Merge pull request #166 from stacklok/remove_models_v2

feat: replace models with lfs pointer files, and fix testing

Author: lukehinds

README.md

@@ -122,6 +122,12 @@ source venv/bin/activate  # On Windows: venv\Scripts\activate
 pip install -e ".[dev]"
 ```
 
+```bash
+# Download models
+git lfs install
+git lfs pull
+```
+
 ### Testing
 ```bash
 pytest

models/all-minilm-L6-v2-q5_k_m.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:60c7e141495321c7d303ec5ccc79296cfeb044263af840c583fed695d423aee8
+size 21717952

models/qwen2.5-coder-1.5b-instruct-q5_k_m.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f15d02e8e51a5c6f9448b972819acfa66aee4d8bb7d881a8b8ba3d90da08ef09
+size 1285494336

tests/conftest.py

@@ -95,8 +95,3 @@ def parse_json_log(log_line: str) -> dict[str, Any]:
         return json.loads(log_line)
     except json.JSONDecodeError as e:
         pytest.fail(f"Invalid JSON log line: {e}")
-
-
-@pytest.fixture
-def inference_engine() -> LlamaCppInferenceEngine:
-    return LlamaCppInferenceEngine()

tests/test_inference.py

@@ -34,23 +34,21 @@ def mock_inference_engine():
 
 @pytest.mark.asyncio
 async def test_search(mock_weaviate_client, mock_inference_engine):
-    # Patch the WeaviateClient and LlamaCppInferenceEngine inside the test function
-    with (
-        patch("weaviate.WeaviateClient", return_value=mock_weaviate_client),
-        patch(
-            "codegate.inference.inference_engine.LlamaCppInferenceEngine",
-            return_value=mock_inference_engine,
-        ),
-    ):
-
-        # Initialize StorageEngine
-        storage_engine = StorageEngine(data_path="./weaviate_data")
-
-        # Invoke the search method
-        results = await storage_engine.search("test query", 5, 0.3)
-
-        # Assertions to validate the expected behavior
-        assert len(results) == 1  # Assert that one result is returned
-        assert results[0]["properties"]["name"] == "test"
-        mock_weaviate_client.connect.assert_called()
-        mock_weaviate_client.close.assert_called()
+    # Patch the LlamaCppInferenceEngine.embed method (not the entire class)
+    with patch("codegate.inference.inference_engine.LlamaCppInferenceEngine.embed",
+               mock_inference_engine.embed):
+
+        # Mock the WeaviateClient as before
+        with patch("weaviate.WeaviateClient", return_value=mock_weaviate_client):
+
+            # Initialize StorageEngine
+            storage_engine = StorageEngine(data_path="./weaviate_data")
+
+            # Invoke the search method
+            results = await storage_engine.search("test query", 5, 0.3)
+
+            # Assertions to validate the expected behavior
+            assert len(results) == 1  # Assert that one result is returned
+            assert results[0]["properties"]["name"] == "test"
+            mock_weaviate_client.connect.assert_called()
+            mock_weaviate_client.close.assert_called()