ebc-jepa-led-gaussian-hs256

h3lloworld/ebc-jepa-led-gaussian-hs256

ebc-jepa-led-gaussian-hs256 is a machine learning model from h3lloworld. 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 cc-by-nc-4.0.

Frozen V-JEPA 2.1 tokens on LED, tokenised as Gaussian encoder, K = 256 (trained Gaussian event encoder (hstokenlorav2), 256 events per token, LoRA-16 trunk, re-laid by tubelet), scored per event: each event looks up…

Downloads · 30 days

Access

Public

Updated Sep 21, 2026

Repo size

149 MB

Likes

Public

Hugging Face

Repo makeup

Click a slice to open those files.

.pt149 MB · 100%

At a glance

Library: pytorch
License: cc-by-nc-4.0
Access: Public
Created: Sep 17, 2026
Updated: Sep 21, 2026
SHA: 3fbe99c8

Datasets

Model card

From the Hugging Face model README

EBC-JEPA per-event LED denoising: carrier "Gaussian encoder, K = 256"

Frozen V-JEPA 2.1 tokens on LED, tokenised as Gaussian encoder, K = 256 (trained Gaussian event encoder (hs_token_lora_v2), 256 events per token, LoRA-16 trunk, re-laid by tubelet), scored per event: each event looks up the token at its (time bin / 2, patch) and is judged with its own E15 neighbourhood patch. Routes in this repo:

V12 AC: token + E15 patch CNN, no history (v12_ac/).
Integrated: History Reader (kv) conditions the token on earlier chunks, a TemporalViT reads the patch, and Pixel-AC fuses the per-event scores into one decision per pixel and chunk (integrated/<run>/; result.json carries the Pixel-AC metrics with the AC-only metrics under ac). Run-name suffixes: _nokv = reader bypassed, _lb = scalar Local branch on, _vit_base = ViT-B/384 tokens.

All units: 6,000 training chunks, 4 epochs, official 713 chunks at a fixed threshold of 0.5, no selection on test.

unit	route	backbone	kv	LB	SR	NR	DA
`integrated/gaussian_hs256_time`	History Reader + TemporalViT AC + Pixel-AC	ViT-L	on	off	0.8409	0.9010	0.8710
`integrated/gaussian_hs256_time_lb`	TemporalViT AC only (no Pixel-AC)	ViT-L	on	on	0.8174	0.9022	0.8598
`e2e/cos`	end-to-end Gaussian stem + LoRA + V12 AC head, alignment=cos	ViT-L (LoRA)	off	off	0.8422	0.8728	0.8575
`v12_ac`	V12 AC: event's V-JEPA token + E15 patch CNN, no history	ViT-L	off	off	0.8727	0.8391	0.8559
`integrated/gaussian_hs256_time_nokv_lb`	TemporalViT AC only (no Pixel-AC)	ViT-L	off	on	0.7822	0.9157	0.8490
`integrated/gaussian_hs256_time_nokv`	TemporalViT AC only (no Pixel-AC)	ViT-L	off	off	0.7780	0.9169	0.8474

Reference on the same protocol: the collaborator's ViT-L histogram integrated run with Pixel-AC, DA .8714. Code: EBC-JEPA-share (denoising-local-run) and its share-contrib-vitb-downstream worktree.

Library: pytorch
License: cc-by-nc-4.0
Created: Sep 17, 2026
Updated: Sep 21, 2026