Create app.py

app.py

@@ -1,6 +1,135 @@
-gradio
-onnxruntime
-huggingface_hub
-transformers
-Pillow
-numpy
+import gradio as gr
+import numpy as np
+import onnxruntime as ort
+from PIL import Image
+from huggingface_hub import hf_hub_download
+from transformers import AutoTokenizer
+
+# ---------------------------------------------------------
+# Download ONNX files directly from Hugging Face model hubs
+# ---------------------------------------------------------
+
+# 1) Multilingual sentiment (DistilBERT)
+multilingual_onnx = hf_hub_download(
+    repo_id="lxyuan/distilbert-base-multilingual-cased-sentiments-student",
+    filename="onnx/model.onnx"
+)
+tokenizer_multilingual = AutoTokenizer.from_pretrained(
+    "lxyuan/distilbert-base-multilingual-cased-sentiments-student"
+)
+session_multilingual = ort.InferenceSession(multilingual_onnx, providers=["CPUExecutionProvider"])
+
+# 2) SDG BERT
+sdgbert_onnx = hf_hub_download(
+    repo_id="sadickam/sdgBERT",
+    filename="onnx/model.onnx"
+)
+tokenizer_sdg = AutoTokenizer.from_pretrained("sadickam/sdgBERT")
+session_sdg = ort.InferenceSession(sdgbert_onnx, providers=["CPUExecutionProvider"])
+
+# 3) German sentiment BERT
+german_onnx = hf_hub_download(
+    repo_id="oliverguhr/german-sentiment-bert",
+    filename="onnx/model.onnx"
+)
+tokenizer_german = AutoTokenizer.from_pretrained("oliverguhr/german-sentiment-bert")
+session_german = ort.InferenceSession(german_onnx, providers=["CPUExecutionProvider"])
+
+# 4) ViT-small image classifier
+vit_onnx = hf_hub_download(
+    repo_id="WinKawaks/vit-small-patch16-224",
+    filename="onnx/model.onnx"
+)
+session_vit = ort.InferenceSession(vit_onnx, providers=["CPUExecutionProvider"])
+
+# Basic preprocessing params (ImageNet)
+IMAGE_SIZE = 224
+MEAN = [0.485, 0.456, 0.406]
+STD  = [0.229, 0.224, 0.225]
+
+
+# ---------------------------------------------------------
+# Inference functions
+# ---------------------------------------------------------
+def softmax(x):
+    e = np.exp(x - np.max(x))
+    return e / e.sum(axis=-1, keepdims=True)
+
+
+def run_multilingual(text):
+    inputs = tokenizer_multilingual(text, return_tensors="np", truncation=True, padding=True)
+    inputs = {k: v.astype(np.int64) for k, v in inputs.items()}
+    logits = session_multilingual.run(None, inputs)[0][0]
+    probs = softmax(logits)
+    labels = tokenizer_multilingual.model.config.id2label
+    return {labels[i]: float(probs[i]) for i in range(len(probs))}
+
+
+def run_sdg(text):
+    inputs = tokenizer_sdg(text, return_tensors="np", truncation=True, padding=True)
+    inputs = {k: v.astype(np.int64) for k, v in inputs.items()}
+    logits = session_sdg.run(None, inputs)[0][0]
+    probs = softmax(logits)
+    labels = tokenizer_sdg.model.config.id2label
+    return {labels[i]: float(probs[i]) for i in range(len(probs))}
+
+
+def run_german(text):
+    inputs = tokenizer_german(text, return_tensors="np", truncation=True, padding=True)
+    inputs = {k: v.astype(np.int64) for k, v in inputs.items()}
+    logits = session_german.run(None, inputs)[0][0]
+    probs = softmax(logits)
+    labels = tokenizer_german.model.config.id2label
+    return {labels[i]: float(probs[i]) for i in range(len(probs))}
+
+
+def preprocess_vit(image):
+    image = image.convert("RGB").resize((IMAGE_SIZE, IMAGE_SIZE))
+    arr = np.array(image).astype(np.float32) / 255.0
+    arr = (arr - MEAN) / STD
+    arr = arr.transpose(2, 0, 1)
+    return arr[np.newaxis, :]
+
+
+def run_vit(image):
+    arr = preprocess_vit(image)
+    input_name = session_vit.get_inputs()[0].name
+    logits = session_vit.run(None, {input_name: arr})[0][0]
+    probs = softmax(logits)
+    top5 = probs.argsort()[::-1][:5]
+    return {f"class_{i}": float(probs[i]) for i in top5}
+
+
+# ---------------------------------------------------------
+# Gradio UI
+# ---------------------------------------------------------
+def inference(model, text, image):
+    if model == "Multilingual Sentiment":
+        return run_multilingual(text)
+    if model == "SDG Classification":
+        return run_sdg(text)
+    if model == "German Sentiment":
+        return run_german(text)
+    if model == "ViT Image Classification":
+        if image is None:
+            return {"error": "Upload an image"}
+        return run_vit(image)
+    return {"error": "Invalid selection"}
+
+
+with gr.Blocks() as demo:
+    gr.Markdown("# 🔍 Multi-Model ONNX Inference (HF-loaded)")
+
+    model_choice = gr.Dropdown(
+        ["Multilingual Sentiment", "SDG Classification", "German Sentiment", "ViT Image Classification"],
+        label="Choose a model"
+    )
+
+    text_in = gr.Textbox(label="Text Input", lines=3)
+    img_in  = gr.Image(label="Image Input", type="pil")
+    output  = gr.JSON(label="Output")
+
+    run_btn = gr.Button("Run Inference")
+    run_btn.click(fn=inference, inputs=[model_choice, text_in, img_in], outputs=[output])
+
+demo.launch()