Add files using upload-large-folder tool

LICENSE

@@ -0,0 +1,31 @@
+MIT License
+
+Copyright (c) 2025 ByteDance Ltd. and/or its affiliates
+Copyright (c) 2025 Xingjian Leng and Jaskirat Singh
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.
+
+---
+
+This license covers the work of the copyright holders named above. This
+repository also vendors third-party files that are distributed under their own
+terms, and some of those terms are more restrictive than the MIT license --
+notably several files that are licensed for non-commercial use only. Those
+files are listed individually in NOTICE, which you should read before using
+this repository.

MANIFEST.md

@@ -0,0 +1,81 @@
+# GenFirst model release manifest
+
+All paths below are relative to the root of this model release. These are the
+checkpoints used for the paper results, not the later validation runs.
+
+## VAE bundles
+
+- `vae/e2e-f8d4/`: ImageNet f8d4 VAE initialized from SD-VAE.
+- `vae/e2e-f16d32/`: decoder-finetuned f16d32 VAE used by the ImageNet-256
+  class-conditional results.
+- `vae/e2e-f16d32-nodecft/`: pre-decoder-finetune f16d32 VAE used by the
+  ImageNet-512 results.
+- `vae/e2e-flux-vae-sit-imagenet-s2/`: T2I variant A VAE (FLUX.1 init).
+- `vae/e2e-flux-vae-mmdit-l2-t2i-s2/`: T2I variant B VAE (FLUX.1 init).
+- `vae/e2e-vae-flux2-sitxl2-auxkl2p5e8-s3-20k/`: 32-channel f8d32 E2E FLUX.2 VAE
+  (three-stage aspect-ratio-bucketed E2E + decoder finetune).
+
+Every VAE directory contains:
+
+- `<name>.pt`: directly loadable VAE state dict.
+- `<name>-latents-stats.pt`: required latent normalization statistics.
+- `<name>-discriminator-ckpt.pt`: only needed to resume end-to-end training.
+
+In addition, `vae/e2e-vae-flux2-sitxl2-auxkl2p5e8-s3-20k/` includes:
+
+- `e2e-vae-flux2-sitxl2-auxkl2p5e8-s3-20k.safetensors` and `diffusion_pytorch_model.safetensors`:
+  diffusers-compatible safetensors weights.
+- `config.json` and `vae_params.json`: diffusers configuration and explicit per-channel
+  `scaling_factor` and `shift_factor`:
+  - `--vae_pth=vae/e2e-vae-flux2-sitxl2-auxkl2p5e8-s3-20k/e2e-vae-flux2-sitxl2-auxkl2p5e8-s3-20k.safetensors`
+  - `--model.vae.scaling_factor=[0.887221,0.886926,0.887096,0.881914,0.886901,0.887307,0.886685,0.886833,0.886540,0.886646,0.887031,0.887073,0.887381,0.886768,0.887299,0.885674,0.886466,0.886795,0.886979,0.886519,0.886949,0.886949,0.887036,0.886828,0.887344,0.887004,0.886816,0.886799,0.887354,0.887523,0.887012,0.886643]`
+  - `--model.vae.shift_factor=[-0.000623,0.000143,0.000282,0.001599,-0.001093,-0.000477,0.000855,-0.000830,-0.000883,0.001076,-0.000417,0.000200,0.000626,-0.000649,-0.000137,0.001453,-0.000816,-0.000307,-0.000290,0.000662,0.000484,0.000993,0.000574,-0.000139,-0.000914,0.000578,-0.000691,-0.000323,0.000619,-0.000606,-0.000372,0.000437]`
+- `e2e-vae-flux2-sitxl2-auxkl2p5e8-s3-20k-stage2-bn-latents-stats.pt`: Stage-2 70k SiT EMA BatchNorm statistics.
+
+Keep each VAE and its latent statistics together.
+
+## Class-conditional priors
+
+ImageNet-256, paired with `vae/e2e-f16d32/`:
+
+- `class-conditional/imagenet256/eit-imagenet256-weak-20ep.pt`: weak model
+  required for the paper's autoguidance evaluation.
+- `class-conditional/imagenet256/eit-imagenet256-80ep.pt`
+- `class-conditional/imagenet256/eit-imagenet256-480ep.pt`
+- `class-conditional/imagenet256/eit-imagenet256-800ep.pt`
+
+ImageNet-512, paired with `vae/e2e-f16d32-nodecft/`:
+
+- `class-conditional/imagenet512/eit-imagenet512-80ep.pt`
+- `class-conditional/imagenet512/eit-imagenet512-400ep.pt`
+
+These prior checkpoints are directly accepted by
+`Prior_only/generate.py --ckpt` and `Prior_only/scripts/eval_fid.sh` through its
+`CKPT` override. Do not convert or strip them.
