Demo online !

app.py

@@ -0,0 +1,74 @@
+import gradio as gr
+from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
+import torch
+
+
+# Load model and tokenizer
+MODEL_NAME = "comma-project/normalization-byt5-small"
+
+tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
+model = AutoModelForSeq2SeqLM.from_pretrained(MODEL_NAME)
+
+
+def normalize_text(text: str) -> str:
+    """
+    Normalize input text using ByT5.
+    """
+
+    if not text.strip():
+        return ""
+
+    # Tokenize
+    inputs = tokenizer(
+        text,
+        return_tensors="pt",
+        truncation=True,
+        padding=True,
+        max_length=1024,
+    )
+
+    # Generate
+    with torch.no_grad():
+        outputs = model.generate(
+            **inputs,
+            max_length=1024,
+            num_beams=2,
+            early_stopping=True,
+        )
+
+    # Decode
+    normalized = tokenizer.decode(
+        outputs[0],
+        skip_special_tokens=True,
+    )
+
+    return normalized
+
+
+# Gradio interface
+demo = gr.Interface(
+    fn=normalize_text,
+    inputs=gr.Textbox(
+        label="Input Text",
+        placeholder="Enter text to normalize...",
+        lines=4,
+    ),
+    outputs=gr.Textbox(
+        label="Normalized Text",
+        lines=4,
+    ),
+    title="Text Normalization with ByT5",
+    description="Normalize noisy or non-standard text using the ByT5 model.",
+    theme="soft",
+    examples=[
+        ["Scͥbo uobiᷤᷤ ñ pauli ł donati."],
+        ["""⁊ pitie mlt' lelasce
+P ities li dist. uai a ton peire
+Nelaissier. """, """Uer̃ ab his qͥ ita dissert̃
+q̃ri debet. qͥd ꝑ amorem dei. quidq ꝑ amorẽ
+boni tẽꝑalis ueluit intellig̾e."""]
+    ],
+)
+
+if __name__ == "__main__":
+    demo.launch()

requirements.txt

@@ -0,0 +1,5 @@
+gradio>=4.0.0
+transformers>=4.35.0
+torch>=2.0.0
+sentencepiece>=0.1.99
+accelerate>=0.25.0