PouyaZX4/persian-5gram-kenlm

Persian 5-gram KenLM Language Model

Trained on over 4,078,254 Persian sentences (~85.5 million tokens) with Kneser-Ney smoothing.

Files Available:

• persian_5gram.binary: Full precision trie model (~1.9 GB).
• persian_5gram_8bit.binary: 8-bit quantized model (~900 MB) - Recommended for production.
• persian_5gram_4bit.binary: 4-bit quantized model (~650 MB) - Recommended for mobile/edge ASR.