gardel/Mofox-Core
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Merge pull request #1057 from Todysheep/dev

Browse Source 
feat: 更新LLMRequest类以支持自定义参数，更新payload键值添加逻辑，兼容不支持某些键值的api
This commit is contained in:
SengokuCola
2025-07-06 22:15:00 +08:00
committed by GitHub
parent a3a3d872fa 7961a1f04c
commit 02b747e2fc
1 changed files with 13 additions and 30 deletions
Download Patch File Download Diff File Expand all files Collapse all files

43
src/llm_models/utils_model.py
View File

@@ -124,14 +124,21 @@ class LLMRequest:
self.model_name: str = model["name"] self.model_name: str = model["name"]
self.params = kwargs self.params = kwargs
self.enable_thinking = model.get("enable_thinking", False) self.enable_thinking = model.get("enable_thinking", None)
self.temp = model.get("temp", 0.7) self.temp = model.get("temp", 0.7)
self.thinking_budget = model.get("thinking_budget", 4096) self.thinking_budget = model.get("thinking_budget", None)
self.stream = model.get("stream", False) self.stream = model.get("stream", False)
self.pri_in = model.get("pri_in", 0) self.pri_in = model.get("pri_in", 0)
self.pri_out = model.get("pri_out", 0) self.pri_out = model.get("pri_out", 0)
self.max_tokens = model.get("max_tokens", global_config.model.model_max_output_length) self.max_tokens = model.get("max_tokens", global_config.model.model_max_output_length)
# print(f"max_tokens: {self.max_tokens}") # print(f"max_tokens: {self.max_tokens}")
custom_params_str = model.get("custom_params", "{}")
try:
self.custom_params = json.loads(custom_params_str)
except json.JSONDecodeError as e:
logger.error(f"Invalid JSON in custom_params for model '{self.model_name}': {custom_params_str}")
self.custom_params = {}
# 获取数据库实例 # 获取数据库实例
self._init_database() self._init_database()
@@ -249,28 +256,6 @@ class LLMRequest:
elif payload is None: elif payload is None:
payload = await self._build_payload(prompt) payload = await self._build_payload(prompt)
if stream_mode:
payload["stream"] = stream_mode
if self.temp != 0.7:
payload["temperature"] = self.temp
# 添加enable_thinking参数（如果不是默认值False）
if not self.enable_thinking:
payload["enable_thinking"] = False
if self.thinking_budget != 4096:
payload["thinking_budget"] = self.thinking_budget
if self.max_tokens:
payload["max_tokens"] = self.max_tokens
# if "max_tokens" not in payload and "max_completion_tokens" not in payload:
# payload["max_tokens"] = global_config.model.model_max_output_length
# 如果 payload 中依然存在 max_tokens 且需要转换，在这里进行再次检查
if self.model_name.lower() in self.MODELS_NEEDING_TRANSFORMATION and "max_tokens" in payload:
payload["max_completion_tokens"] = payload.pop("max_tokens")
return { return {
"policy": policy, "policy": policy,
"payload": payload, "payload": payload,
@@ -670,18 +655,16 @@ class LLMRequest:
if self.temp != 0.7: if self.temp != 0.7:
payload["temperature"] = self.temp payload["temperature"] = self.temp
# 添加enable_thinking参数（如果不是默认值False） # 仅当配置文件中存在参数时，添加对应参数
if not self.enable_thinking: if self.enable_thinking is not None:
payload["enable_thinking"] = False payload["enable_thinking"] = self.enable_thinking
if self.thinking_budget != 4096: if self.thinking_budget is not None:
payload["thinking_budget"] = self.thinking_budget payload["thinking_budget"] = self.thinking_budget
if self.max_tokens: if self.max_tokens:
payload["max_tokens"] = self.max_tokens payload["max_tokens"] = self.max_tokens
# if "max_tokens" not in payload and "max_completion_tokens" not in payload:
# payload["max_tokens"] = global_config.model.model_max_output_length
# 如果 payload 中依然存在 max_tokens 且需要转换，在这里进行再次检查 # 如果 payload 中依然存在 max_tokens 且需要转换，在这里进行再次检查
if self.model_name.lower() in self.MODELS_NEEDING_TRANSFORMATION and "max_tokens" in payload: if self.model_name.lower() in self.MODELS_NEEDING_TRANSFORMATION and "max_tokens" in payload:
payload["max_completion_tokens"] = payload.pop("max_tokens") payload["max_completion_tokens"] = payload.pop("max_tokens")