Create app.py

app.py

@@ -0,0 +1,55 @@
+import os
+import numpy as np
+import torch
+from huggingface_hub import hf_hub_download
+from safetensors.torch import load_file as load_safetensors
+from transformers import AutoTokenizer
+from openvino.runtime import Core
+import gradio as gr
+
+# 1) Model repo on HF Hub
+HF_MODEL = "Kaiyeee/goemotions-multilabel"
+
+# 2) Load tokenizer once
+tokenizer = AutoTokenizer.from_pretrained("roberta-base")
+
+# 3) Load and compile ONNX with OpenVINO on first request
+core = Core()
+# download and cache the .onnx from the model repo
+onnx_path = hf_hub_download(repo_id=HF_MODEL, filename="goemotions_multilabel.onnx")
+ov_model = core.read_model(model=onnx_path)
+compiled = core.compile_model(model=ov_model, device_name="CPU")
+
+# 4) Emotion labels
+emotion_labels = [
+    "admiration","amusement","anger","annoyance","approval","caring","confusion",
+    "curiosity","desire","disappointment","disapproval","disgust","embarrassment",
+    "excitement","fear","gratitude","grief","joy","love","nervousness","optimism",
+    "pride","realization","relief","remorse","sadness","surprise","neutral"
+]
+
+def predict(texts, threshold=0.3):
+    # tokenize to numpy
+    toks = tokenizer(texts, padding="max_length", truncation=True, max_length=128, return_tensors="np")
+    outs = compiled([toks["input_ids"], toks["attention_mask"]])
+    logits = outs[compiled.output(0)]
+    probs  = 1 / (1 + np.exp(-logits))
+    preds  = (probs > threshold).astype(int)
+
+    # map back
+    results = []
+    for i, ps in enumerate(preds):
+        fired = [emotion_labels[j] for j, flag in enumerate(ps) if flag]
+        results.append(", ".join(fired) or "none")
+    return results
+
+# 5) Gradio UI
+with gr.Blocks() as demo:
+    gr.Markdown("# 👀 GoEmotions Multi-Label Demo")
+    inp = gr.Textbox(label="Enter text", lines=3, placeholder="How are you feeling today?")
+    thr = gr.Slider(0.1, 0.9, 0.3, label="Threshold")
+    out = gr.Textbox(label="Predicted emotions")
+    btn = gr.Button("Analyze")
+    btn.click(fn=predict, inputs=[inp, thr], outputs=out)
+
+demo.launch()