aoiandroid/nllb-200-distilled-600M-unsloth-GGUF-mirror

Mirror: NLLB-200-Distilled-600M Unsloth GGUF

This is a personal mirror of AlaminI/nllb-200-distilled-600M-unsloth-GGUF on Hugging Face. Authorship, training, and conversion credit belong to the upstream publisher and Meta (base weights). License remains CC BY-NC 4.0 (same as the base model); keep attribution when redistributing.

Snapshot note: When this mirror was created, the upstream Hub repo contained only the shared ggml-vocab-*.gguf files plus this README (no large NLLB-600M weight file in that snapshot). If you need the full quantized translation weights, confirm what is actually published on the upstream repo or use facebook/nllb-200-distilled-600M with Transformers.

---

NLLB-200-Distilled-600M Unsloth GGUF

GGUF-quantized version of facebook/nllb-200-distilled-600M (No Language Left Behind) produced with Unsloth for efficient inference.

Model

• Base: facebook/nllb-200-distilled-600M
• Format: GGUF (e.g. fast_quantized ~4-bit)
• Use case: Multilingual translation (200+ languages), including English ↔ Hausa (eng_Latn ↔ hau_Latn)

How it was created

1. Loaded the seq2seq model with Unsloth using AutoModelForSeq2SeqLM.
2. Saved a merged 16-bit HF-format checkpoint.
3. Converted that checkpoint to GGUF with Unsloth’s save_pretrained_gguf (e.g. fast_quantized).
4. Uploaded the GGUF file(s) to this repo.

How to use

• GGUF runtimes: Use llama.cpp or any GGUF-compatible runtime that supports this architecture. Download the .gguf file(s) from this repo and run inference there.
• Hugging Face Transformers: For 16-bit inference, use the base model facebook/nllb-200-distilled-600M with the standard NLLB pipeline; for translation, set src_lang and tgt_lang (e.g. eng_Latn → hau_Latn).

License

Same as the base model (see facebook/nllb-200-distilled-600M).