Warm Ollama into VRAM on up and tunnel so Assistent chat is not cold.

A 1-token /api/chat after tags (and again if /api/ps is empty) loads VL weights before the first message. Mid KEEP_ALIVE is 15m so a short image-gen burst does not unload the model.

Co-authored-by: Cursor <cursoragent@cursor.com>

tests/test_ollama_warmup.py

@@ -0,0 +1,101 @@
+"""Unit tests for remote ollama_warmup (stdlib helpers)."""
+
+from importlib.util import module_from_spec, spec_from_file_location
+from pathlib import Path
+
+_ROOT = Path(__file__).resolve().parents[1]
+_SPEC = spec_from_file_location(
+    "ollama_warmup_remote",
+    _ROOT / "src" / "gpu_rent" / "remote" / "ollama_warmup.py",
+)
+assert _SPEC and _SPEC.loader
+_mod = module_from_spec(_SPEC)
+_SPEC.loader.exec_module(_mod)
+
+
+class FakeResp:
+    def __init__(self, body: str):
+        self._body = body.encode("utf-8")
+
+    def read(self) -> bytes:
+        return self._body
+
+    def __enter__(self):
+        return self
+
+    def __exit__(self, *a):
+        return False
+
+
+def test_parse_env_file(tmp_path):
+    path = tmp_path / ".gpu-rent-ollama.env"
+    path.write_text(
+        "# comment\nOLLAMA_KEEP_ALIVE=15m\nOLLAMA_CONTEXT_LENGTH=16384\n",
+        encoding="utf-8",
+    )
+    env = _mod.parse_env_file(path)
+    assert env["OLLAMA_KEEP_ALIVE"] == "15m"
+    assert env["OLLAMA_CONTEXT_LENGTH"] == "16384"
+    assert _mod.parse_env_file(tmp_path / "missing") == {}
+
+
+def test_warmup_skips_when_ps_has_model(monkeypatch):
+    def fake_urlopen(req, timeout=None):
+        url = getattr(req, "full_url", str(req))
+        assert "/api/ps" in url
+        return FakeResp('{"models":[{"name":"huihui_ai/qwen2.5-vl-abliterated:7b"}]}')
+
+    monkeypatch.setattr(_mod.urllib.request, "urlopen", fake_urlopen)
+    msg = _mod.warmup("huihui_ai/qwen2.5-vl-abliterated:7b", keep_alive="15m", num_ctx=16384, timeout=5)
+    assert "skip" in msg
+    assert "VRAM" in msg
+
+
+def test_warmup_posts_one_token_chat(monkeypatch):
+    seen: dict = {}
+
+    def fake_urlopen(req, timeout=None):
+        url = getattr(req, "full_url", str(req))
+        method = getattr(req, "method", "GET")
+        if "/api/ps" in url:
+            return FakeResp('{"models":[]}')
+        seen["url"] = url
+        seen["method"] = method
+        seen["body"] = req.data
+        return FakeResp('{"message":{"role":"assistant","content":"ok"}}')
+
+    monkeypatch.setattr(_mod.urllib.request, "urlopen", fake_urlopen)
+    msg = _mod.warmup("foo:7b", keep_alive="15m", num_ctx=16384, timeout=5)
+    assert msg.startswith("ok foo:7b")
+    assert "/api/chat" in seen["url"]
+    assert seen["method"] == "POST"
+    import json
+
+    body = json.loads(seen["body"].decode())
+    assert body["model"] == "foo:7b"
+    assert body["stream"] is False
+    assert body["keep_alive"] == "15m"
+    assert body["options"]["num_predict"] == 1
+    assert body["options"]["num_ctx"] == 16384
+    assert body["messages"][0]["content"] == "ok"
+
+
+def test_main_never_fails_on_http_error(monkeypatch, tmp_path):
+    jobs = tmp_path / "jobs.json"
+    jobs.write_text('{"model":"foo:7b"}', encoding="utf-8")
+    monkeypatch.setattr(_mod, "JOBS", jobs)
+    monkeypatch.setattr(_mod, "ENV_FILE", tmp_path / "missing.env")
+
+    def boom(*_a, **_k):
+        raise _mod.urllib.error.URLError("down")
+
+    monkeypatch.setattr(_mod.urllib.request, "urlopen", boom)
+    assert _mod.main() == 0
+
+
+def test_warmup_body_matches_assistent_ctx():
+    import json
+
+    body = json.loads(_mod.warmup_body("m:7b", keep_alive="15m", num_ctx=16384))
+    assert body["options"]["num_ctx"] == 16384
+    assert body["options"]["num_predict"] == 1