+
+## Text-to-image priors
+
+- `text-to-image/variant-a/eit-t2i-variant-a-512.pt`, paired with
+  `vae/e2e-flux-vae-sit-imagenet-s2/`.
+- `text-to-image/variant-b/eit-t2i-variant-b-512.pt`, paired with
+  `vae/e2e-flux-vae-mmdit-l2-t2i-s2/`.
+
+These checkpoints are directly accepted by `Prior_only/infer_t2i.py` and the
+GenEval/DPG generation scripts. Pass the matching VAE, latent statistics and
+`--text-encoder-name openai/clip-vit-large-patch14`; no checkpoint conversion
+is required.
+
+## Checkpoint formats produced by the repository
+
+- `train_prior_only.py` and `train_t2i_frozen.py` save final checkpoints in
+  direct evaluation format.
+- Class-conditional `train_e2e.py` checkpoints can be used directly by the
+  top-level `generate.py`; all `train_e2e.py`, `train_e2e_vae.py`, and
+  `train_decoder_ft.py` checkpoints can be evaluated directly for reconstruction.
+  Before using either checkpoint's VAE with a separate prior, run
+  `Prior_only/scripts/run_prior_only_pipeline.sh extract`, which calls
+  `save_vae_weights.py` to produce a standalone VAE bundle.
+- A downloaded VAE bundle is already extracted and must not be extracted again.
+
+Weights are released under CC BY-NC 4.0. See the repository's
+`MODEL_CARD.md`, `NOTICE`, and documentation for complete terms and commands.

MODEL_CARD.md

@@ -0,0 +1,281 @@
+---
+license: cc-by-nc-4.0
+library_name: pytorch
+pipeline_tag: text-to-image
+tags:
+  - diffusion
+  - latent-diffusion
+  - end-to-end-training
+  - variational-autoencoder
+  - text-to-image
+  - class-conditional-image-generation
+  - sit
+  - mmdit
+---
+
+# GenFirst
+
+Generation Before Reconstruction for Stable End-to-End Latent Generative Modeling.
+
+- Paper: [arXiv:2608.29335](https://arxiv.org/abs/2608.29335)
+- Code: [github.com/ByteDance-Seed/GenFirst](https://github.com/ByteDance-Seed/GenFirst)
+- Project page: [tom-zgt.github.io/GenFirst](https://tom-zgt.github.io/GenFirst/)
+
+These weights are released under **CC BY-NC 4.0: non-commercial research use
+only**, which is stricter than the MIT licence on the code. See
+[Licence and third-party terms](#licence-and-third-party-terms) before use.
+
+## What this is
+
+GenFirst trains a latent VAE jointly with a generative prior without the latent
+collapse that usually makes such end-to-end training unstable. The method is a
+two-phase schedule over the prior and entropy weights: phase 1 lets the prior
+shape the latent geometry, phase 2 drops both weights tenfold so the decoder
+recovers detail. The result is a latent space that a diffusion prior fits well
+while remaining a good autoencoder.
+
+This release covers the **EiT** (diffusion-prior) instantiation. The
+autoregressive **EAR** instantiation from the paper is not released.
+
+Two kinds of artifact are published, and they are used together:
+
+1. **End-to-end VAEs** — the encoder/decoder produced by the end-to-end stage,
+   each shipped with its latent normalization statistics.
+2. **Generative priors** — SiT models for class-conditional generation, and
+   MMDiT models for text-to-image, each trained on a frozen VAE from (1).
+
+## Model details
+
+| | |
+|---|---|
+| Developed by | Guangting Zheng, Yiyuan Zhang, Tao Yang, Yunpeng Chen, Rui Zhu, Jiajun Deng, Yanyong Zhang |
+| Model type | Latent diffusion (flow matching / stochastic interpolant) with a jointly trained VAE |
+| Prior architecture | SiT-XL and MMDiT-XL: 28 blocks, width 1152, 16 attention heads (~675M parameters) |
+| Text encoder (text-to-image) | CLIP-L, frozen, not modified by us |
+| Precision | Weights released in fp32; sampling tested in bf16 |
+| Framework | PyTorch |
+| Inference code | Required — see [Usage](#usage). These are raw `.pt` checkpoints, not `diffusers` pipelines. |
+
+### Class-conditional priors
+
+`SiT-XL` on a frozen end-to-end VAE. Conditioning is a single integer class
+label in `[0, 999]` following the standard 1000-class label ordering; there is
+no text input. Epoch counts are prior-only training epochs.
