[Bugfix] Fix correct error message when len(prompt) + max_tokens > max_model_len

vllm/entrypoints/renderer.py

@@ -395,12 +395,14 @@ def _create_tokens_prompt(
     ) -> TokensPrompt:
         """Create validated TokensPrompt."""
         if max_length is not None and len(token_ids) > max_length:
+            max_model_len = self.model_config.max_model_len
+            max_tokens = max_model_len - max_length
             raise VLLMValidationError(
-                f"This model's maximum context length is {max_length} tokens. "
-                f"However, your request has {len(token_ids)} input tokens. "
-                "Please reduce the length of the input messages.",
-                parameter="input_tokens",
-                value=len(token_ids),
+                f"This model's maximum context length is {max_model_len} tokens. "
+                f"However, your request has {len(token_ids)} input tokens plus "
+                f"{max_tokens} 'max_tokens'. Please reduce one or the other.",
+                parameter="input_tokens, max_tokens",
+                value=(len(token_ids), max_tokens),
             )
 
         tokens_prompt = TokensPrompt(prompt_token_ids=token_ids)