Change default embedding model to qwen3-embedding:4b

CHANGELOG.md

@@ -17,8 +17,13 @@
 
 ### Changed
 
-- **Evaluations**: Improved evaluation dataset naming and simplified evaluator
+- **BREAKING: Default Embedding Model**: Changed default embedding model from `qwen3-embedding` to `qwen3-embedding:4b` with vector dimension 2560 (previously 4096)
-- configuration
+  - New installations will use the smaller, more efficient 4B parameter model by default
+  - **Action required**: Existing databases created with the old default will be incompatible. Users must either:
+    - Explicitly set `embeddings.model: "qwen3-embedding"` and `embeddings.vector_dim: 4096` in their config to maintain compatibility with existing databases
+    - Or run `haiku-rag rebuild` to re-embed all documents with the new default
+  - This change provides better performance for most use cases while reducing resource requirements
+- **Evaluations**: Improved evaluation dataset naming and simplified evaluator configuration
   - `EvalDataset` now accepts dataset name for better organization in Logfire
   - Added `--name` CLI parameter to override evaluation run names
   - Removed `IsInstance` evaluator, using only `LLMJudge` for QA evaluation

docs/configuration.md

@@ -36,8 +36,8 @@ environment: production
 
 embeddings:
   provider: ollama
-  model: qwen3-embedding
+  model: qwen3-embedding:4b
-  vector_dim: 4096
+  vector_dim: 2560
 
 qa:
   provider: ollama
@@ -68,8 +68,8 @@ lancedb:
 
 embeddings:
   provider: ollama
-  model: qwen3-embedding
+  model: qwen3-embedding:4b
-  vector_dim: 4096
+  vector_dim: 2560
 
 reranking:
   provider: ""  # Empty to disable, or mxbai, cohere, zeroentropy, vllm
@@ -144,7 +144,7 @@ from haiku.rag.client import HaikuRAG
 # Create custom configuration
 custom_config = AppConfig(
     qa={"provider": "openai", "model": "gpt-4o"},
-    embeddings={"provider": "ollama", "model": "qwen3-embedding"},
+    embeddings={"provider": "ollama", "model": "qwen3-embedding:4b", "vector_dim": 2560},
     processing={"chunk_size": 512}
 )
 

haiku_rag_slim/haiku/rag/config/loader.py

@@ -63,8 +63,8 @@ def generate_default_config() -> dict:
         "lancedb": {"uri": "", "api_key": "", "region": ""},
         "embeddings": {
             "provider": "ollama",
-            "model": "qwen3-embedding",
+            "model": "qwen3-embedding:4b",
-            "vector_dim": 4096,
+            "vector_dim": 2560,
         },
         "reranking": {"provider": "", "model": ""},
         "qa": {"provider": "ollama", "model": "gpt-oss"},

haiku_rag_slim/haiku/rag/config/models.py

@@ -26,8 +26,8 @@ class LanceDBConfig(BaseModel):
 
 class EmbeddingsConfig(BaseModel):
     provider: str = "ollama"
-    model: str = "qwen3-embedding"
+    model: str = "qwen3-embedding:4b"
-    vector_dim: int = 4096
+    vector_dim: int = 2560
 
 
 class RerankingConfig(BaseModel):

tests/conftest.py

@@ -44,8 +44,8 @@ def temp_yaml_config(tmp_path, monkeypatch):
         },
         "embeddings": {
             "provider": "ollama",
-            "model": "qwen3-embedding",
+            "model": "qwen3-embedding:4b",
-            "vector_dim": 4096,
+            "vector_dim": 2560,
         },
         "qa": {"provider": "ollama", "model": "gpt-oss"},
     }

tests/test_config_loader.py

@@ -108,7 +108,7 @@ def test_generate_default_config():
     assert "qa" in config
     assert "providers" in config
     assert config["embeddings"]["provider"] == "ollama"
-    assert config["embeddings"]["vector_dim"] == 4096
+    assert config["embeddings"]["vector_dim"] == 2560
 
 
 def test_config_precedence_cwd_over_user(tmp_path, monkeypatch):