Upgrade to vllm 0.17.0 corex v4.1 overlay

vllm/tokenizers/registry.py

@@ -36,6 +36,7 @@ _VLLM_TOKENIZERS = {
     "grok2": ("grok2", "Grok2Tokenizer"),
     "hf": ("hf", "CachedHfTokenizer"),
     "mistral": ("mistral", "MistralTokenizer"),
+    "qwen_vl": ("qwen_vl", "QwenVLTokenizer"),
 }
 
 
@@ -165,6 +166,10 @@ def resolve_tokenizer_args(
     ):
         tokenizer_mode = "grok2"
 
+    # Model-specific tokenizers
+    if tokenizer_mode == "auto" and "/Qwen-VL" in str(tokenizer_name):
+        tokenizer_mode = "qwen_vl"
+
     # Fallback to HF tokenizer
     if tokenizer_mode == "auto":
         tokenizer_mode = "hf"