From a7983f0550ae38c82bad943aadf37d883e15e9b7 Mon Sep 17 00:00:00 2001
From: yyzxw <1020938856@qq.com>
Date: Thu, 18 Sep 2025 14:41:43 +0800
Subject: [PATCH 1/2] fix: speculative-config-parse-error (#24888)

Signed-off-by: zxw <1020938856@qq.com>
---
 vllm/engine/arg_utils.py | 17 +++++++++++------
 1 file changed, 11 insertions(+), 6 deletions(-)

diff --git a/vllm/engine/arg_utils.py b/vllm/engine/arg_utils.py
index 8c7a1b413cdb..e1699d4e14cf 100644
--- a/vllm/engine/arg_utils.py
+++ b/vllm/engine/arg_utils.py
@@ -1463,12 +1463,17 @@ def _is_v1_supported_oracle(self, model_config: ModelConfig) -> bool:
             return False
 
         # V1 supports N-gram, Medusa, and Eagle speculative decoding.
-        if (self.speculative_config is not None
-                and self.speculative_config.get("method") == "draft_model"):
-            raise NotImplementedError(
-                "Speculative decoding with draft model is not supported yet. "
-                "Please consider using other speculative decoding methods "
-                "such as ngram, medusa, eagle, or deepseek_mtp.")
+        if self.speculative_config is not None:
+            if isinstance(self.speculative_config, dict):
+                method = self.speculative_config.get("method")
+            else:
+                method = getattr(self.speculative_config, "method", None)
+
+            if method == "draft_model":
+                raise NotImplementedError(
+                    "Draft model speculative decoding is not supported yet. "
+                    "Please consider using other speculative decoding methods "
+                    "such as ngram, medusa, eagle, or deepseek_mtp.")
 
         V1_BACKENDS = [
             "FLASH_ATTN_VLLM_V1",

From 9b510f5f4bdebee8b60009726110d1823551023d Mon Sep 17 00:00:00 2001
From: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Date: Thu, 25 Sep 2025 09:57:34 +0100
Subject: [PATCH 2/2] Don't assume the dict will always have `method`

Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
---
 vllm/engine/arg_utils.py | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)

diff --git a/vllm/engine/arg_utils.py b/vllm/engine/arg_utils.py
index e1699d4e14cf..a9c3a80a9c82 100644
--- a/vllm/engine/arg_utils.py
+++ b/vllm/engine/arg_utils.py
@@ -1464,10 +1464,11 @@ def _is_v1_supported_oracle(self, model_config: ModelConfig) -> bool:
 
         # V1 supports N-gram, Medusa, and Eagle speculative decoding.
         if self.speculative_config is not None:
+            # speculative_config could still be a dict at this point
             if isinstance(self.speculative_config, dict):
-                method = self.speculative_config.get("method")
+                method = self.speculative_config.get("method", None)
             else:
-                method = getattr(self.speculative_config, "method", None)
+                method = self.speculative_config.method
 
             if method == "draft_model":
                 raise NotImplementedError(