feat: add governed embedding client with batching

2026-07-21 01:11:48 -04:00
parent 193d67da93
commit 5e01dc738f
5 changed files with 840 additions and 2 deletions
@@ -0,0 +1,83 @@
+"""真实网关 /embeddings 端点探测(M2 设计 §11.6;人类默认口径: 实现时探测)。
+
+对 .env 的 LLM 源网关发一次真实 embeddings 请求: 支持则记录向量证据,
+不支持(404/翻译为领域错误)则 skip 并把响应记录进 tests/outputs/
+(降级证据)。无 EMBED scope 配置时复用 LLM 源的 base_url/api_key。
+"""
+
+from __future__ import annotations
+
+import dataclasses
+import os
+from datetime import datetime
+from pathlib import Path
+
+import pytest
+from dotenv import dotenv_values
+
+from polygateway.errors import PolyGatewayError
+from polygateway.transports.openai_compat import OpenAICompatTransport
+from polygateway.types import SourceConfig
+
+_ENV = {k: v for k, v in {**dotenv_values(".env"), **os.environ}.items() if v is not None}
+
+pytestmark = pytest.mark.skipif(
+    "LLM__MINIMAX__1__BASE_URL" not in _ENV, reason="缺真实网关配置(.env)"
+)
+
+_OUT = Path("tests/outputs/embedding")
+
+
+def _record(name: str, lines: list[str]) -> Path:
+    _OUT.mkdir(parents=True, exist_ok=True)
+    path = _OUT / f"{name}_{datetime.now():%Y%m%d_%H%M%S}.md"
+    path.write_text("\n".join(lines) + "\n", encoding="utf-8")
+    return path
+
+
+async def test_probe_real_gateway_embeddings():
+    source = SourceConfig(
+        name="probe_1",
+        provider="minimax",
+        base_url=_ENV["LLM__MINIMAX__1__BASE_URL"],
+        api_key=_ENV["LLM__MINIMAX__1__API_KEY"],
+        model=_ENV.get("PGW_EMBED_PROBE_MODEL", "text-embedding-v1"),
+        timeout_s=30.0,
+        est_tokens=8,
+    )
+    transport = OpenAICompatTransport()
+    try:
+        result = await transport.embed(
+            texts=["polygateway embedding probe"], source=source, call_id="probe"
+        )
+    except PolyGatewayError as exc:
+        path = _record(
+            "probe_unsupported",
+            [
+                "# Embedding 端点探测: 网关不支持",
+                f"- base_url: {source.base_url}",
+                f"- model: {source.model}",
+                f"- 错误分类: {type(exc).__name__}",
+                f"- status_code: {exc.status_code}",
+                f"- 详情: {exc}",
+                "",
+                "结论: e2e 按设计 §11.6 降级,embedding 行为由 unit 全覆盖。",
+            ],
+        )
+        await transport.aclose()
+        pytest.skip(f"网关不支持 embeddings({type(exc).__name__}),证据: {path}")
+    else:
+        await transport.aclose()
+        assert result.dim > 0 and len(result.vectors) == 1
+        _record(
+            "probe_supported",
+            [
+                "# Embedding 端点探测: 网关支持",
+                f"- base_url: {source.base_url}",
+                f"- model: {source.model}",
+                f"- dim: {result.dim}",
+                f"- usage: {result.prompt_tokens}({result.usage_source})",
+                f"- 向量前 5 维: {result.vectors[0][:5]}",
+                f"- raw: {dataclasses.asdict(result)['raw']}",
+            ],
+        )