Resolve formatting issues

Signed-off-by: weisirui-eng <weisirui@h-partners.com>

Author: weisirui-eng

vllm_ascend/spec_decode/mtp_proposer.py

@@ -1,4 +1,3 @@
-
 import torch
 import torch.nn as nn
 from vllm.attention.layer import Attention
@@ -49,7 +48,6 @@ def __init__(
             dtype=self.runner.dtype,
             device=self.device)
 
-
         # We need +1 here because the arange is used to set query_start_loc,
         # which has one more element than batch_size.
         self.arange = torch.arange(vllm_config.scheduler_config.max_num_seqs +
@@ -102,8 +100,6 @@ def dummy_run(self,
         moe_comm_method = self.runner._select_moe_comm_method(
             num_tokens, with_prefill)
 
-
-
         if skip_attn:
             attn_metadata = None
         else:
@@ -289,7 +285,6 @@ def _propose(
         # Replace the last token with the next token.
         # E.g., [b1, b2, c1, c2, c3, c3] -> [a2, b2, b3, c2, c3, c4]
 
-
         self.input_ids[last_token_indices] = next_token_ids
 
         query_lens = cu_num_tokens[1:] - cu_num_tokens[:-1]
@@ -344,8 +339,6 @@ def _propose(
         for layer_name in self.attn_layer_name:
             attn_metadata[layer_name] = attn_metadata_mtp
 
-
-
         self.positions[:num_tokens] = target_positions
         self.hidden_states[:num_tokens] = target_hidden_states
 
@@ -379,7 +372,6 @@ def _propose(
                     model_kwargs = {}
                     model_kwargs["attn_metadata"] = attn_metadata
 
-
                     hidden_states = self.model(
                         input_ids=self.input_ids[:num_input_tokens],
                         positions=self.positions[:num_input_tokens],
@@ -418,10 +410,8 @@ def _propose(
             if step == self.num_speculative_tokens - 1 or with_prefill:
                 break
 
-
             attn_metadata_i = attn_metadata[self.attn_layer_name[0]]
 
-
             if step == 0:
                 positions = target_positions[last_token_indices]
                 hidden_states = hidden_states[last_token_indices]
@@ -432,7 +422,6 @@ def _propose(
                 if attn_metadata_i.num_decode_tokens != 0:
                     attn_metadata_i.num_decode_tokens = batch_size
 
-
             input_ids = draft_token_ids_list[-1].int()
             positions += 1
 
@@ -489,7 +478,6 @@ def _propose(
         draft_token_ids = torch.stack(draft_token_ids_list, dim=1)
         return draft_token_ids
 
-
     # TODO Using torch instead of triton may result in poor performance
     def _prepare_input_kernel(self, out_ptr: torch.Tensor,
                               cu_query_lens: torch.Tensor,

vllm_ascend/torchair/mtp_torchair_proposer.py

@@ -15,6 +15,7 @@
 from vllm.v1.core.sched.output import SchedulerOutput
 from vllm.v1.sample.metadata import SamplingMetadata
 from vllm.v1.spec_decode.metadata import SpecDecodeMetadata
+
 from vllm_ascend.ascend_config import get_ascend_config
 from vllm_ascend.ascend_forward_context import set_ascend_forward_context
 from vllm_ascend.attention.utils import AscendCommonAttentionMetadata
@@ -56,7 +57,6 @@ def load_model(self, model) -> None:
             self.model = TorchairDeepSeekMTP(
                 vllm_config=self.vllm_config).to(target_device)
 
-
         draft_attn_layer_names = (
             get_layers_from_vllm_config(self.vllm_config, Attention).keys() -
             target_attn_layer_names)
@@ -329,16 +329,12 @@ def _propose(
             num_computed_tokens_cpu=None,
             seq_lens=None)
 
-
-
-
         attn_metadata = self.runner.attn_metadata_builder.build(
             0, common_attn_metadata, self.runner.get_model())
 
         self.positions[:num_tokens] = target_positions
         self.hidden_states[:num_tokens] = target_hidden_states
 
-
         # torchair mode can reuse self.runner.num_tokens_across_dp
         num_tokens_across_dp = self.runner.num_tokens_across_dp
         with_prefill = self.runner.with_prefill
@@ -420,7 +416,6 @@ def _propose(
             if step == self.num_speculative_tokens - 1 or with_prefill:
                 break
 
-
             attn_metadata_i = attn_metadata
 
             if step == 0:
@@ -546,5 +541,3 @@ def _get_torchair_lazy_compiled_model(self, batch_size: int):
                     config=config,
                     ge_cache=False)
             return self.torchair_compiled_models[batch_size]
-
-