """Generate samples from the model and reference model for the given batch of inputs."""
 
         # If one uses `generate_during_eval` with peft + bf16, we need to explicitly call generate with
@@ -957,7 +957,7 @@ def evaluation_loop(
             random_batch = self.data_collator(random_batch_dataset)
             random_batch = self._prepare_inputs(random_batch)
 
-            policy_output_decoded = self.get_batch_samples(self.model, random_batch)
+            policy_output_decoded = self.generate_from_model(self.model, random_batch)
 
             self.log(
                 {