clark-air-sana-1.6b-1.58bit

Clark Air · Sana 1.6B

A Sana 1.6B text-to-image transformer compressed to ternary (~1.85 bits/weight): 8.6× smaller than FP16, near-FP16 quality.

Samples generated by Clark Air Sana 1.6B at 1.58 bits.

Footprint (measured)

Measured ~1.85 bits/weight → 8.6× smaller (374 MB packed ÷ 3.21 GB FP16).

ℹ️ The transformer/ here is the unpacked compatibility format: it loads in stock diffusers with no custom code, so it is the same 3.21 GB as FP16. The actual compressed artifact is the 374 MB packed safetensors in this repo.

Usage (drop-in diffusers)

import torch
from diffusers import SanaPipeline, SanaTransformer2DModel

transformer = SanaTransformer2DModel.from_pretrained(
    "clark-labs/clark-air-sana-1.6b-1.58bit", subfolder="transformer", torch_dtype=torch.bfloat16)
pipe = SanaPipeline.from_pretrained(
    "Efficient-Large-Model/Sana_1600M_512px_diffusers", transformer=transformer, torch_dtype=torch.bfloat16)
pipe.to("cuda")
pipe("a half empty bottle of red wine", num_inference_steps=20, guidance_scale=4.5).images[0].save("out.png")

About

The transformer weights are quantized to ternary with group-wise scales; a small high-precision tail (~5% of parameters, the conditioning and projection layers) is kept at higher precision.

• Base: Sana 1.6B, 512px

License

Apache-2.0 © Clark Labs, Inc.

Quality vs. bit-width

Same prompts and seed across columns: fp16 base | naive 3-bit PTQ | naive 2-bit PTQ | Clark Air 1.58-bit (trained ternary). Naive post-training quantization collapses by 2 bits, while the trained ternary model uses fewer bits yet tracks fp16.