Automatic Speech Recognition
Open on Hugging FaceQwen3-ASR 0.6B — Amharic Mixed
Amharic ASR checkpoint trained for broader generalization across speech domains.
Models / Datasets / Open Source
Open-source models and datasets.
Selected speech recognition, language model, and dataset releases. The full archive is on Hugging Face.
View all on Hugging FaceSpeech Recognition
Automatic Speech Recognition
Open on Hugging FaceQwen3-ASR 0.6B — Amharic Gold + Silver
Amharic ASR checkpoint trained on the curated gold and silver dataset mix.
Language Models
Language Model
Open on Hugging FaceGemma 2 27B — Amharic CPT
Gemma 2 27B adapted for Amharic through tokenizer expansion and continued pretraining.
Language Model
Open on Hugging FaceGemma 2 27B — Amharic SFT
Instruction-tuned Amharic model built on the continued-pretraining checkpoint.
Full Archive
The personal site shows selected releases. Hugging Face remains the complete archive.
View all on Hugging Face