diff --git a/dat/qwen3.7-max.txt b/dat/qwen3.7-max.txt new file mode 100644 index 0000000..f07f135 --- /dev/null +++ b/dat/qwen3.7-max.txt @@ -0,0 +1,53 @@ + 已生成完整的 qwen3.7-max 配置SQL。以下是配置方案: + + 模型摘要 + - 模型名称: 千问3.7-Max + - API model: qwen3.7-max (严格按API文档填写) + - 分类: text2text (文生文) + - 供应商: 阿里百炼 (ali-qwen) + - 接口: OpenAI兼容 /chat/completions,同步流式 + + 复用vs新建 + - upapp: 复用 ali-qwen (阿里百炼) + - upappkey: 复用 gCMl-BnrTrfoqWAEPPoH8 (API Key已配置) + - uapi: 复用 t2t (BY2cA4RD_axNxTG4m9Coa) + - uapiio: 复用 Is8l4TGkcZcqFSjbbeIK2 (文本会话) + - llm: 新建 + - llm_api_map: 新建 + - pricing_program: 新建 + - pricing_program_timing: 新建 + + 生成的SQL (4条) + + sql + -- 1. llm表 + INSERT INTO llm VALUES ('Jt26VKlUFsJxABuVLauBc','千问3.7-Max','qwen3.7-max','通义千问3.7系列旗舰模型,相较3.6版本在推理能力、代码能力、多模态理解等方面全面升级。适用于复杂任务处理、智能体开发、代码生成等高级应用场景。','text2text','qwen','ali-qwen','t2t','6fadgewjraOyvxC_EkHou','0','2026-05-23','9999-12-31','',NULL,10.00); + + -- 2. llm_api_map表 + INSERT INTO llm_api_map VALUES ('rFS0Ag-epBUxE5LCF6glW','Jt26VKlUFsJxABuVLauBc','text2text','t2t',NULL,NULL,'0B-JNSUEelsY5TQ_Sawa7','1'); + + -- 3. pricing_program表 + INSERT INTO pricing_program VALUES ('0B-JNSUEelsY5TQ_Sawa7','qwen3.7-max定价','fields:\r\n formula:\r\n label: 计算公式\r\n type: str\r\n model:\r\n label: 模型\r\n type: str\r\n prompt_tokens:\r\n label: 输入词元数\r\n type: int\r\n value_mode: between\r\nformula: null\r\npricings:\r\n- formula: 2.5 prompt_tokens / 1000000 + completion_tokens 10 /1000000\r\n model: qwen3.7-max\r\n prompt_tokens: -0.1 ~= 32000\r\n- formula: 4 prompt_tokens / 1000000 + completion_tokens 16 /1000000\r\n model: qwen3.7-max\r\n prompt_tokens: 32000 ~= 128000\r\n- formula: 7 prompt_tokens / 1000000 + completion_tokens 28 /1000000\r\n model: qwen3.7-max\r\n prompt_tokens: 128000 ~= 256000\r\n- formula: 10 prompt_tokens / 1000000 + completion_tokens 40 /1000000\r\n model: qwen3.7-max\r\n prompt_tokens: 256000 ~= 1000000','2026-05-23','9999-12-31'); + + -- 4. pricing_program_timing表 + INSERT INTO pricing_program_timing VALUES ('C-IIEG_9_O9beN8S2vdUI','0B-JNSUEelsY5TQ_Sawa7','qwen3.7-max计费','fields:\r\n formula:\r\n label: 计算公式\r\n type: str\r\n model:\r\n label: 模型\r\n type: str\r\n prompt_tokens:\r\n label: 输入词元数\r\n type: int\r\n value_mode: between\r\nformula: null\r\npricings:\r\n- formula: 2.5 prompt_tokens / 1000000 + completion_tokens 10 /1000000\r\n model: qwen3.7-max\r\n prompt_tokens: -0.1 ~= 32000\r\n- formula: 4 prompt_tokens / 1000000 + completion_tokens 16 /1000000\r\n model: qwen3.7-max\r\n prompt_tokens: 32000 ~= 128000\r\n- formula: 7 prompt_tokens / 1000000 + completion_tokens 28 /1000000\r\n model: qwen3.7-max\r\n prompt_tokens: 128000 ~= 256000\r\n- formula: 10 prompt_tokens / 1000000 + completion_tokens 40 /1000000\r\n model: qwen3.7-max\r\n prompt_tokens: 256000 ~= 1000000','2026-05-23','9999-12-31'); + + + 验证查询 + sql + SELECT id, name, model, upappid, apiname FROM llm WHERE model = 'qwen3.7-max'; + SELECT id, llmid, llmcatelogid, apiname, ppid FROM llm_api_map WHERE llmid = 'Jt26VKlUFsJxABuVLauBc'; + SELECT id, name FROM pricing_program WHERE id = '0B-JNSUEelsY5TQ_Sawa7'; + + + ppid三处一致 + - pricing_program.id = 0B-JNSUEelsY5TQ_Sawa7 + - pricing_program_timing.ppid = 0B-JNSUEelsY5TQ_Sawa7 + - llm_api_map.ppid = 0B-JNSUEelsY5TQ_Sawa7 + + 注意事项 + 1. 定价参考qwen3.6-plus的阶梯式计费模式,具体价格需根据百炼官方定价调整 + 2. qwen3.7-max使用已有的t2t uapi,OpenAI兼容接口 + 3. 如果qwen3.7-max有特殊的推理模式(如思考/非思考切换),可能需要额外的配置 + 4. 