[assets] update wechat (#8385)

src/llamafactory/hparams/finetuning_args.py

@@ -204,7 +204,12 @@ class RLHFArguments:
     )
     ld_alpha: Optional[float] = field(
         default=None,
-        metadata={"help": "α parameter from the LD-DPO paper, which controls the weighting of the verbose token log-probabilities in responses"},
+        metadata={
+            "help": (
+                "Alpha parameter from the LD-DPO paper, which controls the weighting of"
+                " the verbose token log-probabilities in responses."
+            )
+        },
     )
 
 

src/llamafactory/train/dpo/trainer.py

@@ -188,8 +188,9 @@ class CustomDPOTrainer(DPOTrainer):
             batch = nested_detach(batch, clone=True)  # avoid error
 
         all_logits: torch.Tensor = model(**batch, return_dict=True, use_cache=False).logits.to(torch.float32)
-        all_logps, valid_length = get_batch_logps(logits=all_logits, labels=batch["labels"],
-                                                  ld_alpha=(self.ld_alpha if not is_ref_model else None))
+        all_logps, valid_length = get_batch_logps(
+            logits=all_logits, labels=batch["labels"], ld_alpha=(self.ld_alpha if not is_ref_model else None)
+        )
         if self.loss_type in ["ipo", "orpo", "simpo"]:
             all_logps = all_logps / valid_length
 
@@ -219,8 +220,9 @@ class CustomDPOTrainer(DPOTrainer):
             ref_context = nullcontext()
 
         with torch.no_grad(), ref_context:
-            reference_chosen_logps, reference_rejected_logps, *_ = self.concatenated_forward(ref_model, batch,
-                                                                                             is_ref_model=True)
+            reference_chosen_logps, reference_rejected_logps, *_ = self.concatenated_forward(
+                ref_model, batch, is_ref_model=True
+            )
 
         return reference_chosen_logps, reference_rejected_logps
 

src/llamafactory/train/trainer_utils.py

@@ -585,7 +585,10 @@ def create_custom_scheduler(
 
 
 def get_batch_logps(
-    logits: "torch.Tensor", labels: "torch.Tensor", label_pad_token_id: int = IGNORE_INDEX, ld_alpha: Optional[float] = None
+    logits: "torch.Tensor",
+    labels: "torch.Tensor",
+    label_pad_token_id: int = IGNORE_INDEX,
+    ld_alpha: Optional[float] = None,
 ) -> tuple["torch.Tensor", "torch.Tensor"]:
     r"""Compute the log probabilities of the given labels under the given logits.