JEPA-CoT
JEPA-CoT is an experimental chained architecture combining JEPA-style latent prediction with chain-of-thought-inspired multi-stage computation.
License
This repository is licensed under the Apache License 2.0.
Download model weights
The complete retrained Scale-1 checkpoint set contains three independent seeds.
Scale-1 configuration
K=8 路 observation dimension=512 路 latent dimension=128 路 T=24 路 batch=512 路 epochs=40.
Verified reference result
Previously verified seed-0 checkpoint: R虏 0.8729 路 cosine similarity 0.9720 路 MSE 0.05534 路 block metric 0.8441.
Source code
Loading
import torch
ckpt = torch.load("jepa_cot_scale1_seed0.pt", map_location="cpu", weights_only=False)
state_dict = ckpt["state_dict"]
config = ckpt["cfg"]
metrics = ckpt["metrics"]
Each checkpoint contains the model identifier, seed, configuration, complete PyTorch state dict, and evaluation metrics.
- Downloads last month
- 76
Inference Providers NEW
This model isn't deployed by any Inference Provider. 馃檵 Ask for provider support