+
+| Checkpoint | Resolution | VAE | gFID | IS | gFID (guided) | IS (guided) |
+|---|---|---|---|---|---|---|
+| EiT, 80 epochs | 256x256 | `f16d32` | 2.79 | 175.4 | 1.38 | 271.3 |
+| EiT, 480 epochs | 256x256 | `f16d32` | 1.60 | 216.8 | 0.988 | 270.7 |
+| EiT, 800 epochs | 256x256 | `f16d32` | **1.45** | 226.7 | **0.969** | 283.4 |
+| EiT, 80 epochs | 512x512 | `f16d32` | – | – | 1.63 | 278.9 |
+| EiT, 400 epochs | 512x512 | `f16d32` | – | – | **1.23** | 286.1 |
+
+The 512 models reuse the 256 `f16d32` VAE unchanged and frozen; only the prior
+is trained at 512.
+
+### Text-to-image priors
+
+`MMDiT-XL/2` with a frozen CLIP-L text encoder, on a frozen end-to-end FLUX
+VAE, at 512x512. Scores are overall GenEval and DPG-Bench, with no prompt
+rewriting. The two variants differ only in which VAE they were trained on.
+
+| Checkpoint | Resolution | GenEval | DPG-Bench |
+|---|---|---|---|
+| EiT T2I, variant A | 512x512 | **0.8987** | **82.60** |
+| EiT T2I, variant B | 512x512 | 0.8920 | 82.06 |
+
+### End-to-end VAEs
+
+Reconstruction quality of the posterior mean, measured at the resolution the
+VAE is used at.
+
+| VAE | Downsample / channels | Initialized from | rFID | PSNR | SSIM | LPIPS |
+|---|---|---|---|---|---|---|
+| `f8d4` | 8x, 4ch | SD-VAE (`stabilityai/sd-vae-ft-ema`) | **0.57** | 25.13 | 0.71 | 0.14 |
+| `f16d32` | 16x, 32ch | VA-VAE (LightningDiT) | 0.36 | 25.82 | 0.74 | 0.12 |
+| `f8d16` variant A | 8x, 16ch | FLUX.1 VAE | 0.23 | 29.43 | 0.85 | 0.07 |
+| `f8d16` variant B | 8x, 16ch | FLUX.1 VAE | 0.21 | 29.37 | 0.85 | 0.07 |
+| `f8d32` (`e2e-vae-flux2-sitxl2-auxkl2p5e8-s3-20k`, @512) | 8x, 32ch | FLUX.2 VAE | **0.041** | **36.46** | **0.953** | **0.026** |
+
+**A VAE and its latent statistics must be kept together.** The prior was
+trained on latents normalized by those exact statistics; pairing a VAE with the
+wrong statistics file produces noise, not a degraded image. Each VAE directory
+is laid out as:
+
+```
+<name>/
+  <name>.pt                     VAE weights
+  <name>.safetensors            VAE weights in diffusers safetensors format (for FLUX.2 release)
+  <name>-latents-stats.pt       latent normalization (scale + bias)
+  <name>-discriminator-ckpt.pt  discriminator, only needed to resume end-to-end training
+  config.json / vae_params.json diffusers config and explicit scaling_factor / shift_factor (for FLUX.2 release)
+```
+
+For `vae/e2e-vae-flux2-sitxl2-auxkl2p5e8-s3-20k`, the per-channel `scaling_factor` and `shift_factor` parameters are:
+
+```bash
+--vae_pth=vae/e2e-vae-flux2-sitxl2-auxkl2p5e8-s3-20k/e2e-vae-flux2-sitxl2-auxkl2p5e8-s3-20k.safetensors
+--model.vae.scaling_factor=[0.887221,0.886926,0.887096,0.881914,0.886901,0.887307,0.886685,0.886833,0.886540,0.886646,0.887031,0.887073,0.887381,0.886768,0.887299,0.885674,0.886466,0.886795,0.886979,0.886519,0.886949,0.886949,0.887036,0.886828,0.887344,0.887004,0.886816,0.886799,0.887354,0.887523,0.887012,0.886643]
+--model.vae.shift_factor=[-0.000623,0.000143,0.000282,0.001599,-0.001093,-0.000477,0.000855,-0.000830,-0.000883,0.001076,-0.000417,0.000200,0.000626,-0.000649,-0.000137,0.001453,-0.000816,-0.000307,-0.000290,0.000662,0.000484,0.000993,0.000574,-0.000139,-0.000914,0.000578,-0.000691,-0.000323,0.000619,-0.000606,-0.000372,0.000437]
+```
+
+## Usage
+
+These checkpoints need the training repository; there is no `diffusers`
+integration and no `from_pretrained` entry point.
