Adjusts

README.md

@@ -4,7 +4,7 @@ emoji: 📉
 colorFrom: gray
 colorTo: indigo
 sdk: gradio
-sdk_version: 5.16.1
+sdk_version: 5.32.1
 pinned: false
 app_port: 8080
 ---

app.py

@@ -4,31 +4,48 @@ import uvicorn
 from sentence_transformers import SentenceTransformer
 from sentence_transformers.util import cos_sim
 from sentence_transformers.quantization import quantize_embeddings
-
-
 import spaces
-
+from gradio_client import Client
+import json
+import os
 
 
 app = FastAPI()
 
 
-@spaces.GPU
-def embed(text):
-    return [0,1]
-    #query_embedding = Embedder.encode(text)
-    #return query_embedding.tolist();
-    
-    
     
 @app.post("/v1/embeddings")
 async def openai_embeddings(request: Request):
     body = await request.json();
+    token = request.headers.get("authorization");
+    apiName = body.get("ApiName");
+    
     print(body);
     
-    model = body['model']
+    BearerToken = None;
+    if not token is None:
+        parts = token.split(' ');
+        BearerToken = parts[1];
+        print("Using token...");
+        
+    SpacePath = body['model']
+    
+    print("Creating client...");
+    SpaceClient = Client(SpacePath, hf_token = BearerToken)
+    
+    
+    if not apiName:
+        apiName = "/embed"
+    
     text = body['input'];
-    embeddings = embed(text)
+    
+    result = SpaceClient.predict(
+            text=text,
+            api_name=apiName
+        )
+    embeddings = json.loads(result);
+ 
+
     return {
 		'object': "list"
 		,'data': [{
@@ -36,45 +53,53 @@ async def openai_embeddings(request: Request):
 			,'embedding': embeddings
 			,'index':0
 		}]
-		,'model':model
+		,'model': SpacePath
 		,'usage':{
 			 'prompt_tokens': 0
 			,'total_tokens': 0
 		}
 	}
+    
+SpaceHost = os.environ.get("SPACE_HOST");
 
-def fn(text):
-    return embed(text);
-
-with gr.Blocks(fill_height=True) as demo:
-    text = gr.Textbox();
-    embeddings = gr.Textbox()
+if not SpaceHost:
+    SpaceHost = "localhost"
     
-    text.submit(fn, [text], [embeddings]);
+with gr.Blocks() as demo:
+    gr.Markdown(f"""
+        This space allow you connect SQL Server 2025 with Hugging Face to generate embeddings!
+        First, create a ZeroGPU Space that export an endpoint called embed.
+        That endpoint must accept a parameter called text.
+        Then, create the external model using T-SQL:
+            
+        ```sql 
+            CREATE EXTERNAL MODEL HuggingFace
+            WITH (
+                  LOCATION = 'https://{SpaceHost}/v1/embeddings',
+                  API_FORMAT = 'OpenAI',
+                  MODEL_TYPE = EMBEDDINGS,
+                  MODEL = 'user/space'
+            );  
+        ```
+        
+        If you prefer, just type the space name into field bellow and we generate the right T-SQL command for you!
+        
     
+    """)
+    
+    SpaceName = gr.Textbox(label="Space")
+    EndpointName = gr.Textbox(value="/embed", label = "EndpointName");
 
-print("Loading embedding model");
-Embedder = None #SentenceTransformer("mixedbread-ai/mxbai-embed-large-v1")
-
-# demo.run_startup_events() 
- 
- 
-#demo.launch(
-#    share=False,
-#    debug=False,
-#    server_port=7860,
-#    server_name="0.0.0.0",
-#    allowed_paths=[]
-#)
 
-print("Demo run...");
-(app2,url,other) = demo.launch(prevent_thread_lock=True, server_name=None, server_port=8000);
+    
+## hack para funcionar com ZeroGPU nesse mesmo space
+#print("Demo run...");
+#(app2,url,other) = demo.launch(prevent_thread_lock=True, server_name=None, server_port=8000);
+# demo.close 
 
 print("Mounting app...");  
-GradioApp = gr.mount_gradio_app(app, demo, path="/", ssr_mode=False); 
-
+GradioApp = gr.mount_gradio_app(app, demo, path="", ssr_mode=False); 
 
-demo.close();
 
 if __name__ == '__main__':
     print("Running uviconr...");

requirements.txt

@@ -1,3 +1,4 @@
 fastapi
 uvicorn
-sentence_transformers
+sentence_transformers
+gradio-client