之前提到的 startReasoning is not defined 前端报错问题需要单独排查,可能与harnessed_reasoning的bricks前端代码有关 + diff --git a/llmage/__pycache__/accounting.cpython-310.pyc b/llmage/__pycache__/accounting.cpython-310.pyc new file mode 100644 index 0000000..efd3830 Binary files /dev/null and b/llmage/__pycache__/accounting.cpython-310.pyc differ diff --git a/llmage/accounting.py b/llmage/accounting.py index e076826..ef9087a 100644 --- a/llmage/accounting.py +++ b/llmage/accounting.py @@ -1,7 +1,7 @@ import asyncio import json import time -from datetime import datetime +from datetime import datetime, timedelta from appPublic.log import exception, debug from appPublic.uniqueID import getID from appPublic.dictObject import DictObject @@ -239,20 +239,18 @@ async def llm_accoung_failed(luid, reason=None): await sor.C('llmusage_accounting_failed', failed_rec) -async def backup_accounted_llmusage(): - """Backup accounted records with use_date before today to history table.""" +async def backup_accounted_llmusage(cutoff_date): + """Backup accounted records with use_date < cutoff_date to history table.""" env = ServerEnv() - today = datetime.now().strftime('%Y-%m-%d') ts = env.timestampstr() batched = 0 async with get_sor_context(env, 'llmage') as sor: - # Select records with use_date < today (i.e. yesterday and earlier) sql = """select * from llmusage where accounting_status='accounted' - and use_date < ${today}$""" - recs = await sor.sqlExe(sql, {'today': today}) + and use_date < ${cutoff_date}$""" + recs = await sor.sqlExe(sql, {'cutoff_date': cutoff_date}) if not recs: - debug(f'backup_accounted_llmusage: no records to backup for use_date < {today}') + debug(f'backup_accounted_llmusage: no records to backup for use_date < {cutoff_date}') return 0 debug(f'backup_accounted_llmusage: {len(recs)} records to backup') for r in recs: @@ -280,7 +278,7 @@ where accounting_status='accounted' # Delete from main table await sor.D('llmusage', {'id': r.id}) batched += 1 - debug(f'backup_accounted_llmusage: backed up {batched} records') + debug(f'backup_accounted_llmusage: backed up {batched} records for use_date < {cutoff_date}') return batched @@ -335,14 +333,13 @@ order by failed_time desc limit {page_size} offset {offset}""" async def backend_accounting(): env = ServerEnv() debug(f'backend accounting started ...') - backup_counter = 0 + last_backup_date = None while True: try: lus = await get_accounting_llmusages() except Exception as e: exception(f'{e}') lus = [] - # debug(f'{len(lus)=} need to accounting........') for lu in lus: try: tpac = await get_user_tpac(lu.userid) @@ -356,12 +353,14 @@ async def backend_accounting(): exception(f'{e}, {lu.id=}') await llm_accoung_failed(lu.id, reason=str(e)) - # Run backup every 30 iterations (~5 minutes) - backup_counter += 1 - if backup_counter >= 30: - backup_counter = 0 + # Check if date changed, trigger backup once per day + today = datetime.now().strftime('%Y-%m-%d') + if today != last_backup_date: + yesterday = (datetime.now() - timedelta(days=1)).strftime('%Y-%m-%d') + last_backup_date = today try: - await backup_accounted_llmusage() + debug(f'date changed to {today}, triggering backup for use_date < {yesterday}') + await backup_accounted_llmusage(yesterday) except Exception as e: exception(f'backup_accounted_llmusage failed: {e}') diff --git a/models/llmusage.json b/models/llmusage.json index 7d0f6ca..3371734 100644 --- a/models/llmusage.json +++ b/models/llmusage.json @@ -123,6 +123,14 @@ "idxfields": [ "userid" ] + }, + { + "name": "idx_llmusage_accounting", + "idxtype": "index", + "idxfields": [ + "accounting_status", + "use_date" + ] } ] } \ No newline at end of file diff --git a/scripts/migrate_llmusage_history.sql b/scripts/migrate_llmusage_history.sql index 5c30e0c..9bebf84 100644 --- a/scripts/migrate_llmusage_history.sql +++ b/scripts/migrate_llmusage_history.sql @@ -69,6 +69,9 @@ CREATE INDEX idx_laf_llmid ON llmusage_accounting_failed(llmid); CREATE INDEX idx_laf_handled ON llmusage_accounting_failed(handled); CREATE INDEX idx_laf_failed_time ON llmusage_accounting_failed(failed_time); +-- 3. 为 llmusage 表添加组合索引(优化备份查询: accounting_status + use_date) +CREATE INDEX idx_llmusage_accounting ON llmusage(accounting_status, use_date); + -- ============================================================ -- 验证步骤(执行后运行): -- 1. 确认表创建成功: