@@ -119,6 +119,7 @@ def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
119119 model_arch = vllm_config .model_config .architectures [0 ]
120120
121121 from atom .config import get_current_atom_config
122+
122123 if self ._is_draft :
123124 self .atom_config = get_current_atom_config ()
124125 else :
@@ -163,7 +164,6 @@ def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
163164 self .pp_group = get_pp_group ()
164165 self .tp_group = get_tp_group ()
165166
166-
167167 def _register_indexer_caches_with_vllm (self ):
168168 """Register DeepseekV32IndexerCache instances with vLLM so that:
169169 1. vLLM discovers them via isinstance(AttentionLayerBase) for KV cache
@@ -220,8 +220,7 @@ def _register_indexer_caches_with_vllm(self):
220220 if prefix not in vllm_sfc :
221221 vllm_sfc [prefix ] = module
222222 logger .info (
223- f"Registered indexer cache in vLLM static_forward_context: "
224- f"{ prefix } "
223+ f"Registered indexer cache in vLLM static_forward_context: { prefix } "
225224 )
226225 else :
227226 logger .warning (
@@ -295,7 +294,6 @@ class ATOMMoEForCausalLM(ATOMModelBase, VllmModelForTextGeneration): ...
295294class ATOMForConditionalGeneration (
296295 ATOMModelBase , VllmModelForTextGeneration , SupportsMultiModal , SupportsMRoPE
297296):
298-
299297 @classmethod
300298 def get_placeholder_str (cls , modality : str , i : int ) -> str | None :
301299 """
0 commit comments