reviewd text

app.py

@@ -1,162 +1,166 @@
-import gradio as gr
-import onnx
-from huggingface_hub import HfApi
-import json
-import sys
-import os
-import io
-import requests
-from urllib.parse import urlparse
-
-
-HfClient = HfApi();
-
-ONNX_PREFERED = [
-    "model.onnx",
-    "onnx/model.onnx"    
-]
-
-ONNX_CACHE = {
-    'models': {
-        
-    }
-}
-
-def is_url(path):
-    try:
-        result = urlparse(path)
-        return all([result.scheme, result.netloc])
-    except ValueError:
-        return False
-
-
-def load_model(path):
-    if is_url(path):
-        print(f"Downloading model from: {path}...", file=sys.stderr)
-        try:
-            response = requests.get(path)
-            response.raise_for_status() # Check for HTTP errors
-            # Load from binary stream
-            return onnx.load(io.BytesIO(response.content))
-        except requests.exceptions.RequestException as e:
-            print(f"Error downloading model: {e}", file=sys.stderr)
-            sys.exit(1)
-    else:
-        # Check if local file exists before loading
-        if not os.path.exists(path):
-            print(f"Error: File not found at {path}", file=sys.stderr)
-            sys.exit(1)
-        
-        return onnx.load(path)
-
-def CheckSqlOnnx(path):
-    
-    OnnxModel = load_model(path);
-    
-    initializer_names = {init.name for init in OnnxModel.graph.initializer}
-    inputs = [inp.name for inp in OnnxModel.graph.input if inp.name not in initializer_names]
-    outputs = [out.name for out in OnnxModel.graph.output]
-
-    required_inputs = {"input_ids", "attention_mask"}
-    required_outputs = {"token_embeddings", "sentence_embedding"}
-
-    is_supported = (
-        required_inputs.issubset(inputs) and
-        required_outputs.issubset(outputs)
-    )
-
-    OnnxInouts = {
-        "supported": is_supported,
-        "inputs": inputs,
-        "outputs": outputs
-    }
-    
-    return OnnxInouts;
-
-def CheckModel(repo_id: str, path: str | None = None):
-    
-    MODELS_CACHE = ONNX_CACHE['models'];
-    
-    CacheSlot = MODELS_CACHE.get(repo_id);
-    
-    if CacheSlot:
-        return json.dumps(CacheSlot, indent = 2);
-    
-    
-    
-    model_info = HfClient.model_info(repo_id=repo_id)
-
-    # Extract filenames from RepoSibling objects
-    sibling_files = [s.rfilename for s in model_info.siblings]
-
-    onnx_path = None
-
-    if path:
-        if path in sibling_files:
-            onnx_path = path
-        else:
-            return f"Error: ONNX file not found: {path}"
-    else:
-        for p in ONNX_PREFERED:
-            if p in sibling_files:
-                onnx_path = p
-                break
-
-        if not onnx_path:
-            onnx_path = next(
-                (f for f in sibling_files if f.lower().endswith(".onnx")),
-                None
-            )
-
-    if not onnx_path:
-        raise "Error: No ONNX model found in repository";
-
-    # Build Hugging Face raw file URL
-    file_url = f"https://huggingface.co/{repo_id}/resolve/main/{onnx_path}"
-
-    # Check SQL ONNX compatibility
-    OnnxInfo = CheckSqlOnnx(file_url)
-    
-    CacheSlot = {
-         'url': file_url
-        ,'onnx': OnnxInfo
-    }
-    
-    MODELS_CACHE[repo_id] = CacheSlot
-
-    return json.dumps({**CacheSlot, 'cached': False}, indent = 2);
-
-
-with gr.Blocks() as demo:
-    
-    gr.Markdown("""
-        This sample app test if a given model repository can be used with SQL Server ONNX.  
-        In some tests, discovered that is not any ONNX model that works with sql server CREATE EXTERNAL MODEL.  
-        For works, the input parameters of neural network must contains specific names, and output also.  
-        I dont know if this behavior will be ketp in future verisons of SQL 2025...  
-        
-        But, while we dont have official doc about this, this repo can help discovery if a given model will work with sql server if you plan download to use with ONNX.  
-        Just input the model name bellow in format user/model-name (check examples).  
-        
-        Look at JSON output. If "supported" is True, then you can use with SQL...
-        
-        Soon bring a default tested list!
-        
-        **IMPORTANT**: To check, this space will attempt donwload the model onnx file. If is big can take several minutes.
