Downloads · 30 days
0
klein4b-qat-ckpt
JoaoZaokk/klein4b-qat-ckpt
klein4b-qat-ckpt is a machine learning model from JoaoZaokk. Use it for the machine learning task on the model card, and read the license before you ship it in a product. The card lists the license as apache-2.0.
Rolling checkpoint of an ongoing quantization-aware-training (QAT) experiment: the transformer of FLUX.2-klein-4B with its 100 block Linear layers trained as ternary (1.58-bit, absmean, group 128 on K) by distillation…
Access
Public
Updated Sep 23, 2026
Repo size
15.6 GB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.pt15.6 GB · 100%
At a glance
- License
- apache-2.0
- Access
- Public
- Created
- Sep 23, 2026
- Updated
- Sep 23, 2026
- SHA
- 49cf1ebc
Base models
Model card
From the Hugging Face model README
klein4b-qat-ckpt — training checkpoint, not a usable model
Rolling checkpoint of an ongoing quantization-aware-training (QAT) experiment: the transformer of FLUX.2-klein-4B with its 100 block Linear layers trained as ternary (1.58-bit, absmean, group 128 on K) by distillation against the BF16 original.
ckpt/ultimo.pt is a raw PyTorch pickle (bf16 master weights in diffusers naming + torchao AdamW8bit
state). It is overwritten every ~25 minutes, is not ternary-packed, and is not meant to be
loaded for inference. Derived from FLUX.2-klein-4B, licensed Apache-2.0.
- License
- apache-2.0
- Created
- Sep 23, 2026
- Updated
- Sep 23, 2026