drop empty token from beginning if tokenizer has no bos_token (in the…

… case of qwen) (#1490)

src/axolotl/core/trainer_builder.py

@@ -23,6 +23,7 @@
 from torch.utils.data import BatchSampler, DataLoader, RandomSampler, SequentialSampler
 from transformers import (
     EarlyStoppingCallback,
+    PreTrainedModel,
     Trainer,
     TrainerCallback,
     TrainingArguments,
@@ -802,6 +803,15 @@ def push_to_hub(self, *args, **kwargs) -> str:
 
         return super().push_to_hub(*args, **kwargs)
 
+    def tokenize_row(
+        self, feature, model: Optional[Union[PreTrainedModel, torch.nn.Module]] = None
+    ) -> Dict:
+        res = super().tokenize_row(feature, model=model)
+        if self.tokenizer.bos_token_id is None and res["prompt_input_ids"][0] is None:
+            for key in res.keys():
+                res[key] = res[key][1:]
+        return res
+
 
 class TrainerBuilderBase(abc.ABC):
     """