axial-dense-jepa-cot-mvp

DKNTZMN/axial-dense-jepa-cot-mvp

axial-dense-jepa-cot-mvp is a machine learning model from DKNTZMN. Use it for the machine learning task on the model card, and read the license before you ship it in a product. It is set up for pytorch. The card lists the license as mit.

Pure Axial Dense Transformer + JEPA-CoT. No MoE. No Engram.

Downloads · 30 days

41% of all-time downloads

All-time downloads

Public

Repo size

809 MB

Likes

Public

Hugging Face

Repo makeup

Click a slice to open those files.

.pt809 MB · 100%

At a glance

Library: pytorch
License: mit
Model type: axial-dense-jepa-cot
Access: Public
Created: Sep 24, 2026
Updated: Sep 25, 2026
SHA: 449ff698

Model card

From the Hugging Face model README

Axial Dense + JEPA-CoT

Pure Axial Dense Transformer + JEPA-CoT. No MoE. No Engram.

Hard Urban 512d

Protocol: 4 s history / 8 s future / 2 Hz / 12 agents. Weight: checkpoints/axial_dense_hard_ad_512.pt

Model	ADE	FDE	Miss@2
CV	5.05 m	9.96 m	57.7%
Axial Dense 512d	0.293 m	0.569 m	1.74%

Per-scene (Axial Dense 512d):

Scene	ADE	FDE	Miss@2	n
cruise	0.198	0.509	0.0%	84
lead_hard_brake	0.134	0.237	0.0%	101
cut_in	0.188	0.432	0.0%	86
unprotected_left	0.335	0.404	1.1%	94
lane_change	0.305	0.690	1.1%	87
roundabout	0.224	0.520	1.1%	92
merge_yield	0.235	0.563	1.9%	103
double_park_swerve	0.413	0.675	3.3%	92
pedestrian_dart	0.494	0.896	3.4%	88
multimodal_fork	0.415	0.798	5.3%	94

huggingface-cli download DKNTZMN/axial-dense-jepa-cot-mvp \
  checkpoints/axial_dense_hard_ad_512.pt --local-dir ./axial-dense-hard

MVP easy set (2s/6s, 6 agents)

Model	Val ADE	Val FDE
Flat Dense + JEPA-CoT	3.35 m	6.15 m
Axial Dense + JEPA-CoT	0.183 m	0.503 m

MVP harder set (same shapes)

Model	Val ADE	Val FDE
Flat Dense + JEPA-CoT	2.26 m	4.38 m
Axial Dense + JEPA-CoT	0.328 m	0.752 m

Library: pytorch
Type: axial-dense-jepa-cot
License: mit
Created: Sep 24, 2026
Updated: Sep 25, 2026