Ivonar Nano

Ivonar Nano is a 349M-parameter language model with natively ternary weights: every large matrix holds only −1, 0 and +1, stored as 2-bit codes with one 16-bit scale per 128 weights. It was trained that way from random initialization on 100 billion tokens, so the ternary weights are what the optimizer produced, not a compressed copy of a larger model. No existing model was used as a starting point, and nothing was quantized after training.

The packed model is 94.6 MB and decodes at about 1,430 tokens per second on an RTX 4060 Ti and about 130 on a Ryzen 9 3900X.

ivonar.com · runtime: github.com/LuisOezdem/ivonar-inference

Benchmarks

Zero-shot multiple choice, full validation splits, one H100. Every option is scored by its summed log probability, the rule the published tables below use, so each column means the same thing. Best per row in bold.

TriLM, FloatLM and QuantLM are the Spectra suite: one lab, one recipe, one 300B-token run, differing only in how the weights are stored. Nano leads TriLM 390M, the only other natively ternary model of its size, by 2.7 points at a third of its training tokens, and it leads QuantLM 390M 4-bit by 1.9, so training in the format beats quantising into it afterwards. It matches BitNet b1.58 700M with half the parameters and stands level with FloatLM 390M. Only LLaMA 700M is ahead, by 1.3 points at twice the parameters.

Format size is parameters × bit width. TriLM and QuantLM ship unpacked in FP16 at 785 MB, BitNet 700M and LLaMA 700M were never released, and Nano's 95 MB is the measured file. Sources: Spectra (arXiv 2407.12327) tables 6 and 7, BitNet b1.58 (arXiv 2402.17764) table 2, raw-accuracy columns.

The supervised pass keeps what pretraining built: under the project's own scoring rule the eight-task mean is 43.8 against 44.0 for the pretrained checkpoint, a 285-question knowledge probe reads 66.0 against 66.3, and validation loss is 1.148. Standard errors run ±1.5 to ±2.6 per task, so differences under 3 points are noise. Both scoring rules sit in nano_sft_base.csv.

Speed and size

On the GPU and on the CPU the kernels compute directly on the packed 2-bit codes, with INT8 activations, integer dot products and one scale multiply per 128 weights.

Quick start

# macOS and Linux
curl -fsSL https://raw.githubusercontent.com/LuisOezdem/ivonar-inference/main/install.sh | sh

# Windows, in PowerShell
irm https://raw.githubusercontent.com/LuisOezdem/ivonar-inference/main/install.ps1 | iex

The installer sets up its own Python and the matching PyTorch build, then opens the chat page; Download Ivonar Nano fetches the model into ~/.ivonar/models. Afterwards ivonar serve starts the chat page and an OpenAI-compatible API on http://127.0.0.1:8000/, and ivonar chat runs in the terminal. An NVIDIA GPU is used when present, otherwise the CPU.

Without a system message the model answers "who are you" with a persona from its training data, since it carries no identity of its own. The runtime ships one by default.

Files

Only the first two are needed to run the model. Keep attribution_bundle.md beside them when you redistribute the weights.

The full training checkpoint, 4.19 GB, sits on the step2100 branch of this repository, for anyone continuing the training:

from huggingface_hub import hf_hub_download
hf_hub_download("Ivonar/ivonar-nano", "checkpoint_step_00002100.pt", revision="step2100")

Architecture

Training

One H100 80GB, 147,456 tokens per optimizer step, FP32 shadow weights and a straight-through estimator. DPO kept the model's knowledge but did not improve its answers at this size, so the supervised checkpoint ships and the preference data never entered the released weights. The supervised checkpoint that produced them, optimizer state included, is on the step2100 branch.

Pretraining mixture by configured weight: FineWeb-Edu 46%, FineWeb-Edu score-2 30%, FineMath 3+ 10%, Cosmopedia 5%, permissively licensed source code 5%, Project Gutenberg 4%. Fine-tuning by realized token share: Magpie-Ultra 62%, OpenCodeInstruct 12%, science answers 7%, summaries 6%, rewrites 4%, persona instructions 3%, OpenThoughts3 3%, system chats 2%, rewriting drills 0.6%, small talk 0.1%. All SmolTalk2 no_think splits apart from OpenCodeInstruct, because a model this size imitates a reasoning trace without being able to carry it.

Training data

Every source carries a permissive license and a pinned revision; exact revisions and required wording are in attribution_bundle.md.

Benchmark datasets were used for evaluation only.

Intended use and limitations

A reference for native ternary training and a small, fast local chat model: 2-bit inference experiments, on-device demos, short answers, rewriting, summarizing a paragraph, simple code.

• Facts are right about two thirds of the time and arithmetic about a third, on the runtime's own test suite, and it invents details fluently, so read what it writes before acting on it.
• English only, 4,096-token context, no tool use, no retrieval.
• It sometimes declines a question it could answer, and repeats itself at low temperature; the runtime's repetition penalty and system message reduce both.

License

Copyright 2026 Luis Oezdem. The weights, tokenizer and this card are released under the Apache License 2.0 (LICENSE). The training data licenses require attribution when the weights are redistributed: keep attribution_bundle.md with the model. One fine-tuning source, 0.19% of all training tokens, carries conversations generated with Llama 3.1: built with Llama.

Citation

@misc{ivonar-nano-2026,
  title  = {Ivonar Nano: a 349M natively ternary language model},
  author = {Oezdem, Luis},
  year   = {2026},
  note   = {Apache-2.0. Trained from scratch on 100B tokens, 2.17 bits per parameter, 94.6 MB packed.}
}