Model Detail
Ternary-Bonsai-2-27B-mlx-2bit
—Ternary-Bonsai-2-27B-mlx-2bit is a large language model with 27B parameters released by prism-ml. The model is registered under the text-generation pipeline tag on Hugging Face, distributed under the permissive apache-2.0 license.
Ternary-Bonsai-2-27B-mlx-2bit ships with 27B parameters. Total weight footprint is approximately 27.4 GB, which is the relevant figure when planning local-inference VRAM. The apache-2.0 license is permissive, allowing commercial deployment and derivative work without per-seat fees, though attribution requirements still apply.
Downloads of Ternary-Bonsai-2-27B-mlx-2bit have moved +8.5% over the trailing seven days. The trend is mildly positive, consistent with a model that is being picked up incrementally rather than going viral. These numbers are signal, not guarantee — week-over-week download counts on Hugging Face also reflect mirror traffic, CI scrapes, and one-off benchmarking runs.
Ternary-Bonsai-2-27B-mlx-2bit is best fit for general-purpose chat and instruction-following workloads. Treat this as a starting matrix rather than a benchmark verdict — the right deployment usually depends on the specific evaluation suite that mirrors your workload.
Embeddings Are Overpriced? A Frozen 16 MB Vocabulary for Small and Ternary Language Models
arXiv:2610.22428v1 Announce Type: new Abstract: In a small language model the vocabulary is the single largest part. In Gemma 3 270M the embedding table holds 168M of 268M parameters, and because quantisation methods leave the embedding in 16-bit, it is also most of the file once the rest of the mod
Noise-shaped Sampling and Annealing for Scalable On-the-edge Training of Binary and Ternary AI Models
arXiv:2610.22621v1 Announce Type: new Abstract: The training loss of binary and ternary neural networks, expanded as a multilinear polynomial over the discrete weights, is structurally equivalent to the energy of a hierarchical high-order Ising model. We exploit this equivalence to apply Ising solve
RoleQ: Role-Aligned Scales for Converting Language Models to Ternary
arXiv:2610.21808v1 Announce Type: new Abstract: Weight quantization below four bits shrinks memory and bandwidth at a perplexity that looks acceptable, but often at the cost of factual knowledge and task accuracy. Ternary weights have mostly been obtained by training from scratch or long fine-tuning
Lost in the bf16 Cast: Exporting Ternary Language Models Can Revert Most Low-Learning-Rate Code Changes
arXiv:2610.07853v2 Announce Type: replace-cross Abstract: Ternary language models such as BitNet b1.58, Falcon-E and BitCPM are fine-tuned with higher-precision latent weights and deployed as ternary codes produced by an export step that, in the labs' documented pipelines, first casts the latents to
TACO: Ternary Absolute-max Column-wise One-sparse Optimizer for LLM Fine-Tuning
arXiv:2610.02199v2 Announce Type: replace Abstract: Full-parameter fine-tuning of large language models (LLMs) incurs substantial optimizer state memory overhead, limiting the model sizes that fit on modern GPUs. Existing approaches either compress optimizer state, abandon first-order gradients, or
