red-future/model-gateway
4
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

refactor(model): 重构模型实体和数据访问层

Browse Source
This commit is contained in:
王利钊
2026-05-21 10:41:37 +08:00
parent a080a5536d
commit 170568e03e
35 changed files with 903 additions and 1072 deletions
Download Patch File Download Diff File Expand all files Collapse all files

28
update.sql
View File

@@ -40,9 +40,18 @@ CREATE TABLE IF NOT EXISTS asynch_models (
retry_queue_max_seconds INT NOT NULL DEFAULT 600, -- 失败重试最大排队时间(秒 0=插队到队首；>0=排队超过该时间后插队，否则仍到队尾)
auto_clean_seconds INT NOT NULL DEFAULT 86400, -- 已下载(state=4 后的保留时间(秒)，到期清理)
remark TEXT DEFAULT '' -- 备注
token_mapping VARCHAR(128) NOT NULL DEFAULT ''; -- token 映射
);
response_token_field VARCHAR(128) NOT NULL DEFAULT ''; -- 响应中消耗token的字段映射
operator_name VARCHAR(64) NOT NULL DEFAULT '', -- 运营商名称
token_config JSONB NOT NULL DEFAULT '{
"zh_ratio": 1.0,
"en_ratio": 1.3,
"space_ratio": 0.1,
"punctuation_ratio": 0.1,
"max_window_size": 8192,
"reserve_ratio": 0.2,
"min_reserve": 512,
}'::jsonb -- Token配置
);
CREATE UNIQUE INDEX IF NOT EXISTS uk_asynch_models_tenant_creator_chat ON asynch_models(tenant_id, creator) WHERE is_chat_model = 1 AND deleted_at IS NULL;
CREATE UNIQUE INDEX IF NOT EXISTS uk_asynch_models_tenant_model_name ON asynch_models(tenant_id, creator, model_name);
CREATE INDEX IF NOT EXISTS idx_asynch_models_tenant_id ON asynch_models(tenant_id);
@@ -83,8 +92,17 @@ COMMENT ON COLUMN asynch_models.retry_times IS '失败重试次数';
COMMENT ON COLUMN asynch_models.retry_queue_max_seconds IS '失败重试最大排队时间(秒 0=插队到队首；>0=排队超过该时间后插队，否则仍到队尾)';
COMMENT ON COLUMN asynch_models.auto_clean_seconds IS '已下载(state=4 后的保留时间(秒)，到期清理)';
COMMENT ON COLUMN asynch_models.remark IS '备注';
COMMENT ON COLUMN asynch_models.token_mapping IS 'token映射';
COMMENT ON COLUMN asynch_models.response_token_field IS '响应中消耗token的字段映射';
COMMENT ON COLUMN asynch_models.operator_name IS '运营商名称';
COMMENT ON COLUMN asynch_models.token_config IS '{
"zh_ratio": 1.0, // 中文字符→token系数
"en_ratio": 1.3, // 英文单词→token系数
"space_ratio": 0.1, // 空格系数
"punctuation_ratio": 0.1, // 标点系数
"max_window_size": 8192, // 模型最大窗口
"reserve_ratio": 0.2, // 预留回复空间比例
"min_reserve": 512, // 最少预留token数
}';
-- =========================