80 lines
3.1 KiB
Plaintext
80 lines
3.1 KiB
Plaintext
# completions.dspy — OpenAI 兼容 LLM 推理端点（产线平台统一模型入口，分类照 llmage）
|
||
#
|
||
# URL: /pipeline_llm/api/v1/chat/completions
|
||
# OpenAI 兼容客户端配置：
|
||
# base_url = http://<host>/pipeline_llm/api/v1
|
||
# api_key = <短期 token>（pipeline_llm_tokens 签发，非真实模型 key）
|
||
#
|
||
# 鉴权：Authorization: Bearer *** token>，不依赖登录会话（运行环境无 session）。
|
||
# 机构隔离：按 token 绑定的 org_id 走治理链解析真实模型与 key，真 key 不出服务进程。
|
||
#
|
||
# 调用链：token 校验 → chat_inference（门禁链①-⑥+预授权）→ 上游调用 → 结算（双维度记账）。
|
||
# 失败返回 OpenAI 错误结构，消息真实可行动。
|
||
|
||
auth = ''
|
||
try:
|
||
auth = request.headers.get('Authorization', '') or ''
|
||
except Exception:
|
||
auth = ''
|
||
if not auth:
|
||
auth = (params_kw or {}).get('api_key', '') or ''
|
||
|
||
if not auth:
|
||
return json.dumps({"error": {"message": "缺少 Authorization Bearer token",
|
||
"type": "invalid_request_error", "code": "missing_token"}},
|
||
ensure_ascii=False)
|
||
|
||
payload = None
|
||
try:
|
||
payload = await request.json()
|
||
except Exception:
|
||
payload = None
|
||
if not isinstance(payload, dict):
|
||
try:
|
||
_raw = await request.text()
|
||
payload = json.loads(_raw) if _raw else None
|
||
except Exception:
|
||
payload = None
|
||
if not isinstance(payload, dict):
|
||
_pk = dict(params_kw or {})
|
||
_msgs = _pk.get('messages')
|
||
if isinstance(_msgs, str):
|
||
try:
|
||
_msgs = json.loads(_msgs)
|
||
except Exception:
|
||
_msgs = None
|
||
payload = {"model": _pk.get('model', ''), "messages": _msgs} if _msgs else None
|
||
|
||
if not isinstance(payload, dict) or not payload.get('messages'):
|
||
return json.dumps({"error": {"message": "请求体缺 messages",
|
||
"type": "invalid_request_error", "code": "missing_messages"}},
|
||
ensure_ascii=False)
|
||
|
||
# token 校验（沿用现有短期 token 机制：签发/校验/吊销/机构隔离齐全）
|
||
ok, info = await verify_llm_token(auth)
|
||
if not ok:
|
||
return json.dumps({"error": {"message": str(info),
|
||
"type": "invalid_request_error", "code": "invalid_token"}},
|
||
ensure_ascii=False)
|
||
|
||
org_id = info.get('org_id', '') or ''
|
||
user_id = info.get('created_by', '') or ''
|
||
# token 绑定 model_name 则强制（防运行环境越权指定贵模型）；否则用请求里的 model
|
||
task_ref = 'v1:%s' % (info.get('project_id') or '')
|
||
|
||
try:
|
||
data = await llm_chat_inference(
|
||
org_id, user_id, payload,
|
||
model_name=(info.get('model_name') or payload.get('model') or ''),
|
||
task_ref=task_ref)
|
||
# 回填 token 用量
|
||
try:
|
||
await record_llm_token_usage(info.get('id', ''), data.get('usage') or {})
|
||
except Exception:
|
||
pass
|
||
return json.dumps(data, ensure_ascii=False, default=str)
|
||
except Exception as e:
|
||
return json.dumps({"error": {"message": str(e),
|
||
"type": "invalid_request_error", "code": "govern_error"}},
|
||
ensure_ascii=False)
|