[Liger] add native liger-kernel orpo loss

tests/test_orpo_trainer.py

@@ -19,7 +19,7 @@
 from datasets import load_dataset
 from parameterized import parameterized
 from transformers import AutoModelForCausalLM, AutoModelForSeq2SeqLM, AutoTokenizer
-from transformers.testing_utils import require_peft
+from transformers.testing_utils import require_liger_kernel, require_peft
 
 from trl import ORPOConfig, ORPOTrainer
 from trl.trainer.utils import SIMPLE_CHAT_TEMPLATE
@@ -148,3 +148,39 @@ def test_orpo_trainer_with_lora(self, config_name):
                     # check the params have changed - ignore 0 biases
                     if param.sum() != 0:
                         self.assertFalse(torch.equal(param, new_param))
+
+    @require_liger_kernel
+    def test_orpo_trainer_with_liger(self):
+        """Test ORPO trainer with Liger loss enabled."""
+        with tempfile.TemporaryDirectory() as tmp_dir:
+            training_args = ORPOConfig(
+                output_dir=tmp_dir,
+                report_to="none",
+                use_liger_loss=True,  # Enable Liger loss
+            )
+
+            dummy_dataset = load_dataset("trl-internal-testing/zen", "standard_preference")
+
+            trainer = ORPOTrainer(
+                model=self.model,
+                args=training_args,
+                processing_class=self.tokenizer,
+                train_dataset=dummy_dataset["train"],
+                eval_dataset=dummy_dataset["test"],
+            )
+
+            # Verify Liger is being used
+            self.assertTrue(trainer._using_liger)
+
+            previous_trainable_params = {n: param.clone() for n, param in trainer.model.named_parameters()}
+
+            trainer.train()
+
+            self.assertIsNotNone(trainer.state.log_history[-1]["train_loss"])
+
+            # check the params have changed
+            for n, param in previous_trainable_params.items():
+                new_param = trainer.model.get_parameter(n)
+                # check the params have changed - ignore 0 biases
+                if param.sum() != 0:
+                    self.assertFalse(torch.equal(param, new_param))

trl/trainer/orpo_config.py

@@ -61,6 +61,8 @@ class ORPOConfig(TrainingArguments):
             string.
         dataset_num_proc (`Optional[int]`, *optional*, defaults to `None`):
             Number of processes to use for processing the dataset.
+        use_liger_loss (`bool`, *optional*, defaults to `False`):
+            Whether to use Liger loss.
     """
 
     learning_rate: float = 1e-6
@@ -76,3 +78,4 @@ class ORPOConfig(TrainingArguments):
     is_encoder_decoder: Optional[bool] = None
     model_init_kwargs: Optional[dict[str, Any]] = None
     dataset_num_proc: Optional[int] = None
+    use_liger_loss: bool = False

trl/trainer/orpo_trainer.py

@@ -47,7 +47,7 @@
 )
 from transformers.trainer_callback import TrainerCallback
 from transformers.trainer_utils import EvalLoopOutput
-from transformers.utils import is_peft_available, is_torch_fx_proxy
+from transformers.utils import is_liger_kernel_available, is_peft_available, is_torch_fx_proxy
 from transformers.utils.deprecation import deprecate_kwarg
 
 from ..data_utils import maybe_apply_chat_template, maybe_extract_prompt
@@ -68,7 +68,6 @@
 if is_peft_available():
     from peft import PeftModel, get_peft_model, prepare_model_for_kbit_training
 
-
 if is_wandb_available():
     import wandb
 
@@ -357,6 +356,26 @@ def make_inputs_require_grad(module, input, output):
                 "Your `Trainer` does not have an `accelerator` object. Consider upgrading `transformers`."
             )
 