+
+```bash
+git clone https://github.com/ByteDance-Seed/GenFirst
+cd GenFirst
+pip install -r requirements.txt          # install torch first
+```
+
+Class-conditional sampling, writing a small batch of images rather than a full
+evaluation run:
+
+```bash
+cd Prior_only
+python generate.py \
+    --model SiT-XL/1 --resolution 256 \
+    --ckpt /path/to/prior.pt \
+    --vae-arch f16d32 \
+    --vae-ckpt /path/to/e2e-f16d32.pt \
+    --latents-stats /path/to/e2e-f16d32-latents-stats.pt \
+    --num-fid-samples 64 --label-sampling random \
+    --cfg-scale 1.5 --sample-dir samples
+```
+
+`--model` must match the VAE: `SiT-XL/2` for `f8d4`, `SiT-XL/1` for `f16d32`.
+Text-to-image sampling, the guidance settings that reproduce the numbers above,
+and the full evaluation harness are documented in the repository under `docs/`.
+
+## Evaluation
+
+Reported metrics and their protocols:
+
+- **gFID / IS** — 50,000 generated samples scored with
+  [torch-fidelity](https://github.com/LTH14/torch-fidelity) against the
+  reference statistics used by [MAR](https://github.com/LTH14/mar), at the
+  stated resolution. Sampling is SDE with 250 steps. Guided rows use
+  [autoguidance](https://arxiv.org/abs/2406.02507) (pointing away from an
+  early checkpoint of the same run) with CFG on top, except the 512 rows, which
+  use plain CFG.
+- **rFID / PSNR / SSIM / LPIPS** — posterior-mean reconstructions over a
+  50,000-image validation split, following
+  [latent-diffusion](https://github.com/CompVis/latent-diffusion). rFID here is
+  measured against those validation images, not a training-set reference, so it
+  is not comparable to numbers computed the other way.
+- **GenEval / DPG-Bench** — overall scores, Euler sampling with 50 steps and
+  CFG 7.0 on the interval `[0.35, 1.0]`, no prompt rewriting.
+
+Numbers were produced with the versions pinned in `environment.yml`. FID is
+sensitive to reference statistics and to the resizing path, so comparisons
+against other work are only meaningful under the same protocol.
+
+## Training
+
+Training proceeds in two stages. The end-to-end stage trains the VAE jointly
+with a prior under the two-phase GenFirst schedule (global batch 256; 400k
+steps at the high weight, then 200k at the low one). The prior-only stage then
+freezes the VAE, caches its latents, and trains from scratch the prior that is
+actually evaluated (4M steps at 256, 2M at 512). An optional 10k-step
+decoder-only finetune between the stages improves unguided gFID by roughly 0.2.
+Text-to-image follows the same structure, with an `MMDiT-L/2` prior shaping the
+latent during the end-to-end stage and the released `MMDiT-XL/2` trained on the
+frozen result.
+
+Training data consists of publicly available datasets. Individual corpora are
+not enumerated here; each is subject to its own licence and terms of use, which
+govern independently of the licence on these weights.
+
+## Intended use
+
+Intended for research: reproducing the paper, studying end-to-end VAE and prior
+training, latent-space analysis, and use as a baseline or initialization for
+further non-commercial research.
+
+**Out of scope.** Any commercial use. Generating content about real,
+identifiable people. Generating sexual, violent, hateful, or otherwise harmful
+imagery. Any use presenting outputs as real photographs or as factual records.
+Any safety-critical or decision-making pipeline. These models produce synthetic
+images with no factual grounding and must not be treated as a source of truth.
+
+No safety filter, watermarking, or content classifier is included. Anyone
+deploying these weights in a setting where outputs reach other people is
+responsible for adding those controls.
+
+## Limitations and biases
+
+- The class-conditional models generate from a fixed 1000-class label set and
+  cannot follow text. Their outputs reflect the visual distribution and the
+  category taxonomy they were fit to, including the geographic and cultural
+  skew documented for that style of curated image collection.
+- The text-to-image models inherit the biases of both their training data and
+  the frozen CLIP-L text encoder. Expect uneven representation across
+  demographics, professions, and cultures, and expect prompt-following to
+  degrade on long, compositional, or counting-heavy prompts. GenEval and
+  DPG-Bench scores measure prompt alignment on those benchmarks only and do not
+  certify safety or fairness.
+- Text rendering, hands, and fine structure remain unreliable, as with
+  comparable models at this scale.
+- Metrics reported here are distribution-level. A good FID says nothing about
+  the quality or appropriateness of any single sample.
