Stack
Built on open models and open data.
The base models, test sets, app and hosting behind Nzinga.
Gemma 3
Llama 3.1
Hugging Face
FLORES+
NTREX-128
Expo
Cloudflare
7 pieces · one Nzinga stack
Works with
Another base model
Build an extension pack for another base with the same script and corpus.
Extension pack
Extend a tokenizer, keep every ID.
Train word pieces on the corpus, drop any that fire on English or French, and append the rest.
from tokenizers import Tokenizer
tok = Tokenizer.from_file("gemma3-afx32k.json")
def count(text: str) -> int:
"""Tokens Nzinga needs for a sentence."""
return len(tok.encode(text).ids)
Count your own sentence in the browser.
Paste a sentence into the live comparison. Every figure here is read from a results file.
$ bash scripts/fetch_data.sh
$ export HF_TOKEN=hf_••••••••
$ python scripts/premium_v2.py
✓ NTREX-128 · ✓ FLORES+ · ✓ baselines
● 5 results files · about 4 min