clark-labs/clark-air-sana-1.6b-gemlite-2bit

Clark Air — Sana 1.6B · GemLite INT2 (ternary)

A self-contained text-to-image pipeline at 3.2 GB: the Sana 1.6B transformer trunk runs ternary {-1, 0, +1} weights on real GemLite INT2 CUDA kernels, the Gemma-2 text encoder is bitsandbytes NF4 (4-bit), and the text encoder + VAE are otherwise the original Sana 1.6B components — all bundled so it loads in one call.

import torch
from diffusers import DiffusionPipeline

pipe = DiffusionPipeline.from_pretrained(
    "clark-labs/clark-air-sana-1.6b-gemlite-2bit", trust_remote_code=True, torch_dtype=torch.bfloat16
).to("cuda")

img = pipe("a corgi astronaut floating in space, photorealistic, 8k",
           num_inference_steps=20, guidance_scale=4.5).images[0]
img.save("out.png")

trust_remote_code=True is required: the transformer is a custom component (ClarkAirSanaGemliteTransformer) that builds the GemLite kernels on load. The 4-bit text encoder loads automatically from its embedded quantization_config.

Requirements

• —NVIDIA CUDA GPU — both GemLite INT2 and bitsandbytes NF4 are CUDA-only.
• —gemlite, bitsandbytes, diffusers >= 0.38, transformers, accelerate, torch, safetensors.

What's inside

The transformer's attention + GLU-FFN trunk is ternary, packed two bits per weight and re-quantized at each layer's largest power-of-two GemLite group (g64 attention / g32 FFN); the small "island" layers (embeddings, norms, modulation, depthwise conv) stay bf16. The transformer is 519 MB vs ~9 GB at fp16. The text encoder + VAE are the original Sana 1.6B weights — only the transformer is quantized (ternary) and the text encoder is NF4-compressed. The NF4 text encoder's 2.1 GB is dominated by the 256k-token embedding, which stays bf16 (quantizing it hurts caption quality).

Standalone packed file (alternative)

The repo also carries clark_air_sana_gemlite_2bit.safetensors + load_clark_air_gemlite.py — a single-file packed transformer with a minimal loader, for when you already have a Sana pipeline and just want to swap in the gemlite trunk. The from_pretrained route above is the self-contained option.