+- The 512 class-conditional models are reported with guidance only; their
+  unguided behaviour is untuned.
+- A VAE paired with mismatched latent statistics fails silently in the sense
+  that it produces output rather than an error. Check the pairing.
+
+## Licence and third-party terms
+
+**Weights: [CC BY-NC 4.0](https://creativecommons.org/licenses/by-nc/4.0/) —
+non-commercial research use only, with attribution.** This is stricter than the
+MIT licence on the code and is the binding constraint on these artifacts.
+
+The code in the GitHub repository is MIT, but the repository as a whole is not
+usable commercially as it stands: the data preprocessing derives from
+[NVIDIA EDM2](https://github.com/NVlabs/edm2) under CC BY-NC-SA 4.0, and
+several vendored encoders and samplers trace back to Meta projects under
+CC BY-NC 4.0. The repository's `NOTICE` lists this component by component.
+
+These weights are also derivative of upstream models, each carrying its own
+terms that continue to apply and that you should verify for your use:
+
+| Component | Upstream |
+|---|---|
+| `f8d4` VAE initialization | [stabilityai/sd-vae-ft-ema](https://huggingface.co/stabilityai/sd-vae-ft-ema) |
+| `f16d32` VAE initialization | VA-VAE, from [LightningDiT](https://github.com/hustvl/LightningDiT) |
+| `f8d16` VAE initialization | FLUX.1 VAE, [black-forest-labs](https://huggingface.co/black-forest-labs) |
+| Text encoder | [OpenAI CLIP](https://github.com/openai/CLIP) |
+| Prior architecture lineage | [REPA](https://github.com/sihyun-yu/REPA), [REPA-E](https://github.com/End2End-Diffusion/REPA-E), and DiT/SiT before them |
+
+Training datasets carry their own terms independently of this licence, and
+those terms may further restrict what you may do with a model trained on them.
+
+## Security
+
+A `.pt` checkpoint is a pickle: loading one executes arbitrary Python. These
+checkpoints store an `args` namespace alongside the tensors, so they cannot be
+loaded with `weights_only=True` and require `weights_only=False` on torch 2.6
+and newer. Load checkpoints only from sources you trust, and verify you are
+downloading from the official repository.
+
+To report a security vulnerability, follow `SECURITY.md` in the code
+repository: contact the ByteDance security team rather than opening a public
+issue.
+
+## Citation
+
+```bibtex
+@article{zheng2026genfirst,
+  title={GenFirst: Generation Before Reconstruction for Stable End-to-End Latent Generative Modeling},
+  author={Zheng, Guangting and Zhang, Yiyuan and Yang, Tao and Chen, Yunpeng and Zhu, Rui and Deng, Jiajun and Zhang, Yanyong},
+  journal={arXiv preprint arXiv:2608.29335},
+  year={2026}
+}
+```
+
+## Contact
+
+Questions through [GitHub Issues](https://github.com/ByteDance-Seed/GenFirst/issues) or
+mail to [zgt@mail.ustc.edu.cn](mailto:zgt@mail.ustc.edu.cn). Security reports
+go to the address in `SECURITY.md`, not to Issues.

README.md

@@ -0,0 +1,281 @@
+---
+license: cc-by-nc-4.0
+library_name: pytorch
+pipeline_tag: text-to-image
+tags:
+  - diffusion
+  - latent-diffusion
+  - end-to-end-training
+  - variational-autoencoder
+  - text-to-image
+  - class-conditional-image-generation
+  - sit
+  - mmdit
+---
+
+# GenFirst
+
+Generation Before Reconstruction for Stable End-to-End Latent Generative Modeling.
+
+- Paper: [arXiv:2608.29335](https://arxiv.org/abs/2608.29335)
+- Code: [github.com/ByteDance-Seed/GenFirst](https://github.com/ByteDance-Seed/GenFirst)
+- Project page: [tom-zgt.github.io/GenFirst](https://tom-zgt.github.io/GenFirst/)
+
+These weights are released under **CC BY-NC 4.0: non-commercial research use
+only**, which is stricter than the MIT licence on the code. See
+[Licence and third-party terms](#licence-and-third-party-terms) before use.
+
+## What this is
+
+GenFirst trains a latent VAE jointly with a generative prior without the latent
+collapse that usually makes such end-to-end training unstable. The method is a
+two-phase schedule over the prior and entropy weights: phase 1 lets the prior
+shape the latent geometry, phase 2 drops both weights tenfold so the decoder
+recovers detail. The result is a latent space that a diffusion prior fits well
+while remaining a good autoencoder.
+
+This release covers the **EiT** (diffusion-prior) instantiation. The
+autoregressive **EAR** instantiation from the paper is not released.
