chore: support `bge-m3` and `gte-multilingual-base` models

README.md

@@ -21,7 +21,7 @@ header: mini
 - **Reranking Support**: Includes powerful reranking capabilities for both text and image inputs.
 - **Optimized for Speed**: Built with lightweight transformer models and efficient backends for rapid inference, even on low-resource systems.
 - **Flexible Model Support**: Use a range of transformer models tailored to diverse use cases:
-  - Text models: `multilingual-e5-small`, `multilingual-e5-base`, `multilingual-e5-large`, `snowflake-arctic-embed-l-v2.0`, `paraphrase-multilingual-MiniLM-L12-v2`, `paraphrase-multilingual-mpnet-base-v2`, `bge-m3`.
+  - Text models: `snowflake-arctic-embed-l-v2.0`, `bge-m3`, `gte-multilingual-base`, `paraphrase-multilingual-MiniLM-L12-v2`, `paraphrase-multilingual-mpnet-base-v2`, `multilingual-e5-small`, `multilingual-e5-base`, `multilingual-e5-large`.
   - Image model: `siglip-base-patch16-256-multilingual`
 - **Production-Ready**: Easily deploy anywhere with Docker for hassle-free setup.
 - **Interactive Playground**: Test embeddings and reranking directly via a **Gradio-powered interface** alongside detailed REST API documentation.

lightweight_embeddings/__init__.py

@@ -13,6 +13,7 @@ Supported text model IDs:
 - "paraphrase-multilingual-MiniLM-L12-v2"
 - "paraphrase-multilingual-mpnet-base-v2"
 - "bge-m3"
+- "gte-multilingual-base"
 
 Supported image model ID:
 - "siglip-base-patch16-256-multilingual"
@@ -135,12 +136,13 @@ def create_main_interface():
     # Available model options for the dropdown
     model_options = [
         "snowflake-arctic-embed-l-v2.0",
+        "bge-m3",
+        "gte-multilingual-base",
+        "paraphrase-multilingual-MiniLM-L12-v2",
+        "paraphrase-multilingual-mpnet-base-v2",
         "multilingual-e5-small",
         "multilingual-e5-base",
         "multilingual-e5-large",
-        "paraphrase-multilingual-MiniLM-L12-v2",
-        "paraphrase-multilingual-mpnet-base-v2",
-        "bge-m3",
         "siglip-base-patch16-256-multilingual",
     ]
 

lightweight_embeddings/router.py

@@ -11,6 +11,7 @@ Supported Text Model IDs:
 - "paraphrase-multilingual-MiniLM-L12-v2"
 - "paraphrase-multilingual-mpnet-base-v2"
 - "bge-m3"
+- "gte-multilingual-base"
 
 Supported Image Model IDs:
 - "siglip-base-patch16-256-multilingual"

lightweight_embeddings/service.py

@@ -18,6 +18,7 @@ Supported Text Model IDs:
 - "paraphrase-multilingual-MiniLM-L12-v2"
 - "paraphrase-multilingual-mpnet-base-v2"
 - "bge-m3"
+- "gte-multilingual-base"
 
 Supported Image Model IDs:
 - "google/siglip-base-patch16-256-multilingual" (default, but extensible)
@@ -56,6 +57,7 @@ class TextModelType(str, Enum):
     PARAPHRASE_MULTILINGUAL_MINILM_L12_V2 = "paraphrase-multilingual-MiniLM-L12-v2"
     PARAPHRASE_MULTILINGUAL_MPNET_BASE_V2 = "paraphrase-multilingual-mpnet-base-v2"
     BGE_M3 = "bge-m3"
+    GTE_MULTILINGUAL_BASE = "gte-multilingual-base"
 
 
 class ImageModelType(str, Enum):
@@ -122,8 +124,12 @@ class ModelConfig:
                 onnx_file="onnx/model_quantized.onnx",
             ),
             TextModelType.BGE_M3: ModelInfo(
-                model_id="BAAI/bge-m3",
-                onnx_file="onnx/model.onnx",
+                model_id="Xenova/bge-m3",
+                onnx_file="onnx/model_quantized.onnx",
+            ),
+            TextModelType.GTE_MULTILINGUAL_BASE: ModelInfo(
+                model_id="onnx-community/gte-multilingual-base",
+                onnx_file="onnx/model_quantized.onnx",
             ),
         }
         return text_configs[self.text_model_type]