feat(qwen): enforce 1024-dimensional embeddings

scripts/test-retrieval-server.py

@@ -16,6 +16,10 @@ MODULE = importlib.util.module_from_spec(SPEC)
 SPEC.loader.exec_module(MODULE)
 
 
+def embedding():
+    return array('f', [1.0] + [0.0] * (MODULE.EMBEDDING_DIMENSIONS - 1))
+
+
 class RetrievalServerTests(unittest.TestCase):
     @unittest.skipUnless(os.environ.get('QWEN_TEST_MODEL_CACHE'), 'Set QWEN_TEST_MODEL_CACHE to run the real-model check.')
     def test_real_models(self):
@@ -52,7 +56,7 @@ class RetrievalServerTests(unittest.TestCase):
             def encode(values, **kwargs):
                 self.assertEqual(kwargs.get('prompt_name'), 'query' if values == ['query'] else None)
                 encoded.extend(values)
-                return [array('f', [1.0]) for _ in values]
+                return [embedding() for _ in values]
 
             for index in indexes:
                 index.encoder = lambda: SimpleNamespace(encode=encode)
@@ -88,7 +92,7 @@ class RetrievalServerTests(unittest.TestCase):
             root = Path(temporary)
             (root / 'a.md').write_text('before', encoding='utf-8')
             index = MODULE.Index(root, root / 'data')
-            index.encoder = lambda: SimpleNamespace(encode=lambda values, **kwargs: [array('f', [1.0]) for _ in values])
+            index.encoder = lambda: SimpleNamespace(encode=lambda values, **kwargs: [embedding() for _ in values])
             try:
                 index.rebuild()
                 (root / 'a.md').write_text('after', encoding='utf-8')
@@ -97,7 +101,7 @@ class RetrievalServerTests(unittest.TestCase):
                 def encode(values, **kwargs):
                     if values == ['fail']:
                         raise RuntimeError('encoder failed')
-                    return [array('f', [1.0]) for _ in values]
+                    return [embedding() for _ in values]
 
                 index.encoder = lambda: SimpleNamespace(encode=encode)
                 with self.assertRaisesRegex(RuntimeError, 'encoder failed'):

shared/retrieval/server.py

@@ -15,6 +15,7 @@ import threading
 
 MODEL = "Qwen/Qwen3-Embedding-0.6B"
 RERANKER_MODEL = "Qwen/Qwen3-Reranker-0.6B"
+EMBEDDING_DIMENSIONS = 1024
 RETRIEVAL_INSTRUCTION = "Given a codebase question, retrieve relevant code and documentation that answer the question."
 TEXT_EXTENSIONS = {".c", ".cpp", ".cs", ".go", ".java", ".js", ".json", ".md", ".py", ".ps1", ".rs", ".sh", ".toml", ".ts", ".tsx", ".txt", ".yaml", ".yml"}
 
@@ -31,7 +32,7 @@ class Index:
         self.root = root.resolve()
         self.data = data.resolve()
         self.data.mkdir(parents=True, exist_ok=True)
-        identity = os.path.normcase(str(self.root)) + "\n" + MODEL
+        identity = os.path.normcase(str(self.root)) + "\n" + MODEL + "\n" + str(EMBEDDING_DIMENSIONS)
         database = hashlib.sha256(identity.encode()).hexdigest() + ".sqlite3"
         self.db = sqlite3.connect(self.data / database, check_same_thread=False, timeout=30)
         self.db.execute("create table if not exists chunks (path text, text text, vector blob)")
@@ -62,7 +63,8 @@ class Index:
     def encoder(self):
         if self.model is None:
             from sentence_transformers import SentenceTransformer
-            self.model = SentenceTransformer(MODEL, prompts={"query": f"Instruct: {RETRIEVAL_INSTRUCTION}\nQuery: "})
+            self.model = SentenceTransformer(MODEL, truncate_dim=EMBEDDING_DIMENSIONS,
+                                             prompts={"query": f"Instruct: {RETRIEVAL_INSTRUCTION}\nQuery: "})
         return self.model
 
     def reranker(self):