Update text_generation.py

text_generation.py

@@ -66,8 +66,8 @@ class TextGenerationPipeline(Pipeline):
         return {"english_tokens": english_tokens, "bio_tokens": bio_tokens}
 
     def _forward(self, model_inputs: dict, max_num_tokens_to_decode: int = 50) -> dict:
-        english_tokens = model_inputs["english_tokens"].clone().to(torch.bfloat16)
-        bio_tokens = model_inputs["bio_tokens"].clone().to(torch.bfloat16)
+        english_tokens = model_inputs["english_tokens"].clone()
+        bio_tokens = model_inputs["bio_tokens"].clone()
         projected_bio_embeddings = None
 
         actual_num_steps = 0