use onnx backend

main.py

@@ -49,7 +49,7 @@ app.add_middleware(
 )
 
 # Initialize model and DuckDB
-model = SentenceTransformer("nomic-ai/modernbert-embed-base", device="cpu")
+model = SentenceTransformer("nomic-ai/modernbert-embed-base", backend="onnx")
 embedding_dim = model.get_sentence_embedding_dimension()
 
 # Database setup with fallback

requirements.in

@@ -8,7 +8,7 @@ pandas
 pydantic
 polars
 python-dotenv
-sentence-transformers
+sentence-transformers[onnx]
 stamina
 tqdm
 uvicorn

requirements.txt

@@ -3,7 +3,10 @@
 aiohappyeyeballs==2.4.6
     # via aiohttp
 aiohttp==3.11.12
-    # via -r requirements.in
+    # via
+    #   -r requirements.in
+    #   datasets
+    #   fsspec
 aiosignal==1.3.2
     # via aiohttp
 annotated-types==0.7.0
@@ -25,10 +28,23 @@ charset-normalizer==3.4.1
     # via requests
 click==8.1.8
     # via uvicorn
+coloredlogs==15.0.1
+    # via onnxruntime
+datasets==2.14.4
+    # via
+    #   evaluate
+    #   optimum
+dill==0.3.7
+    # via
+    #   datasets
+    #   evaluate
+    #   multiprocess
 duckdb==1.2.0
     # via -r requirements.in
 einops==0.8.1
     # via -r requirements.in
+evaluate==0.4.3
+    # via optimum
 fastapi==0.115.8
     # via -r requirements.in
 filelock==3.17.0
@@ -36,12 +52,16 @@ filelock==3.17.0
     #   huggingface-hub
     #   torch
     #   transformers
+flatbuffers==25.2.10
+    # via onnxruntime
 frozenlist==1.5.0
     # via
     #   aiohttp
     #   aiosignal
 fsspec==2025.2.0
     # via
+    #   datasets
+    #   evaluate
     #   huggingface-hub
     #   torch
 h11==0.14.0
@@ -61,9 +81,14 @@ httpx==0.28.1
 huggingface-hub==0.28.1
     # via
     #   -r requirements.in
+    #   datasets
+    #   evaluate
+    #   optimum
     #   sentence-transformers
     #   tokenizers
     #   transformers
+humanfriendly==10.0
+    # via coloredlogs
 hyperframe==6.1.0
     # via h2
 idna==3.10
@@ -84,20 +109,42 @@ multidict==6.1.0
     # via
     #   aiohttp
     #   yarl
+multiprocess==0.70.15
+    # via
+    #   datasets
+    #   evaluate
 networkx==3.4.2
     # via torch
 numpy==2.2.2
     # via
+    #   datasets
+    #   evaluate
+    #   onnx
+    #   onnxruntime
+    #   optimum
     #   pandas
     #   scikit-learn
     #   scipy
     #   transformers
+onnx==1.17.0
+    # via optimum
+onnxruntime==1.20.1
+    # via optimum
+optimum==1.24.0
+    # via sentence-transformers
 packaging==24.2
     # via
+    #   datasets
+    #   evaluate
     #   huggingface-hub
+    #   onnxruntime
+    #   optimum
     #   transformers
 pandas==2.2.3
-    # via -r requirements.in
+    # via
+    #   -r requirements.in
+    #   datasets
+    #   evaluate
 pillow==11.1.0
     # via sentence-transformers
 polars==1.22.0
@@ -106,6 +153,13 @@ propcache==0.2.1
     # via
     #   aiohttp
     #   yarl
+protobuf==5.29.3
+    # via
+    #   onnx
+    #   onnxruntime
+    #   optimum
+pyarrow==19.0.0
+    # via datasets
 pydantic==2.10.6
     # via
     #   -r requirements.in
@@ -120,12 +174,15 @@ pytz==2025.1
     # via pandas
 pyyaml==6.0.2
     # via
+    #   datasets
     #   huggingface-hub
     #   transformers
 regex==2024.11.6
     # via transformers
 requests==2.32.3
     # via
+    #   datasets
+    #   evaluate
     #   huggingface-hub
     #   transformers
 safetensors==0.5.2
@@ -149,7 +206,9 @@ stamina==24.3.0
 starlette==0.45.3
     # via fastapi
 sympy==1.13.1
-    # via torch
+    # via
+    #   onnxruntime
+    #   torch
 tenacity==9.0.0
     # via stamina
 threadpoolctl==3.5.0
@@ -157,15 +216,21 @@ threadpoolctl==3.5.0
 tokenizers==0.21.0
     # via transformers
 torch==2.6.0
-    # via sentence-transformers
+    # via
+    #   optimum
+    #   sentence-transformers
 tqdm==4.67.1
     # via
     #   -r requirements.in
+    #   datasets
+    #   evaluate
     #   huggingface-hub
     #   sentence-transformers
     #   transformers
 transformers==4.48.3
-    # via sentence-transformers
+    # via
+    #   optimum
+    #   sentence-transformers
 typing-extensions==4.12.2
     # via
     #   anyio
@@ -180,5 +245,9 @@ urllib3==2.3.0
     # via requests
 uvicorn==0.34.0
     # via -r requirements.in
+xxhash==3.5.0
+    # via
+    #   datasets
+    #   evaluate
 yarl==1.18.3
     # via aiohttp