+
+Two kinds of artifact are published, and they are used together:
+
+1. **End-to-end VAEs** — the encoder/decoder produced by the end-to-end stage,
+   each shipped with its latent normalization statistics.
+2. **Generative priors** — SiT models for class-conditional generation, and
+   MMDiT models for text-to-image, each trained on a frozen VAE from (1).
+
+## Model details
+
+| | |
+|---|---|
+| Developed by | Guangting Zheng, Yiyuan Zhang, Tao Yang, Yunpeng Chen, Rui Zhu, Jiajun Deng, Yanyong Zhang |
+| Model type | Latent diffusion (flow matching / stochastic interpolant) with a jointly trained VAE |
+| Prior architecture | SiT-XL and MMDiT-XL: 28 blocks, width 1152, 16 attention heads (~675M parameters) |
+| Text encoder (text-to-image) | CLIP-L, frozen, not modified by us |
+| Precision | Weights released in fp32; sampling tested in bf16 |
+| Framework | PyTorch |
+| Inference code | Required — see [Usage](#usage). These are raw `.pt` checkpoints, not `diffusers` pipelines. |
+
+### Class-conditional priors
+
+`SiT-XL` on a frozen end-to-end VAE. Conditioning is a single integer class
+label in `[0, 999]` following the standard 1000-class label ordering; there is
+no text input. Epoch counts are prior-only training epochs.
+
+| Checkpoint | Resolution | VAE | gFID | IS | gFID (guided) | IS (guided) |
+|---|---|---|---|---|---|---|
+| EiT, 80 epochs | 256x256 | `f16d32` | 2.79 | 175.4 | 1.38 | 271.3 |
+| EiT, 480 epochs | 256x256 | `f16d32` | 1.60 | 216.8 | 0.988 | 270.7 |
+| EiT, 800 epochs | 256x256 | `f16d32` | **1.45** | 226.7 | **0.969** | 283.4 |
+| EiT, 80 epochs | 512x512 | `f16d32` | – | – | 1.63 | 278.9 |
+| EiT, 400 epochs | 512x512 | `f16d32` | – | – | **1.23** | 286.1 |
+
+The 512 models reuse the 256 `f16d32` VAE unchanged and frozen; only the prior
+is trained at 512.
+
+### Text-to-image priors
+
+`MMDiT-XL/2` with a frozen CLIP-L text encoder, on a frozen end-to-end FLUX
+VAE, at 512x512. Scores are overall GenEval and DPG-Bench, with no prompt
+rewriting. The two variants differ only in which VAE they were trained on.
+
+| Checkpoint | Resolution | GenEval | DPG-Bench |
+|---|---|---|---|
+| EiT T2I, variant A | 512x512 | **0.8987** | **82.60** |
+| EiT T2I, variant B | 512x512 | 0.8920 | 82.06 |
+
+### End-to-end VAEs
+
+Reconstruction quality of the posterior mean, measured at the resolution the
+VAE is used at.
+
+| VAE | Downsample / channels | Initialized from | rFID | PSNR | SSIM | LPIPS |
+|---|---|---|---|---|---|---|
+| `f8d4` | 8x, 4ch | SD-VAE (`stabilityai/sd-vae-ft-ema`) | **0.57** | 25.13 | 0.71 | 0.14 |
+| `f16d32` | 16x, 32ch | VA-VAE (LightningDiT) | 0.36 | 25.82 | 0.74 | 0.12 |
+| `f8d16` variant A | 8x, 16ch | FLUX.1 VAE | 0.23 | 29.43 | 0.85 | 0.07 |
+| `f8d16` variant B | 8x, 16ch | FLUX.1 VAE | 0.21 | 29.37 | 0.85 | 0.07 |
+| `f8d32` (`e2e-vae-flux2-sitxl2-auxkl2p5e8-s3-20k`, @512) | 8x, 32ch | FLUX.2 VAE | **0.041** | **36.46** | **0.953** | **0.026** |
+
+**A VAE and its latent statistics must be kept together.** The prior was
+trained on latents normalized by those exact statistics; pairing a VAE with the
+wrong statistics file produces noise, not a degraded image. Each VAE directory
+is laid out as:
+
+```
+<name>/
+  <name>.pt                     VAE weights
+  <name>.safetensors            VAE weights in diffusers safetensors format (for FLUX.2 release)
+  <name>-latents-stats.pt       latent normalization (scale + bias)
+  <name>-discriminator-ckpt.pt  discriminator, only needed to resume end-to-end training
+  config.json / vae_params.json diffusers config and explicit scaling_factor / shift_factor (for FLUX.2 release)
+```
+
+For `vae/e2e-vae-flux2-sitxl2-auxkl2p5e8-s3-20k`, the per-channel `scaling_factor` and `shift_factor` parameters are:
+
+```bash
+--vae_pth=vae/e2e-vae-flux2-sitxl2-auxkl2p5e8-s3-20k/e2e-vae-flux2-sitxl2-auxkl2p5e8-s3-20k.safetensors
