Tiny cleanup some eagle unused codes (#11660)

python/sglang/srt/speculative/draft_utils.py

@@ -96,7 +96,6 @@ class DraftBackendFactory:
                 FlashInferMultiStepDraftBackend,
             )
 
-            self.has_prefill_wrapper_verify = True
             return FlashInferMultiStepDraftBackend(
                 self.draft_model_runner, self.topk, self.speculative_num_steps
             )
@@ -105,7 +104,6 @@ class DraftBackendFactory:
                 FlashInferMLAMultiStepDraftBackend,
             )
 
-            self.has_prefill_wrapper_verify = True
             return FlashInferMLAMultiStepDraftBackend(
                 self.draft_model_runner, self.topk, self.speculative_num_steps
             )
@@ -149,7 +147,6 @@ class DraftBackendFactory:
             TRTLLMHAAttnMultiStepDraftBackend,
         )
 
-        self.has_prefill_wrapper_verify = True
         return TRTLLMHAAttnMultiStepDraftBackend(
             self.draft_model_runner, self.topk, self.speculative_num_steps
         )
@@ -164,7 +161,6 @@ class DraftBackendFactory:
             TRTLLMMLAMultiStepDraftBackend,
         )
 
-        self.has_prefill_wrapper_verify = True
         return TRTLLMMLAMultiStepDraftBackend(
             self.draft_model_runner, self.topk, self.speculative_num_steps
         )

python/sglang/srt/speculative/eagle_draft_cuda_graph_runner.py

@@ -41,7 +41,6 @@ class EAGLEDraftCudaGraphRunner:
         # Parse args
         self.eagle_worker = eagle_worker
         self.model_runner = model_runner = eagle_worker.model_runner
-        self.model_runner: EAGLEWorker
         self.graphs = {}
         self.output_buffers = {}
         self.enable_torch_compile = model_runner.server_args.enable_torch_compile

python/sglang/srt/speculative/eagle_worker.py

@@ -192,10 +192,6 @@ class EAGLEWorker(TpModelWorker):
 
     def init_attention_backend(self):
         # Create multi-step attn backends and cuda graph runners
-
-        self.has_prefill_wrapper_verify = False
-        self.draft_extend_attn_backend = None
-
         draft_backend_factory = DraftBackendFactory(
             self.server_args,
             self.draft_model_runner,