+        # Import Liger loss if enabled
+        if self.args.use_liger_loss:
+            if not is_liger_kernel_available():
+                raise ValueError(
+                    "You set `use_liger_loss=True` but the liger kernel is not available. "
+                    "Please install liger-kernel first: `pip install liger-kernel`"
+                )
+            try:
+                from liger_kernel.chunked_loss import LigerFusedLinearORPOLoss
+
+                self.orpo_loss_fn = LigerFusedLinearORPOLoss(ignore_index=self.label_pad_token_id, beta=self.beta)
+                self._using_liger = True
+            except ImportError:
+                raise ImportError(
+                    "Failed to import LigerFusedLinearORPOLoss from liger-kernel. "
+                    "Please ensure you have the correct liger-kernel version installed."
+                )
+        else:
+            self._using_liger = False
+
     def _prepare_deepspeed(self, model: PreTrainedModelWrapper):
         # Adapted from accelerate: https://github.com/huggingface/accelerate/blob/739b135f8367becb67ffaada12fe76e3aa60fefd/src/accelerate/accelerator.py#L1473
         deepspeed_plugin = self.accelerator.state.deepspeed_plugin
@@ -756,51 +775,74 @@ def concatenated_forward(
             concatenated_batch["concatenated_input_ids"],
             attention_mask=concatenated_batch["concatenated_attention_mask"],
             use_cache=False,
+            output_hidden_states=True if self._using_liger else False,
             **model_kwargs,
         )
-        all_logits = outputs.logits
-
-        def cross_entropy_loss(logits, labels):
-            if not self.is_encoder_decoder:
-                # Shift so that tokens < n predict n
-                logits = logits[..., :-1, :].contiguous()
-                labels = labels[..., 1:].contiguous()
-            # Flatten the tokens
-            loss_fct = nn.CrossEntropyLoss()
-            logits = logits.view(-1, logits.shape[-1])
-            labels = labels.view(-1)
-            # Enable model parallelism
-            labels = labels.to(logits.device)
-            loss = loss_fct(logits, labels)
-            return loss
 
-        if self.is_encoder_decoder:
-            labels = concatenated_batch["concatenated_labels"].clone()
+        if self._using_liger:
+            lm_head = model.get_output_embeddings()
+
+            # Get the last hidden state from hidden_states tuple
+            last_hidden_state = outputs.hidden_states[-1]
+
+            # return the final loss and aux_outputs tuple
+            return self.orpo_loss_fn(
+                lm_head.weight,
+                last_hidden_state,
+                concatenated_batch["concatenated_labels"],
+                lm_head.bias if hasattr(lm_head, "bias") else None,
+            )
         else:
-            labels = concatenated_batch["concatenated_input_ids"].clone()
-            attention_mask = concatenated_batch["concatenated_attention_mask"]
-            labels = torch.where(attention_mask == 1, labels, self.label_pad_token_id)
+            all_logits = outputs.logits
+
+            def cross_entropy_loss(logits, labels):
+                if not self.is_encoder_decoder:
+                    # Shift so that tokens < n predict n
+                    logits = logits[..., :-1, :].contiguous()
+                    labels = labels[..., 1:].contiguous()
+                # Flatten the tokens
+                loss_fct = nn.CrossEntropyLoss()
+                logits = logits.view(-1, logits.shape[-1])
+                labels = labels.view(-1)
+                # Enable model parallelism
+                labels = labels.to(logits.device)
+                loss = loss_fct(logits, labels)
+                return loss
+
+            if self.is_encoder_decoder:
+                labels = concatenated_batch["concatenated_labels"].clone()
+            else:
+                labels = concatenated_batch["concatenated_input_ids"].clone()
+                attention_mask = concatenated_batch["concatenated_attention_mask"]
+                labels = torch.where(attention_mask == 1, labels, self.label_pad_token_id)
 
-        chosen_nll_loss = cross_entropy_loss(all_logits[:len_chosen], labels[:len_chosen])
+            chosen_nll_loss = cross_entropy_loss(all_logits[:len_chosen], labels[:len_chosen])
 
-        all_logps = self.get_batch_logps(
-            all_logits,
-            concatenated_batch["concatenated_labels"],
-            average_log_prob=True,
-            is_encoder_decoder=self.is_encoder_decoder,
-            label_pad_token_id=self.label_pad_token_id,
-        )
+            all_logps = self.get_batch_logps(
+                all_logits,
+                concatenated_batch["concatenated_labels"],
+                average_log_prob=True,
+                is_encoder_decoder=self.is_encoder_decoder,
+                label_pad_token_id=self.label_pad_token_id,
+            )
 
