return error for cache_salt with V0 engine

Signed-off-by: Marko Rosenmueller <[email protected]>

Author: dr75

docs/source/design/v1/prefix_caching.md

@@ -92,6 +92,8 @@ To improve privacy in shared environments, vLLM supports isolating prefix cache
 
 With this setup, cache sharing is limited to users or requests that explicitly agree on a common salt, enabling cache reuse within a trust group while isolating others.
 
+> **Note:** Cache isolation is not supported in engine V0.
+
 ## Data Structure
 
 The prefix caching in vLLM v1 is implemented in the KV cache manager. The basic building block is the “Block” data class (simplified):

tests/entrypoints/openai/test_serving_chat.py

@@ -273,6 +273,26 @@ def test_serving_chat_could_load_correct_generation_config():
     assert mock_engine.generate.call_args.args[1].temperature == 0.0
     assert mock_engine.generate.call_args.args[1].repetition_penalty == 1.05
 
+
+def test_serving_chat_did_set_correct_cache_salt():
+    mock_model_config = MockModelConfig()
+
+    mock_engine = MagicMock(spec=MQLLMEngineClient)
+    mock_engine.get_tokenizer.return_value = get_tokenizer(MODEL_NAME)
+    mock_engine.errored = False
+
+    # Initialize the serving chat
+    models = OpenAIServingModels(engine_client=mock_engine,
+                                 base_model_paths=BASE_MODEL_PATHS,
+                                 model_config=mock_model_config)
+    serving_chat = OpenAIServingChat(mock_engine,
+                                     mock_model_config,
+                                     models,
+                                     response_role="assistant",
+                                     chat_template=CHAT_TEMPLATE,
+                                     chat_template_content_format="auto",
+                                     request_logger=None)
+
     # Test cache_salt
     req = ChatCompletionRequest(
         model=MODEL_NAME,

vllm/entrypoints/openai/protocol.py

@@ -13,6 +13,7 @@
                       ValidationInfo, field_validator, model_validator)
 from typing_extensions import TypeAlias
 
+from vllm import envs
 from vllm.entrypoints.chat_utils import ChatCompletionMessageParam
 from vllm.logger import init_logger
 from vllm.pooling_params import PoolingParams
@@ -392,7 +393,7 @@ class ChatCompletionRequest(OpenAIBaseModel):
             "environments. The salt should be random, protected from "
             "access by 3rd parties, and long enough to be "
             "unpredictable (e.g., 43 characters base64-encoded, corresponding "
-            "to 256 bit)."))
+            "to 256 bit). Not supported by vLLM engine V0."))
 
     # doc: end-chat-completion-extra-params
 
@@ -704,6 +705,20 @@ def check_generation_prompt(cls, data):
                              "`add_generation_prompt` to True.")
         return data
 
+    @model_validator(mode="before")
+    @classmethod
+    def check_cache_salt_support(cls, data):
+        if data.get("cache_salt") is not None:
+            if not envs.VLLM_USE_V1:
+                raise ValueError(
+                    "Parameter 'cache_salt' is not supported with "
+                    "this instance of vLLM, which uses engine V0.")
+            if not isinstance(data["cache_salt"], str) or len(
+                    data["cache_salt"]) == 0:
+                raise ValueError("Parameter 'cache_salt' must be a "
+                                 "non-empty string if provided.")
+        return data
+
 
 class CompletionRequest(OpenAIBaseModel):
     # Ordered by official OpenAI API documentation
@@ -1599,9 +1614,9 @@ class TranscriptionRequest(OpenAIBaseModel):
 
     # doc: begin-transcription-extra-params
     stream: Optional[bool] = False
-    """Custom field not present in the original OpenAI definition. When set, 
+    """Custom field not present in the original OpenAI definition. When set,
     it will enable output to be streamed in a similar fashion as the Chat
-    Completion endpoint. 
+    Completion endpoint.
     """
     # Flattened stream option to simplify form data.
     stream_include_usage: Optional[bool] = False
@@ -1619,15 +1634,15 @@ class TranscriptionRequest(OpenAIBaseModel):
     """
 
     top_p: Optional[float] = None
-    """Enables nucleus (top-p) sampling, where tokens are selected from the 
+    """Enables nucleus (top-p) sampling, where tokens are selected from the
     smallest possible set whose cumulative probability exceeds `p`.
     """
 
     top_k: Optional[int] = None
     """Limits sampling to the `k` most probable tokens at each step."""
 
     min_p: Optional[float] = None
-    """Filters out tokens with a probability lower than `min_p`, ensuring a 
+    """Filters out tokens with a probability lower than `min_p`, ensuring a
     minimum likelihood threshold during sampling.
     """