+--model.vae.scaling_factor=[0.887221,0.886926,0.887096,0.881914,0.886901,0.887307,0.886685,0.886833,0.886540,0.886646,0.887031,0.887073,0.887381,0.886768,0.887299,0.885674,0.886466,0.886795,0.886979,0.886519,0.886949,0.886949,0.887036,0.886828,0.887344,0.887004,0.886816,0.886799,0.887354,0.887523,0.887012,0.886643]
+--model.vae.shift_factor=[-0.000623,0.000143,0.000282,0.001599,-0.001093,-0.000477,0.000855,-0.000830,-0.000883,0.001076,-0.000417,0.000200,0.000626,-0.000649,-0.000137,0.001453,-0.000816,-0.000307,-0.000290,0.000662,0.000484,0.000993,0.000574,-0.000139,-0.000914,0.000578,-0.000691,-0.000323,0.000619,-0.000606,-0.000372,0.000437]
+```
+
+## Usage
+
+These checkpoints need the training repository; there is no `diffusers`
+integration and no `from_pretrained` entry point.
+
+```bash
+git clone https://github.com/ByteDance-Seed/GenFirst
+cd GenFirst
+pip install -r requirements.txt          # install torch first
+```
+
+Class-conditional sampling, writing a small batch of images rather than a full
+evaluation run:
+
+```bash
+cd Prior_only
+python generate.py \
+    --model SiT-XL/1 --resolution 256 \
+    --ckpt /path/to/prior.pt \
+    --vae-arch f16d32 \
+    --vae-ckpt /path/to/e2e-f16d32.pt \
+    --latents-stats /path/to/e2e-f16d32-latents-stats.pt \
+    --num-fid-samples 64 --label-sampling random \
+    --cfg-scale 1.5 --sample-dir samples
+```
+
+`--model` must match the VAE: `SiT-XL/2` for `f8d4`, `SiT-XL/1` for `f16d32`.
+Text-to-image sampling, the guidance settings that reproduce the numbers above,
+and the full evaluation harness are documented in the repository under `docs/`.
+
+## Evaluation
+
+Reported metrics and their protocols:
+
+- **gFID / IS** — 50,000 generated samples scored with
+  [torch-fidelity](https://github.com/LTH14/torch-fidelity) against the
+  reference statistics used by [MAR](https://github.com/LTH14/mar), at the
+  stated resolution. Sampling is SDE with 250 steps. Guided rows use
+  [autoguidance](https://arxiv.org/abs/2406.02507) (pointing away from an
+  early checkpoint of the same run) with CFG on top, except the 512 rows, which
+  use plain CFG.
+- **rFID / PSNR / SSIM / LPIPS** — posterior-mean reconstructions over a
+  50,000-image validation split, following
+  [latent-diffusion](https://github.com/CompVis/latent-diffusion). rFID here is
+  measured against those validation images, not a training-set reference, so it
+  is not comparable to numbers computed the other way.
+- **GenEval / DPG-Bench** — overall scores, Euler sampling with 50 steps and
+  CFG 7.0 on the interval `[0.35, 1.0]`, no prompt rewriting.
+
+Numbers were produced with the versions pinned in `environment.yml`. FID is
+sensitive to reference statistics and to the resizing path, so comparisons
+against other work are only meaningful under the same protocol.
+
+## Training
+
+Training proceeds in two stages. The end-to-end stage trains the VAE jointly
+with a prior under the two-phase GenFirst schedule (global batch 256; 400k
+steps at the high weight, then 200k at the low one). The prior-only stage then
+freezes the VAE, caches its latents, and trains from scratch the prior that is
+actually evaluated (4M steps at 256, 2M at 512). An optional 10k-step
+decoder-only finetune between the stages improves unguided gFID by roughly 0.2.
+Text-to-image follows the same structure, with an `MMDiT-L/2` prior shaping the
+latent during the end-to-end stage and the released `MMDiT-XL/2` trained on the
+frozen result.
+
+Training data consists of publicly available datasets. Individual corpora are
+not enumerated here; each is subject to its own licence and terms of use, which
+govern independently of the licence on these weights.
+
+## Intended use
+
+Intended for research: reproducing the paper, studying end-to-end VAE and prior
+training, latent-space analysis, and use as a baseline or initialization for
+further non-commercial research.