-    """)
-    
-    ModelPath = gr.Textbox(label="Model Repository", submit_btn = True);
-    ModelInfoOut = gr.Textbox(label="Model Info", lines = 10)
-    
-    ModelPath.submit(fn=CheckModel, inputs=ModelPath, outputs=ModelInfoOut)
-    
-    gr.Examples([
-         ["intfloat/multilingual-e5-large"]
-        ,["mixedbread-ai/mxbai-embed-xsmall-v1"]
-        ,["nsense/all-MiniLM-L6-v2-onnx"]
-    
-    ], ModelPath)
-
-demo.launch(
-    server_name = '0.0.0.0'
+import gradio as gr
+import onnx
+from huggingface_hub import HfApi
+import json
+import sys
+import os
+import io
+import requests
+from urllib.parse import urlparse
+
+
+HfClient = HfApi();
+
+ONNX_PREFERED = [
+    "model.onnx",
+    "onnx/model.onnx"    
+]
+
+ONNX_CACHE = {
+    'models': {
+        
+    }
+}
+
+def is_url(path):
+    try:
+        result = urlparse(path)
+        return all([result.scheme, result.netloc])
+    except ValueError:
+        return False
+
+
+def load_model(path):
+    if is_url(path):
+        print(f"Downloading model from: {path}...", file=sys.stderr)
+        try:
+            response = requests.get(path)
+            response.raise_for_status() # Check for HTTP errors
+            # Load from binary stream
+            return onnx.load(io.BytesIO(response.content))
+        except requests.exceptions.RequestException as e:
+            print(f"Error downloading model: {e}", file=sys.stderr)
+            sys.exit(1)
+    else:
+        # Check if local file exists before loading
+        if not os.path.exists(path):
+            print(f"Error: File not found at {path}", file=sys.stderr)
+            sys.exit(1)
+        
+        return onnx.load(path)
+
+def CheckSqlOnnx(path):
+    
+    OnnxModel = load_model(path);
+    
+    initializer_names = {init.name for init in OnnxModel.graph.initializer}
+    inputs = [inp.name for inp in OnnxModel.graph.input if inp.name not in initializer_names]
+    outputs = [out.name for out in OnnxModel.graph.output]
+
+    required_inputs = {"input_ids", "attention_mask"}
+    required_outputs = {"token_embeddings", "sentence_embedding"}
+
+    is_supported = (
+        required_inputs.issubset(inputs) and
+        required_outputs.issubset(outputs)
+    )
+
+    OnnxInouts = {
+        "supported": is_supported,
+        "inputs": inputs,
+        "outputs": outputs
+    }
+    
+    return OnnxInouts;
+
+def CheckModel(repo_id: str, path: str | None = None):
+    
+    MODELS_CACHE = ONNX_CACHE['models'];
+    
+    CacheSlot = MODELS_CACHE.get(repo_id);
+    
+    if CacheSlot:
+        return json.dumps(CacheSlot, indent = 2);
+    
+    
+    
+    model_info = HfClient.model_info(repo_id=repo_id)
+
+    # Extract filenames from RepoSibling objects
+    sibling_files = [s.rfilename for s in model_info.siblings]
+
+    onnx_path = None
+
+    if path:
+        if path in sibling_files:
+            onnx_path = path
+        else:
+            return f"Error: ONNX file not found: {path}"
+    else:
+        for p in ONNX_PREFERED:
+            if p in sibling_files:
+                onnx_path = p
+                break
+
+        if not onnx_path:
+            onnx_path = next(
+                (f for f in sibling_files if f.lower().endswith(".onnx")),
+                None
+            )
+
+    if not onnx_path:
+        raise "Error: No ONNX model found in repository";
+
+    # Build Hugging Face raw file URL
+    file_url = f"https://huggingface.co/{repo_id}/resolve/main/{onnx_path}"
+
+    # Check SQL ONNX compatibility
+    OnnxInfo = CheckSqlOnnx(file_url)
+    
+    CacheSlot = {
+         'url': file_url
+        ,'onnx': OnnxInfo
+    }
+    
+    MODELS_CACHE[repo_id] = CacheSlot
+
+    return json.dumps({**CacheSlot, 'cached': False}, indent = 2);
+
+
+with gr.Blocks() as demo:
+    
+    gr.Markdown("""
+        This sample app tests whether a given model repository can be used with SQL Server ONNX.  
+        
+        In some tests, it was discovered that not every ONNX model works with SQL Server `CREATE EXTERNAL MODEL`.  
+        For it to work, the input parameters of the neural network must contain specific names, and the output as well.  
+        
+        I don’t know if this behavior will be kept in future versions of SQL Server 2025.  
+        
+        However, while we don’t have official documentation about this, this repository can help discover whether a given model will work with SQL Server if you plan to download and use it with ONNX.  
+        
+        Just input the model name below in the format `user/model-name` (check the examples).  
+        
+        Look at the JSON output. If `"supported"` is `true`, then you can use it with SQL Server.  
+        
+        Soon, I will bring a default tested list!  
+        
+        **IMPORTANT**: To check this, the space will attempt to download the model ONNX file. If it is large, it can take several minutes.  
+
+    """)
+    
+    ModelPath = gr.Textbox(label="Model Repository", submit_btn = True);
+    ModelInfoOut = gr.Textbox(label="Model Info", lines = 10)
+    
+    ModelPath.submit(fn=CheckModel, inputs=ModelPath, outputs=ModelInfoOut)
+    
+    gr.Examples([
+         ["intfloat/multilingual-e5-large"]
+        ,["mixedbread-ai/mxbai-embed-xsmall-v1"]
+        ,["nsense/all-MiniLM-L6-v2-onnx"]
+    
+    ], ModelPath)
+
+demo.launch(
+    server_name = '0.0.0.0'
 )