Gemma 4 E2B Ndizi Swahili ASR (LiteRT-LM, slim)
On-device bundle (~2.6 GB target): LiteRT shell from litert-community/gemma-4-E2B-it-litert-lm with prefill/decode LLM weights from smutuvi/gemma-4-e2b-sw-asr-ndizi-merged.
- File:
sunflower-gemma4-e2b-litert-lm.litertlm - Adapter (GPU/Colab): smutuvi/gemma-4-e2b-sw-asr-ndizi
- Full export (~5 GB): smutuvi/gemma-4-e2b-sw-asr-ndizi-litert-lm
Inference (Swahili ASR)
Use the same audio-first chat turn and Swahili ASR instruction as training (ndizi_mlops_gemma-4).
Build
Reproduced with python scripts/build_litert_lm_slim.py in ndizi_mlops_gemma-4.
- Downloads last month
- -