-        chosen_logps = all_logps[:len_chosen]
-        rejected_logps = all_logps[len_chosen:]
+            chosen_logps = all_logps[:len_chosen]
+            rejected_logps = all_logps[len_chosen:]
 
-        chosen_logits = all_logits[:len_chosen]
-        rejected_logits = all_logits[len_chosen:]
+            chosen_logits = all_logits[:len_chosen]
+            rejected_logits = all_logits[len_chosen:]
 
-        if self.aux_loss_enabled:
-            return (chosen_logps, rejected_logps, chosen_logits, rejected_logits, chosen_nll_loss, outputs.aux_loss)
+            if self.aux_loss_enabled:
+                return (
+                    chosen_logps,
+                    rejected_logps,
+                    chosen_logits,
+                    rejected_logits,
+                    chosen_nll_loss,
+                    outputs.aux_loss,
+                )
 
-        return (chosen_logps, rejected_logps, chosen_logits, rejected_logits, chosen_nll_loss)
+            return (chosen_logps, rejected_logps, chosen_logits, rejected_logits, chosen_nll_loss)
 
     def get_batch_loss_metrics(
         self,
@@ -812,21 +854,41 @@ def get_batch_loss_metrics(
         metrics = {}
 
         forward_output = self.concatenated_forward(model, batch)
-        (
-            policy_chosen_logps,
-            policy_rejected_logps,
-            policy_chosen_logits,
-            policy_rejected_logits,
-            policy_nll_loss,
-        ) = forward_output[:5]
-        if self.aux_loss_enabled:
-            aux_loss = forward_output[5]
+        if self._using_liger:
+            # full ORPO loss and aux outputs
+            (
+                loss,
+                (
+                    policy_chosen_logps,
+                    policy_rejected_logps,
+                    policy_chosen_logits,
+                    policy_rejected_logits,
+                    policy_nll_loss,
+                    chosen_rewards,
+                    rejected_rewards,
+                    log_odds_ratio,
+                    log_odds_chosen,
+                ),
+            ) = forward_output
+        else:
+            (
+                policy_chosen_logps,
+                policy_rejected_logps,
+                policy_chosen_logits,
+                policy_rejected_logits,
+                policy_nll_loss,
+            ) = forward_output[:5]
+            if self.aux_loss_enabled:
+                aux_loss = forward_output[5]
+
+            losses, chosen_rewards, rejected_rewards, log_odds_ratio, log_odds_chosen = self.odds_ratio_loss(
+                policy_chosen_logps, policy_rejected_logps
+            )
+            # full ORPO loss
+            loss = policy_nll_loss - losses.mean()
 
-        losses, chosen_rewards, rejected_rewards, log_odds_ratio, log_odds_chosen = self.odds_ratio_loss(
-            policy_chosen_logps, policy_rejected_logps
-        )
-        # full ORPO loss
-        loss = policy_nll_loss - losses.mean()
+            if self.aux_loss_enabled:
+                loss += self.aux_loss_coef * aux_loss
 
         reward_accuracies = (chosen_rewards > rejected_rewards).float()
 
@@ -846,8 +908,6 @@ def get_batch_loss_metrics(
             xm.mark_step()  # needed because .item() calls
         for k, v in metrics.items():
             metrics[k] = v.item()
-        if self.aux_loss_enabled:
-            loss += self.aux_loss_coef * aux_loss
 
         return loss, metrics
 
@@ -859,7 +919,6 @@ def compute_loss(
         num_items_in_batch=None,
     ) -> Union[torch.Tensor, tuple[torch.Tensor, dict[str, torch.Tensor]]]:
         compute_loss_context_manager = amp.autocast("cuda") if self._peft_has_been_casted_to_bf16 else nullcontext()
-
         with compute_loss_context_manager:
             loss, metrics = self.get_batch_loss_metrics(model, inputs, train_eval="train")