irepa-irepa_ckpts-ckpts

Training checkpoints from a representation-alignment study on diffusion transformers, moved off a shared research filesystem to free space. This repository is an archive: the checkpoints are kept so the experiments stay reproducible, not because they are intended as ready-to-use models.

39 files, 76.84 GiB, 26 research line(s). The probed checkpoint is at step 40000.

Training configuration

Read from the checkpoints themselves (args in the payload):

setting	value
model	SiT-B/2
dataset resolution	256
batch size	32
max train steps	400000
encoder (reference only)	pe-vit-g
encoder depth	4
projection coeff	1.0
path type	linear
prediction	v
mixed precision	fp16
seed	0

Checkpoint payload keys: args, ema, model, opt, steps

Runs included

• irepa/irepa_ckpts/A0/seed0/checkpoints
• irepa/irepa_ckpts/A1/seed0/checkpoints
• irepa/irepa_ckpts/A2/seed0/checkpoints
• irepa/irepa_ckpts/B1/seed0/checkpoints
• irepa/irepa_ckpts/B2/seed0/checkpoints
• irepa/irepa_ckpts/B3/seed0/checkpoints
• irepa/irepa_ckpts/fast_A0_20260224_122807/seed0/checkpoints
• irepa/irepa_ckpts/fast_A0_nosampling_20260226_013015/seed0/checkpoints
• irepa/irepa_ckpts/fast_A1_20260224_122807/seed0/checkpoints
• irepa/irepa_ckpts/fast_A1_canonical_noise_std0p0025_20260403_a1_canonical_rerun/seed0/checkpoints
• irepa/irepa_ckpts/fast_A1_canonical_noise_std0p005_20260403_a1_canonical_rerun/seed0/checkpoints
• irepa/irepa_ckpts/fast_A1_canonical_noise_std0p01_20260403_a1_canonical_rerun/seed0/checkpoints
• irepa/irepa_ckpts/fast_A1_canonical_noise_std0p02_20260403_a1_canonical_rerun/seed0/checkpoints
• irepa/irepa_ckpts/fast_A1_canonical_noise_std0p02_20260403_a1n0p02_8g_eff256_retry2/seed0/checkpoints
• irepa/irepa_ckpts/fast_A1_canonical_noise_std0p02_20260420_214210_a1n0p02_4g_eff256_fixgrad/seed0/checkpoints
• irepa/irepa_ckpts/fast_A1_canonical_noise_std0p03_20260403_a1_canonical_rerun/seed0/checkpoints
• irepa/irepa_ckpts/fast_A1_canonical_noise_std0p05_20260403_a1_canonical_rerun/seed0/checkpoints
• irepa/irepa_ckpts/fast_A1_noise_std0p0025_20260320_022425/seed0/checkpoints
• irepa/irepa_ckpts/fast_A1_noise_std0p005_20260320_022425/seed0/checkpoints
• irepa/irepa_ckpts/fast_A1_noise_std0p01_20260320_022425/seed0/checkpoints
• irepa/irepa_ckpts/fast_A1_noise_std0p02_20260320_022425/seed0/checkpoints
• irepa/irepa_ckpts/fast_A1_noise_std0p03_20260320_022425/seed0/checkpoints
• irepa/irepa_ckpts/fast_A1_noise_std0p05_20260320_022425/seed0/checkpoints
• irepa/irepa_ckpts/fast_A1_nosampling_20260226_013015/seed0/checkpoints
• irepa/irepa_ckpts/fast_A2_nosampling_20260225_133652/seed0/checkpoints
• irepa/irepa_ckpts/fast_A2_nosampling_20260225_165224/seed0/checkpoints
• irepa/irepa_ckpts/fast_A2_nosampling_20260226_013015/seed0/checkpoints
• irepa/irepa_ckpts/fast_A2_preenc_20260319_000649/fast_A2_preenc_20260319_000649_seed0/checkpoints
• irepa/irepa_ckpts/fast_A2_preenc_20260320_011812/seed0/checkpoints
• irepa/irepa_ckpts/fast_A2_preenc_20260324_033637/seed0/checkpoints
• irepa/irepa_ckpts/fast_B1_20260224_122807/seed0/checkpoints
• irepa/irepa_ckpts/fast_B1_nosampling_20260226_013015/seed0/checkpoints
• irepa/irepa_ckpts/fast_B2_20260224_122807/seed0/checkpoints
• irepa/irepa_ckpts/fast_B2_nosampling_20260225_165224/seed0/checkpoints
• irepa/irepa_ckpts/fast_B2_nosampling_20260226_013015/seed0/checkpoints
• irepa/irepa_ckpts/fast_B3_nosampling_20260225_143836/seed0/checkpoints
• irepa/irepa_ckpts/fast_B3_nosampling_20260226_013015/seed0/checkpoints
• irepa/irepa_ckpts/sitb2-pe-vit-g-irepa/checkpoints
• irepa/irepa_ckpts/sitb2-pe-vit-g-repa/checkpoints

Paths inside this repository mirror the original paths relative to the project root, so a restore puts every file back where the project's provenance checks expect it.

External pretrained models

These runs were trained against: pe-vit-g.

Those third-party weights are not included here. The checkpoints name the encoder in their config and load it at runtime from a local pretrained-models directory; only parameters trained in this project are published. Verified by inspecting the parameter names in every uploaded checkpoint group: no external encoder tensors are present.

Restore

offload_manifest.tsv lists every file with its sha256, byte size, path in this repository and original absolute path. To restore:

from huggingface_hub import hf_hub_download
p = hf_hub_download("sjrksjek/irepa-irepa_ckpts-ckpts", "<repo path from offload_manifest.tsv>")

Then verify the sha256 against the manifest before use.

Status and licence

Research artifacts, published as-is under CC BY-NC 4.0, with no guarantee of correctness, reproducibility or fitness for any purpose. The numbers derived from these checkpoints live with the project, not in this repository. The non-commercial licence follows the terms of the datasets and pretrained encoders these runs were trained against.