clef-oQ2.5e

Updated 2026-11-19: Re-quantized with the refined oQe rounding in oMLX 0.8.0, which lowers quantization error at the same size and speed. Please re-download if you have an earlier copy.

This model was quantized using oQ (oMLX v0.8.0) mixed-precision quantization with improved oQe (jundot/omlx#4930) and oQe rounding refinement (jundot/omlx#5292).

A mixed-precision MLX checkpoint of Cloudflare/clef, a 27B decision model post-trained from Qwen/Qwen3.8-27B that turns a state and a schema of typed questions into decisions, returning a probability for every allowed option of every question in a single forward pass. It includes the Qwen3.8-27B language backbone, the vision encoder and the joint schema head.

The joint schema head (joint_head.safetensors, joint_head_config.json) is copied unchanged from the source. oMLX serves this checkpoint through POST /v1/systemone.

The effective average is 3.431 bits/weight for the language backbone and 3.700 bits/weight for the complete checkpoint.

Quantization and Bit Distribution

oQ2.5e allocates bits by measured layer sensitivity, then rounds each group with an importance matrix (128 samples x 512 tokens) and a weighted least-squares refit of its scale and bias.

Storage format Logical weights Share of total Tensor storage Effective bits/weight
Affine 2-bit, group 64 2.333B 8.489% 0.679 GiB 2.50
Affine 3-bit, group 128 20.502B 74.592% 7.757 GiB 3.25
Affine 3-bit, group 64 1.271B 4.626% 0.518 GiB 3.50
Affine 4-bit, group 64 0.005B 0.020% 0.003 GiB 4.50
Affine 5-bit, group 64 2.781B 10.120% 1.781 GiB 5.50
Affine 6-bit, group 64 0.000B 0.001% 0.000 GiB 6.50
Affine 8-bit, group 64 0.000B 0.001% 0.000 GiB 8.50
BF16 0.591B 2.152% 1.102 GiB 16.00
Total 27.485B 100% 11.840 GiB 3.700

Storage Breakdown

Component Logical weights Storage
Language backbone 26.896B 11.535 GB / 10.743 GiB
Vision encoder 0.461B 0.921 GB / 0.858 GiB
Joint schema head 0.128B 0.256 GB / 0.239 GiB
Total 27.485B 12.713 GB / 11.840 GiB

License

Apache-2.0, following the source model. See LICENSE.

Downloads last month
137
Safetensors
Model size
27B params
Tensor type
U32
ยท
BF16
ยท
MLX
Hardware compatibility
Log In to add your hardware

2-bit

Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Model tree for Jundot/clef-oQ2.5e

Base model

Qwen/Qwen3.8-27B
Finetuned
Cloudflare/clef
Quantized
(46)
this model