Stop stacking Ollama -cpu embed tags and verify Assistent Sqlite after seed.

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-08-23 06:45:04 +03:00
parent 2d404436bd
commit ca11467a0d
6 changed files with 367 additions and 43 deletions
@@ -54,3 +54,56 @@ def test_write_preset_includes_memory(tmp_path: Path):
    entries = parse_ollama_models(path)
    assert preferred_ollama_model(path) is not None
    assert preferred_ollama_model(path) != MEMORY_EMBED_MODEL
+
+
+def test_cpu_ollama_tag_does_not_stack():
+    from gpu_rent.llm_runtime import (
+        cpu_ollama_tag,
+        finalize_memory_role_tags,
+        is_stacked_cpu_ollama_tag,
+        ollama_model_root,
+        pick_ollama_from_tag,
+    )
+
+    assert ollama_model_root("nomic-embed-text-cpu-cpu-cpu") == "nomic-embed-text"
+    assert cpu_ollama_tag("nomic-embed-text") == "nomic-embed-text-cpu"
+    assert cpu_ollama_tag("nomic-embed-text-cpu") == "nomic-embed-text-cpu"
+    assert cpu_ollama_tag("nomic-embed-text-cpu-cpu:latest") == "nomic-embed-text-cpu"
+    assert is_stacked_cpu_ollama_tag("nomic-embed-text-cpu-cpu")
+    assert not is_stacked_cpu_ollama_tag("nomic-embed-text-cpu")
+
+    have = {
+        "nomic-embed-text",
+        "nomic-embed-text-cpu",
+        "nomic-embed-text-cpu-cpu",
+        "nomic-embed-text-cpu-cpu-cpu",
+    }
+    assert pick_ollama_from_tag(have, "nomic-embed-text") == "nomic-embed-text"
+    assert pick_ollama_from_tag(have, "nomic-embed-text-cpu") == "nomic-embed-text"
+    # Re-provision must not FROM a stacked tag
+    assert pick_ollama_from_tag(
+        {"nomic-embed-text-cpu-cpu", "nomic-embed-text-cpu"},
+        "nomic-embed-text-cpu",
+    ) == "nomic-embed-text-cpu"
+
+    assert finalize_memory_role_tags(
+        ["nomic-embed-text", "nomic-embed-text-cpu-cpu-cpu"],
+        prefer_cpu=True,
+    ) == ["nomic-embed-text-cpu"]
+    assert finalize_memory_role_tags(
+        ["nomic-embed-text-cpu-cpu", "nomic-embed-text"],
+        prefer_cpu=True,
+    ) == ["nomic-embed-text-cpu"]
+
+
+def test_roles_payload_collapses_duplicate_memory_roots():
+    from gpu_rent.llm_runtime import OllamaModelEntry, ollama_roles_payload
+
+    roles = ollama_roles_payload(
+        [
+            OllamaModelEntry(name="chat:7b", default=True, use="chat"),
+            OllamaModelEntry(name="nomic-embed-text", use="memory"),
+            OllamaModelEntry(name="nomic-embed-text-cpu-cpu", use="memory"),
+        ]
+    )
+    assert roles["memory"] == ["nomic-embed-text"]