Fg-selective-arabic.bin ((top)) -

def get_fine_grained_embedding(text: str) -> np.ndarray: """ Returns a dense vector that captures root-pattern + selective dialect features. """ # Preprocessing: Keep diacritics (don't strip tashkeel) processed_text = text # Assume input is properly normalized but diacritized

# The .bin file has a custom method: get_fg_vector() # Selective attention automatically prunes irrelevant subword units embedding = model.get_sentence_vector(processed_text) Fg-selective-arabic.bin

| Metric | Fg-selective-arabic.bin | GPT‑4‑Turbo (Arabic) | LLaMA‑2‑13B‑Arabic | MPT‑7B‑Arabic | |--------|---------------------------|---------------------|-------------------|---------------| | | 13.7 | 13.9 | 16.4 | 19.1 | | BLEU (Summarization) | 35.2 | 34.8 | 30.7 | 28.3 | | ROUGE‑L (QA) | 48.5 | 48.1 | 44.0 | 41.6 | | Inference Latency (RTX 4090, 1‑token) | 9 ms | 12 ms | 13 ms | 15 ms | | VRAM Footprint (FP16) | 7.8 GB | 9.2 GB | 9.8 GB | 8.6 GB | | Dialectal Accuracy (Egyptian) | 92 % | 90 % | 84 % | 80 % | def get_fine_grained_embedding(text: str) -> np