[V1][Spec Decode] KV cache slots for eagle heads (#16370)

Signed-off-by: LiuXiaoxuanPKU <lilyliupku@gmail.com>
2025-04-12 19:42:51 -07:00
parent 6c11ecf8d3
commit f49e5aff11
4 changed files with 98 additions and 18 deletions
--- a/vllm/v1/engine/core.py
+++ b/vllm/v1/engine/core.py
@@ -98,6 +98,7 @@ class EngineCore:
            cache_config=vllm_config.cache_config,
            lora_config=vllm_config.lora_config,
            kv_cache_config=kv_cache_config,
+            speculative_config=vllm_config.speculative_config,
            structured_output_manager=self.structured_output_manager,
            include_finished_set=vllm_config.parallel_config.data_parallel_size
            > 1,