Dynamic embeddings model (#556)

* Dynamic embeddings model selection

* Added tests

* HF embeddings are skipped, tests don't run with that package currently tests

trustgraph-flow/trustgraph/embeddings/fastembed/processor.py

@@ -27,10 +27,31 @@ class Processor(EmbeddingsService):
             **params | { "model": model }
         )
 
-        logger.info("Loading FastEmbed model...")
-        self.embeddings = TextEmbedding(model_name = model)
+        self.default_model = model
 
-    async def on_embeddings(self, text):
+        # Cache for currently loaded model
+        self.cached_model_name = None
+        self.embeddings = None
+
+        # Load the default model
+        self._load_model(model)
+
+    def _load_model(self, model_name):
+        """Load a model, caching it for reuse"""
+        if self.cached_model_name != model_name:
+            logger.info(f"Loading FastEmbed model: {model_name}")
+            self.embeddings = TextEmbedding(model_name=model_name)
+            self.cached_model_name = model_name
+            logger.info(f"FastEmbed model {model_name} loaded successfully")
+        else:
+            logger.debug(f"Using cached model: {model_name}")
+
+    async def on_embeddings(self, text, model=None):
+
+        use_model = model or self.default_model
+
+        # Reload model if it has changed
+        self._load_model(use_model)
 
         vecs = self.embeddings.embed([text])
 

trustgraph-flow/trustgraph/embeddings/ollama/processor.py

@@ -28,12 +28,14 @@ class Processor(EmbeddingsService):
         )
 
         self.client = Client(host=ollama)
-        self.model = model
+        self.default_model = model
 
-    async def on_embeddings(self, text):
+    async def on_embeddings(self, text, model=None):
+
+        use_model = model or self.default_model
 
         embeds = self.client.embed(
-            model = self.model,
+            model = use_model,
             input = text
         )