[Core] [Frontend] Priority scheduling for embeddings and in the OpenAI-API (#8965)

This commit is contained in:
Sebastian Schoennenbeck
2024-10-01 09:58:06 +00:00
committed by GitHub
parent 1fe0a4264a
commit 35bd215168
8 changed files with 53 additions and 5 deletions
+16 -4
View File
@@ -380,6 +380,7 @@ class MQLLMEngineClient:
lora_request: Optional[LoRARequest] = None,
trace_headers: Optional[Mapping[str, str]] = None,
prompt_adapter_request: Optional[PromptAdapterRequest] = None,
priority: int = 0,
) -> AsyncGenerator[RequestOutput, None]:
...
@@ -392,6 +393,7 @@ class MQLLMEngineClient:
lora_request: Optional[LoRARequest] = None,
trace_headers: Optional[Mapping[str, str]] = None,
prompt_adapter_request: Optional[PromptAdapterRequest] = None,
priority: int = 0,
) -> AsyncGenerator[RequestOutput, None]:
...
@@ -407,6 +409,7 @@ class MQLLMEngineClient:
lora_request: Optional[LoRARequest] = None,
trace_headers: Optional[Mapping[str, str]] = None,
prompt_adapter_request: Optional[PromptAdapterRequest] = None,
priority: int = 0,
*,
inputs: Optional[PromptType] = None # DEPRECATED
) -> AsyncGenerator[RequestOutput, None]:
@@ -425,6 +428,9 @@ class MQLLMEngineClient:
trace_headers: OpenTelemetry trace headers.
prompt_adapter_request: Prompt Adapter request to use
for generation, if any.
priority: Priority of the request (lower means earlier handling).
Any priority other than 0 will lead to an error if the
scheduling policy is not "priority".
"""
if inputs is not None:
prompt = inputs
@@ -433,7 +439,7 @@ class MQLLMEngineClient:
return self._process_request(prompt, sampling_params, request_id,
lora_request, trace_headers,
prompt_adapter_request)
prompt_adapter_request, priority)
@overload # DEPRECATED
def encode(
@@ -444,6 +450,7 @@ class MQLLMEngineClient:
request_id: str,
lora_request: Optional[LoRARequest] = None,
trace_headers: Optional[Mapping[str, str]] = None,
priority: int = 0,
) -> AsyncGenerator[EmbeddingRequestOutput, None]:
...
@@ -455,6 +462,7 @@ class MQLLMEngineClient:
request_id: str,
lora_request: Optional[LoRARequest] = None,
trace_headers: Optional[Mapping[str, str]] = None,
priority: int = 0,
) -> AsyncGenerator[EmbeddingRequestOutput, None]:
...
@@ -469,6 +477,7 @@ class MQLLMEngineClient:
request_id: Optional[str] = None,
lora_request: Optional[LoRARequest] = None,
trace_headers: Optional[Mapping[str, str]] = None,
priority: int = 0,
*,
inputs: Optional[PromptType] = None # DEPRECATED
) -> AsyncGenerator[EmbeddingRequestOutput, None]:
@@ -496,7 +505,7 @@ class MQLLMEngineClient:
and request_id is not None)
return self._process_request(prompt, pooling_params, request_id,
lora_request, trace_headers)
lora_request, trace_headers, priority)
async def _process_request(
self,
@@ -505,7 +514,8 @@ class MQLLMEngineClient:
request_id: str,
lora_request: Optional[LoRARequest] = None,
trace_headers: Optional[Mapping[str, str]] = None,
prompt_adapter_request: Optional[PromptAdapterRequest] = None
prompt_adapter_request: Optional[PromptAdapterRequest] = None,
priority: int = 0,
) -> Union[AsyncGenerator[RequestOutput, None], AsyncGenerator[
EmbeddingRequestOutput, None]]:
"""Send an RPCGenerateRequest to the RPCServer and stream responses."""
@@ -550,7 +560,9 @@ class MQLLMEngineClient:
request_id=request_id,
lora_request=lora_request,
trace_headers=trace_headers,
prompt_adapter_request=prompt_adapter_request))
prompt_adapter_request=prompt_adapter_request,
priority=priority,
))
# 3) Send the RPCGenerateRequest to the MQLLMEngine.
parts = (request_bytes,