server: support Vertex AI compatible API (#22545)

* server: support Vertex AI compatible API * a bit safer * support other AIP_* env var * various fixes * if AIP_MODE is unset, do nothing * fix test case * fix windows build
2026-05-08 15:23:04 +02:00
parent 9dcf835528
commit 29debb3a6a
5 changed files with 296 additions and 4 deletions
@@ -0,0 +1,60 @@
+import pytest
+from utils import *
+
+server: ServerProcess
+
+
+@pytest.fixture(autouse=True)
+def create_server():
+    global server
+    server = ServerPreset.tinyllama2()
+    server.gcp_compat = True
+
+
+def test_gcp_predict_camel_case():
+    global server
+    server.start()
+    res = server.make_request("POST", "/predict", data={
+        "instances": [
+            {
+                "@requestFormat": "chatCompletions",
+                "max_tokens": 8,
+                "messages": [
+                    {"role": "user", "content": "What is the meaning of life?"},
+                ],
+            }
+        ],
+    })
+    assert res.status_code == 200
+    assert "predictions" in res.body
+    assert len(res.body["predictions"]) == 1
+    prediction = res.body["predictions"][0]
+    assert "choices" in prediction
+    assert len(prediction["choices"]) == 1
+    assert prediction["choices"][0]["message"]["role"] == "assistant"
+    assert len(prediction["choices"][0]["message"]["content"]) > 0
+
+
+def test_gcp_predict_multiple_instances():
+    global server
+    server.n_slots = 2
+    server.start()
+    res = server.make_request("POST", "/predict", data={
+        "instances": [
+            {
+                "@requestFormat": "chatCompletions",
+                "max_tokens": 8,
+                "messages": [{"role": "user", "content": "Say hello"}],
+            },
+            {
+                "@requestFormat": "chatCompletions",
+                "max_tokens": 8,
+                "messages": [{"role": "user", "content": "Say world"}],
+            },
+        ],
+    })
+    assert res.status_code == 200
+    assert len(res.body["predictions"]) == 2
+    for prediction in res.body["predictions"]:
+        assert "choices" in prediction
+        assert len(prediction["choices"][0]["message"]["content"]) > 0