| --- |
| license: cc-by-nc-4.0 |
| base_model: facebook/nllb-200-distilled-1.3B |
| library_name: peft |
| language: |
| - en |
| - ks |
| tags: |
| - translation |
| - machine-translation |
| - kashmiri |
| - nllb |
| - lora |
| - peft |
| pipeline_tag: translation |
| datasets: |
| - ai4bharat/BPCC |
| metrics: |
| - chrf |
| - bleu |
| --- |
| |
| # kāv: English to Kashmiri Machine Translation |
|
|
| A LoRA adapter fine-tuning `facebook/nllb-200-distilled-1.3B` for English to Kashmiri |
| (Perso-Arabic script, `kas_Arab`) translation. |
|
|
| Full training methodology, data details, decoding configuration, and the |
| post-processing pipeline this adapter is meant to be used with are documented |
| in the companion GitHub repository:**[aym-n/kav](https://github.com/aym-n/kav)** |
|
|
| ## Adapter details |
|
|
| - **Base model:** `facebook/nllb-200-distilled-1.3B` |
| - **Method:** LoRA (PEFT), `r=32`, `alpha=64`, `dropout=0.05` |
| - **Target modules:** `q_proj, k_proj, v_proj, out_proj, fc1, fc2` (attention |
| and feed-forward layers) |
| - **Source language:** `eng_Latn` — **Target language:** `kas_Arab` |
|
|
