Merge 5ce77694018bd9b17d82318b8fe8b84f5453fe78 into 254f6b986720c92ddf97fbb1a6a6465da8e87e29

vllm/config/speculative.py

@@ -618,7 +618,7 @@ class SpeculativeConfig:
                 f"{self.disable_by_batch_size=}"
             )
 
-        eagle3_target_supported = ["llama", "qwen", "minicpm", "gpt_oss"]
+        eagle3_target_supported = ["llama", "qwen", "minicpm", "gpt_oss", "kimi_k2"]
         if (
             self.method == "eagle3"
             and self.target_model_config

vllm/model_executor/models/deepseek_v2.py

@@ -85,7 +85,13 @@ from vllm.v1.attention.backends.mla.indexer import (
 from vllm.v1.kv_cache_interface import KVCacheSpec, MLAAttentionSpec
 from vllm.v1.worker.workspace import current_workspace_manager
 
-from .interfaces import MixtureOfExperts, SupportsEagle, SupportsLoRA, SupportsPP
+from .interfaces import (
+    MixtureOfExperts,
+    SupportsEagle,
+    SupportsEagle3,
+    SupportsLoRA,
+    SupportsPP,
+)
 from .utils import (
     PPMissingLayer,
     is_pp_missing_parameter,
@@ -1381,7 +1387,12 @@ class DeepseekV2MixtureOfExperts(MixtureOfExperts):
 
 
 class DeepseekV2ForCausalLM(
-    nn.Module, SupportsPP, DeepseekV2MixtureOfExperts, SupportsLoRA, SupportsEagle
+    nn.Module,
+    SupportsPP,
+    DeepseekV2MixtureOfExperts,
+    SupportsLoRA,
+    SupportsEagle,
+    SupportsEagle3,
 ):
     packed_modules_mapping = {
         "gate_up_proj": ["gate_proj", "up_proj"],
@@ -1463,6 +1474,13 @@ class DeepseekV2ForCausalLM(
     def embed_input_ids(self, input_ids: torch.Tensor) -> torch.Tensor:
         return self.model.embed_input_ids(input_ids)
 
+    def set_aux_hidden_state_layers(self, layers: tuple[int, ...]) -> None:
+        self.model.aux_hidden_state_layers = layers
+
+    def get_eagle3_aux_hidden_state_layers(self) -> tuple[int, ...]:
+        num_layers = len(self.model.layers)
+        return (2, num_layers // 2, num_layers - 3)
+
     def forward(
         self,
         input_ids: torch.Tensor,