Spaces:
Sleeping
Sleeping
| import gradio as gr | |
| import spaces | |
| import torch | |
| from transformers import AutoTokenizer, AutoModelForTokenClassification | |
| model_id = "HaniaEmaan/code-switching-codesaviours-si26-hania" | |
| tokenizer = AutoTokenizer.from_pretrained(model_id) | |
| model = AutoModelForTokenClassification.from_pretrained(model_id) | |
| def classify_text(text): | |
| if not text.strip(): | |
| return "" | |
| words = text.strip().split() | |
| tokenized = tokenizer( | |
| words, | |
| is_split_into_words=True, | |
| truncation=True, | |
| max_length=128, | |
| return_tensors="pt" | |
| ) | |
| with torch.no_grad(): | |
| outputs = model(**tokenized) | |
| probs = torch.softmax(outputs.logits, dim=2)[0] | |
| predictions = torch.argmax(outputs.logits, dim=2)[0] | |
| word_ids = tokenized.word_ids(batch_index=0) | |
| id2label = model.config.id2label | |
| results = [] | |
| seen = set() | |
| for idx, word_idx in enumerate(word_ids): | |
| if word_idx is None or word_idx in seen: | |
| continue | |
| seen.add(word_idx) | |
| label = id2label[predictions[idx].item()] | |
| score = probs[idx][predictions[idx]].item() | |
| results.append(f"Word: '{words[word_idx]}' | Label: {label} | Score: {score:.2f}") | |
| return "\n".join(results) | |
| demo = gr.Interface( | |
| fn=classify_text, | |
| inputs=gr.Textbox(placeholder="Type here... e.g. you don't listen me. me tmse bohat tang hun"), | |
| outputs="textbox", | |
| title="Code-Switched Language Identification (ZeroGPU)", | |
| description="XLM-RoBERTa Token Classification for Roman Urdu and English — word-level tagging." | |
| ) | |
| demo.launch() |