BruceSun
/
MaiBot
mirror of https://github.com/Mai-with-u/MaiBot.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

修复gemini_client的小错误

pull/1211/head
UnCLAS-Prommer 2025-08-21 23:26:45 +08:00
parent 957f5f93dd
commit 49cb5e9b28
No known key found for this signature in database
1 changed files with 21 additions and 29 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

50
src/llm_models/model_client/gemini_client.py
View File

@ -47,10 +47,13 @@ logger = get_logger("Gemini客户端")
# gemini_thinking参数（默认范围） # gemini_thinking参数（默认范围）
# 不同模型的思考预算范围配置 # 不同模型的思考预算范围配置
THINKING_BUDGET_LIMITS = { THINKING_BUDGET_LIMITS = {
"gemini-2.5-flash": {"min": 1, "max": 24576, "can_disable": True}, "gemini-2.5-flash": {"min": 1, "max": 24576, "can_disable": True},
"gemini-2.5-flash-lite": {"min": 512, "max": 24576, "can_disable": True}, "gemini-2.5-flash-lite": {"min": 512, "max": 24576, "can_disable": True},
"gemini-2.5-pro": {"min": 128, "max": 32768, "can_disable": False}, "gemini-2.5-pro": {"min": 128, "max": 32768, "can_disable": False},
} }
# 思维预算特殊值
THINKING_BUDGET_AUTO = -1 # 自动调整思考预算，由模型决定
THINKING_BUDGET_DISABLED = 0 # 禁用思考预算（如果模型允许禁用）
gemini_safe_settings = [ gemini_safe_settings = [
SafetySetting(category=HarmCategory.HARM_CATEGORY_HATE_SPEECH, threshold=HarmBlockThreshold.BLOCK_NONE), SafetySetting(category=HarmCategory.HARM_CATEGORY_HATE_SPEECH, threshold=HarmBlockThreshold.BLOCK_NONE),
@ -91,9 +94,7 @@ def _convert_messages(
for item in message.content: for item in message.content:
if isinstance(item, tuple): if isinstance(item, tuple):
image_format = "jpeg" if item[0].lower() == "jpg" else item[0].lower() image_format = "jpeg" if item[0].lower() == "jpg" else item[0].lower()
content.append( content.append(Part.from_bytes(data=base64.b64decode(item[1]), mime_type=f"image/{image_format}"))
Part.from_bytes(data=base64.b64decode(item[1]), mime_type=f"image/{image_format}")
)
elif isinstance(item, str): elif isinstance(item, str):
content.append(Part.from_text(text=item)) content.append(Part.from_text(text=item))
else: else:
@ -336,47 +337,40 @@ class GeminiClient(BaseClient):
api_key=api_provider.api_key, api_key=api_provider.api_key,
) # 这里和openai不一样，gemini会自己决定自己是否需要retry ) # 这里和openai不一样，gemini会自己决定自己是否需要retry
# 思维预算特殊值
THINKING_BUDGET_AUTO = -1 # 自动调整思考预算，由模型决定
THINKING_BUDGET_DISABLED = 0 # 禁用思考预算（如果模型允许禁用）
@staticmethod @staticmethod
def clamp_thinking_budget(tb: int, model_id: str): def clamp_thinking_budget(tb: int, model_id: str) -> int:
""" """
按模型限制思考预算范围，仅支持指定的模型（支持带数字后缀的新版本） 按模型限制思考预算范围，仅支持指定的模型（支持带数字后缀的新版本）
""" """
limits = None limits = None
matched_key = None
# 优先尝试精确匹配 # 优先尝试精确匹配
if model_id in THINKING_BUDGET_LIMITS: if model_id in THINKING_BUDGET_LIMITS:
limits = THINKING_BUDGET_LIMITS[model_id] limits = THINKING_BUDGET_LIMITS[model_id]
matched_key = model_id
else: else:
# 按 key 长度倒序，保证更长的（更具体的，如 -lite）优先 # 按 key 长度倒序，保证更长的（更具体的，如 -lite）优先
sorted_keys = sorted(THINKING_BUDGET_LIMITS.keys(), key=len, reverse=True) sorted_keys = sorted(THINKING_BUDGET_LIMITS.keys(), key=len, reverse=True)
for key in sorted_keys: for key in sorted_keys:
# 必须满足：完全等于 或者 前缀匹配（带 "-" 边界） # 必须满足：完全等于 或者 前缀匹配（带 "-" 边界）
if model_id == key or model_id.startswith(key + "-"): if model_id == key or model_id.startswith(f"{key}-"):
limits = THINKING_BUDGET_LIMITS[key] limits = THINKING_BUDGET_LIMITS[key]
matched_key = key break
break
# 特殊值处理 # 特殊值处理
if tb == GeminiClient.THINKING_BUDGET_AUTO: if tb == THINKING_BUDGET_AUTO:
return GeminiClient.THINKING_BUDGET_AUTO return THINKING_BUDGET_AUTO
if tb == GeminiClient.THINKING_BUDGET_DISABLED: if tb == THINKING_BUDGET_DISABLED:
if limits and limits.get("can_disable", False): if limits and limits.get("can_disable", False):
return GeminiClient.THINKING_BUDGET_DISABLED return THINKING_BUDGET_DISABLED
return limits["min"] if limits else GeminiClient.THINKING_BUDGET_AUTO return limits["min"] if limits else THINKING_BUDGET_AUTO
# 已知模型裁剪到范围 # 已知模型裁剪到范围
if limits: if limits:
return max(limits["min"], min(tb, limits["max"])) return max(limits["min"], min(tb, limits["max"]))
# 未知模型，返回动态模式 # 未知模型，返回动态模式
logger.warning(f"模型 {model_id} 未在 THINKING_BUDGET_LIMITS 中定义，将使用动态模式 tb=-1 兼容。") logger.warning(f"模型 {model_id} 未在 THINKING_BUDGET_LIMITS 中定义，将使用动态模式 tb=-1 兼容。")
return GeminiClient.THINKING_BUDGET_AUTO return THINKING_BUDGET_AUTO
async def get_response( async def get_response(
self, self,
@ -424,15 +418,13 @@ class GeminiClient(BaseClient):
# 将tool_options转换为Gemini API所需的格式 # 将tool_options转换为Gemini API所需的格式
tools = _convert_tool_options(tool_options) if tool_options else None tools = _convert_tool_options(tool_options) if tool_options else None
tb = GeminiClient.THINKING_BUDGET_AUTO tb = THINKING_BUDGET_AUTO
#空处理 # 空处理
if extra_params and "thinking_budget" in extra_params: if extra_params and "thinking_budget" in extra_params:
try: try:
tb = int(extra_params["thinking_budget"]) tb = int(extra_params["thinking_budget"])
except (ValueError, TypeError): except (ValueError, TypeError):
logger.warning( logger.warning(f"无效的 thinking_budget 值 {extra_params['thinking_budget']}，将使用默认动态模式 {tb}")
f"无效的 thinking_budget 值 {extra_params['thinking_budget']}，将使用默认动态模式 {tb}"
)
# 裁剪到模型支持的范围 # 裁剪到模型支持的范围
tb = self.clamp_thinking_budget(tb, model_info.model_identifier) tb = self.clamp_thinking_budget(tb, model_info.model_identifier)