apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2025-10-05 00:33:03 +08:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

[format] Valid para format error info (#4035)

Browse Source 
* feat(log):add_request_and_response_log

* 报错信息与OpenAI对齐
This commit is contained in:
xiaolei373
2025-09-12 19:05:17 +08:00
committed by GitHub
parent 88ea565aba
commit 9ac539471d
11 changed files with 435 additions and 90 deletions
Download Patch File Download Diff File Expand all files Collapse all files

50
fastdeploy/entrypoints/engine_client.py
View File

@@ -31,7 +31,12 @@ from fastdeploy.inter_communicator import IPCSignal, ZmqClient
from fastdeploy.metrics.work_metrics import work_process_metrics
from fastdeploy.multimodal.registry import MultimodalRegistry
from fastdeploy.platforms import current_platform
from fastdeploy.utils import EngineError, StatefulSemaphore, api_server_logger
from fastdeploy.utils import (
EngineError,
ParameterError,
StatefulSemaphore,
api_server_logger,
)
class EngineClient:
@@ -218,42 +223,21 @@ class EngineClient:
def valid_parameters(self, data):
"""
Validate stream options
超参数（top_p、seed、frequency_penalty、temperature、presence_penalty）的校验逻辑
前置到了ChatCompletionRequest/CompletionRequest中
"""
if data.get("n") is not None:
if data["n"] != 1:
raise ValueError("n only support 1.")
raise ParameterError("n", "n only support 1.")
if data.get("max_tokens") is not None:
if data["max_tokens"] < 1 or data["max_tokens"] >= self.max_model_len:
raise ValueError(f"max_tokens can be defined [1, {self.max_model_len}).")
raise ParameterError("max_tokens", f"max_tokens can be defined [1, {self.max_model_len}).")
if data.get("reasoning_max_tokens") is not None:
if data["reasoning_max_tokens"] > data["max_tokens"] or data["reasoning_max_tokens"] < 1:
raise ValueError("reasoning_max_tokens must be between max_tokens and 1")
if data.get("top_p") is not None:
if data["top_p"] > 1 or data["top_p"] < 0:
raise ValueError("top_p value can only be defined [0, 1].")
if data.get("frequency_penalty") is not None:
if not -2.0 <= data["frequency_penalty"] <= 2.0:
raise ValueError("frequency_penalty must be in [-2, 2]")
if data.get("temperature") is not None:
if data["temperature"] < 0:
raise ValueError("temperature must be non-negative")
if data.get("presence_penalty") is not None:
if not -2.0 <= data["presence_penalty"] <= 2.0:
raise ValueError("presence_penalty must be in [-2, 2]")
if data.get("seed") is not None:
if not 0 <= data["seed"] <= 922337203685477580:
raise ValueError("seed must be in [0, 922337203685477580]")
if data.get("stream_options") and not data.get("stream"):
raise ValueError("Stream options can only be defined when `stream=True`.")
raise ParameterError("reasoning_max_tokens", "reasoning_max_tokens must be between max_tokens and 1")
# logprobs
logprobs = data.get("logprobs")
@@ -263,35 +247,35 @@ class EngineClient:
if not self.enable_logprob:
err_msg = "Logprobs is disabled, please enable it in startup config."
api_server_logger.error(err_msg)
raise ValueError(err_msg)
raise ParameterError("logprobs", err_msg)
top_logprobs = data.get("top_logprobs")
elif isinstance(logprobs, int):
top_logprobs = logprobs
elif logprobs:
raise ValueError("Invalid type for 'logprobs'")
raise ParameterError("logprobs", "Invalid type for 'logprobs'")
# enable_logprob
if top_logprobs:
if not self.enable_logprob:
err_msg = "Logprobs is disabled, please enable it in startup config."
api_server_logger.error(err_msg)
raise ValueError(err_msg)
raise ParameterError("logprobs", err_msg)
if not isinstance(top_logprobs, int):
err_type = type(top_logprobs).__name__
err_msg = f"Invalid type for 'top_logprobs': expected int but got {err_type}."
api_server_logger.error(err_msg)
raise ValueError(err_msg)
raise ParameterError("top_logprobs", err_msg)
if top_logprobs < 0:
err_msg = f"Invalid 'top_logprobs': must be >= 0, got {top_logprobs}."
api_server_logger.error(err_msg)
raise ValueError(err_msg)
raise ParameterError("top_logprobs", err_msg)
if top_logprobs > 20:
err_msg = "Invalid value for 'top_logprobs': must be <= 20."
api_server_logger.error(err_msg)
raise ValueError(err_msg)
raise ParameterError("top_logprobs", err_msg)
def check_health(self, time_interval_threashold=30):
"""