Add Sequence-Level KD

docs/source/gkd_trainer.md

@@ -19,8 +19,9 @@ This post-training method was contributed by [Kashif Rasul](https://huggingface.
 
 ## Usage tips
 
-The GKD Trainer is a wrapper around the [`SFTTrainer`] class that takes in a teacher model argument. It needs two parameters to be set via the [`GKDConfig`] namely:
+The [`GKDTrainer`] is a wrapper around the [`SFTTrainer`] class that takes in a teacher model argument. It needs three parameters to be set via the [`GKDConfig`] namely:
 * `lmbda`:  controls the student data fraction, i.e., the proportion of on-policy student-generated outputs. When `lmbda=0.0`, the loss reduces to supervised JSD where the student is trained with the token-level probabilities of the teacher. When `lmbda=1.0`, the loss reduces to on-policy JSD, where the student generates output sequences and token-specific feedback on these sequences from the teacher. For values in between [0, 1] it is random between the two based on the `lmbda` value for each batch.
+* `seq_kd`:  controls whether to perform Sequence-Level KD (can be viewed as supervised FT on teacher-generated out). When `seq_kd=True` and `lmbda=0.0`, the loss reduces to supervised JSD, where the teacher generates output sequences and the student receives token-specific feedback on these sequences from the teacher. 
 * `beta`: controls the interpolation in the generalized Jensen-Shannon Divergence.  When `beta=0.0` the loss approximates forward KL divergence, while for `beta=1.0` the loss approximates reverse KL divergence. For values in between [0, 1] it interpolates between the two.
 
 The authors find that on-policy data (high `lmbda`) performs better and the optimal `beta` varied depending on the task and evaluation method.

trl/trainer/gkd_config.py

@@ -41,6 +41,9 @@ class GKDConfig(SFTConfig):
             from a string.
         disable_dropout (`bool`, *optional*, defaults to `True`):
             Whether or not to disable dropouts in `model`.
+        seq_kd (`bool`, *optional*, defaults to `False`):
+            Seq_kd parameter that controls whether to perform Sequence-Level KD (can be viewed as supervised FT
+            on teacher-generated output).
     """
 
     temperature: float = 0.9
@@ -50,6 +53,7 @@ class GKDConfig(SFTConfig):
     teacher_model_name_or_path: Optional[str] = None
     teacher_model_init_kwargs: Optional[Dict[str, Any]] = None
     disable_dropout: bool = True
+    seq_kd: bool = False
 
     def __post_init__(self):
         super().__post_init__()

trl/trainer/gkd_trainer.py

@@ -136,6 +136,7 @@ def __init__(
         self.lmbda = args.lmbda
         self.beta = args.beta
         self.temperature = args.temperature
+        self.seq_kd = args.seq_kd
 
         self.generation_config = GenerationConfig(
             max_new_tokens=args.max_new_tokens,
@@ -280,6 +281,14 @@ def training_step(self, model: nn.Module, inputs: Dict[str, Union[torch.Tensor,
         With probability `self.lmbda`, it generates new responses using the student model,
         which are then used for training instead of the original inputs.
         """
+        if self.seq_kd:
+            with unwrap_model_for_generation(self.teacher_model, self.accelerator) as unwrapped_model:
+                new_input_ids, new_attention_mask, new_labels = self.generate_on_policy_outputs(
+                    unwrapped_model, inputs, self.generation_config, self.processing_class.pad_token_id
+                )
+            inputs["input_ids"] = new_input_ids
+            inputs["attention_mask"] = new_attention_mask
+            inputs["labels"] = new_labels
         if random.random() <= self.lmbda:
             with unwrap_model_for_generation(model, self.accelerator) as unwrapped_model:
                 new_input_ids, new_attention_mask, new_labels = self.generate_on_policy_outputs(