Fix missing PEFT availability check when passing peft_config to experimental trainers

trl/experimental/bco/bco_trainer.py

@@ -469,13 +469,14 @@ def __init__(
         if isinstance(ref_model, str):
             ref_model = AutoModelForCausalLM.from_pretrained(ref_model, **model_init_kwargs)
 
+        # PEFT
         # Initialize this variable to False. This helps tracking the case when `peft_module_casting_to_bf16`
         # has been called in order to properly call autocast if needed.
         self._peft_has_been_casted_to_bf16 = False
-
-        if not is_peft_available() and peft_config is not None:
-            raise ValueError(
-                "PEFT is not installed and you passed a `peft_config` in the trainer's kwargs, please install it with `pip install peft` to use the PEFT models"
+        if peft_config is not None and not is_peft_available():
+            raise ImportError(
+                "You passed `peft_config` but the `peft` library is not installed. "
+                "Install it with `pip install trl[peft]`."
             )
         elif is_peft_available() and peft_config is not None:
             if isinstance(model, PeftModel):

trl/experimental/cpo/cpo_trainer.py

@@ -169,13 +169,14 @@ def __init__(
         if isinstance(model, str):
             model = AutoModelForCausalLM.from_pretrained(model, **model_init_kwargs)
 
+        # PEFT
         # Initialize this variable to False. This helps tracking the case when `peft_module_casting_to_bf16`
         # has been called in order to properly call autocast if needed.
         self._peft_has_been_casted_to_bf16 = False
-
-        if not is_peft_available() and peft_config is not None:
-            raise ValueError(
-                "PEFT is not installed and you passed a `peft_config` in the trainer's kwargs, please install it to use the PEFT models"
+        if peft_config is not None and not is_peft_available():
+            raise ImportError(
+                "You passed `peft_config` but the `peft` library is not installed. "
+                "Install it with `pip install trl[peft]`."
             )
         elif is_peft_available() and peft_config is not None:
             if isinstance(model, PeftModel):

trl/experimental/distillation/distillation_trainer.py

@@ -429,6 +429,11 @@ def __init__(
                 processing_class.pad_token = processing_class.eos_token
 
         # ── PEFT ──
+        if peft_config is not None and not is_peft_available():
+            raise ImportError(
+                "You passed `peft_config` but the `peft` library is not installed. "
+                "Install it with `pip install trl[peft]`."
+            )
         if peft_config is not None:
             model = get_peft_model(model, peft_config)
 

trl/experimental/kto/kto_trainer.py

@@ -289,13 +289,14 @@ def __init__(
         if tokenizer.pad_token is None:
             tokenizer.pad_token = tokenizer.eos_token
 
+        # PEFT
         # Initialize this variable to False. This helps tracking the case when `peft_module_casting_to_bf16`
         # has been called in order to properly call autocast if needed.
         self._peft_has_been_casted_to_bf16 = False
-
-        if not is_peft_available() and peft_config is not None:
-            raise ValueError(
-                "PEFT is not installed and you passed a `peft_config` in the trainer's kwargs, please install it with `pip install peft` to use the PEFT models"
+        if peft_config is not None and not is_peft_available():
+            raise ImportError(
+                "You passed `peft_config` but the `peft` library is not installed. "
+                "Install it with `pip install trl[peft]`."
             )
         if is_peft_available() and isinstance(model, PeftModel) and peft_config is not None:
             raise ValueError(

trl/experimental/orpo/orpo_trainer.py

@@ -178,13 +178,14 @@ def __init__(
         if isinstance(model, str):
             model = AutoModelForCausalLM.from_pretrained(model, **model_init_kwargs)
 
+        # PEFT
         # Initialize this variable to False. This helps tracking the case when `peft_module_casting_to_bf16`
         # has been called in order to properly call autocast if needed.
         self._peft_has_been_casted_to_bf16 = False
-
-        if not is_peft_available() and peft_config is not None:
-            raise ValueError(
-                "PEFT is not installed and you passed a `peft_config` in the trainer's kwargs, please install it to use the PEFT models"
+        if peft_config is not None and not is_peft_available():
+            raise ImportError(
+                "You passed `peft_config` but the `peft` library is not installed. "
+                "Install it with `pip install trl[peft]`."
             )
         elif is_peft_available() and peft_config is not None:
             if isinstance(model, PeftModel):

trl/experimental/ppo/ppo_trainer.py

@@ -416,10 +416,11 @@ def __init__(
                 "[Approximating KL Divergence](http://joschu.net/blog/kl-approx.html) for details."
             )
 
-        # peft support
-        if not is_peft_available() and peft_config is not None:
+        # PEFT
+        if peft_config is not None and not is_peft_available():
             raise ImportError(
-                "PEFT is not installed and you passed a `peft_config` in the trainer's kwargs, please install it to use the PEFT models"
+                "You passed `peft_config` but the `peft` library is not installed. "
+                "Install it with `pip install trl[peft]`."
             )
         elif is_peft_available() and peft_config is not None:
             if isinstance(self.policy_model, PeftModel):

trl/experimental/tpo/tpo_trainer.py

@@ -350,6 +350,12 @@ def __init__(
         if tokenizer.pad_token is None:
             tokenizer.pad_token = tokenizer.eos_token
 
+        # PEFT
+        if peft_config is not None and not is_peft_available():
+            raise ImportError(
+                "You passed `peft_config` but the `peft` library is not installed. "
+                "Install it with `pip install trl[peft]`."
+            )
         if is_peft_available() and is_peft_model(model) and peft_config is not None:
             raise ValueError(
                 "You passed a `PeftModel` instance together with a `peft_config` to the trainer. Please first merge "