Formatting 2

vllm/worker/hpu_enc_dec_model_runner.py

@@ -27,7 +27,6 @@
     _add_sampling_metadata_broadcastable_dict)
 from vllm.distributed import broadcast_tensor_dict
 
-
 if TYPE_CHECKING:
     from vllm.attention.backends.abstract import AttentionBackend
 
@@ -540,7 +539,6 @@ def _check_config(self, batch_size, seq_len, is_prompt, warmup_mode):
             logger.warning("Configuration: (%s, %s, %s) was not warmed-up!",
                            phase, batch_size, seq_len)
 
-
     def add_dummy_seq(self, seq_group_metadata_list, is_prompt):
         real_batch_size = len(seq_group_metadata_list)
         batch_size_padded = self.bucketing_ctx.get_padded_batch_size(
@@ -554,7 +552,6 @@ def add_dummy_seq(self, seq_group_metadata_list, is_prompt):
                                            for _ in range(batch_size_padding))
         return seq_group_metadata_list
 
-
     @torch.inference_mode()
     def execute_model(
         self,
@@ -593,7 +590,6 @@ def execute_model(
             use_graphs = self._use_graphs(batch_size, seq_len, is_prompt)
             self._check_config(batch_size, seq_len, is_prompt, warmup_mode)
 
-
             execute_model_kwargs = {
                 "input_ids": input_tokens,
                 "positions": input_positions,
@@ -776,7 +772,6 @@ def try_revert_dummy_output_tokens():
 
         return output if type(output) is list else [output]
 
-
     def _decode_sampler_outputs(self, model_input):
         use_async_out_proc = model_input.async_callback is not None
         sampler_outputs = []
@@ -806,7 +801,6 @@ def _decode_sampler_outputs(self, model_input):
         else:
             return sampler_outputs
 
-
     def _make_decode_output(
         self,
         next_token_ids: List[List[int]],
@@ -827,4 +821,4 @@ def _make_decode_output(
             sampler_outputs.append(
                 CompletionSequenceGroupOutput(seq_outputs, None))
         return SamplerOutput(sampler_outputs)
-
+