+
+**Out of scope.** Any commercial use. Generating content about real,
+identifiable people. Generating sexual, violent, hateful, or otherwise harmful
+imagery. Any use presenting outputs as real photographs or as factual records.
+Any safety-critical or decision-making pipeline. These models produce synthetic
+images with no factual grounding and must not be treated as a source of truth.
+
+No safety filter, watermarking, or content classifier is included. Anyone
+deploying these weights in a setting where outputs reach other people is
+responsible for adding those controls.
+
+## Limitations and biases
+
+- The class-conditional models generate from a fixed 1000-class label set and
+  cannot follow text. Their outputs reflect the visual distribution and the
+  category taxonomy they were fit to, including the geographic and cultural
+  skew documented for that style of curated image collection.
+- The text-to-image models inherit the biases of both their training data and
+  the frozen CLIP-L text encoder. Expect uneven representation across
+  demographics, professions, and cultures, and expect prompt-following to
+  degrade on long, compositional, or counting-heavy prompts. GenEval and
+  DPG-Bench scores measure prompt alignment on those benchmarks only and do not
+  certify safety or fairness.
+- Text rendering, hands, and fine structure remain unreliable, as with
+  comparable models at this scale.
+- Metrics reported here are distribution-level. A good FID says nothing about
+  the quality or appropriateness of any single sample.
+- The 512 class-conditional models are reported with guidance only; their
+  unguided behaviour is untuned.
+- A VAE paired with mismatched latent statistics fails silently in the sense
+  that it produces output rather than an error. Check the pairing.
+
+## Licence and third-party terms
+
+**Weights: [CC BY-NC 4.0](https://creativecommons.org/licenses/by-nc/4.0/) —
+non-commercial research use only, with attribution.** This is stricter than the
+MIT licence on the code and is the binding constraint on these artifacts.
+
+The code in the GitHub repository is MIT, but the repository as a whole is not
+usable commercially as it stands: the data preprocessing derives from
+[NVIDIA EDM2](https://github.com/NVlabs/edm2) under CC BY-NC-SA 4.0, and
+several vendored encoders and samplers trace back to Meta projects under
+CC BY-NC 4.0. The repository's `NOTICE` lists this component by component.
+
+These weights are also derivative of upstream models, each carrying its own
+terms that continue to apply and that you should verify for your use:
+
+| Component | Upstream |
+|---|---|
+| `f8d4` VAE initialization | [stabilityai/sd-vae-ft-ema](https://huggingface.co/stabilityai/sd-vae-ft-ema) |
+| `f16d32` VAE initialization | VA-VAE, from [LightningDiT](https://github.com/hustvl/LightningDiT) |
+| `f8d16` VAE initialization | FLUX.1 VAE, [black-forest-labs](https://huggingface.co/black-forest-labs) |
+| Text encoder | [OpenAI CLIP](https://github.com/openai/CLIP) |
+| Prior architecture lineage | [REPA](https://github.com/sihyun-yu/REPA), [REPA-E](https://github.com/End2End-Diffusion/REPA-E), and DiT/SiT before them |
+
+Training datasets carry their own terms independently of this licence, and
+those terms may further restrict what you may do with a model trained on them.
+
+## Security
+
+A `.pt` checkpoint is a pickle: loading one executes arbitrary Python. These
+checkpoints store an `args` namespace alongside the tensors, so they cannot be
+loaded with `weights_only=True` and require `weights_only=False` on torch 2.6
+and newer. Load checkpoints only from sources you trust, and verify you are
+downloading from the official repository.
+
+To report a security vulnerability, follow `SECURITY.md` in the code
+repository: contact the ByteDance security team rather than opening a public
+issue.
+
+## Citation
+
+```bibtex
+@article{zheng2026genfirst,
+  title={GenFirst: Generation Before Reconstruction for Stable End-to-End Latent Generative Modeling},
+  author={Zheng, Guangting and Zhang, Yiyuan and Yang, Tao and Chen, Yunpeng and Zhu, Rui and Deng, Jiajun and Zhang, Yanyong},
+  journal={arXiv preprint arXiv:2608.29335},
+  year={2026}
+}
+```
+
+## Contact
+
+Questions through [GitHub Issues](https://github.com/ByteDance-Seed/GenFirst/issues) or
+mail to [zgt@mail.ustc.edu.cn](mailto:zgt@mail.ustc.edu.cn). Security reports
+go to the address in `SECURITY.md`, not to Issues.

class-conditional/imagenet256/eit-imagenet256-480ep.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8282239e746393f3e9e051083487f37bc6c0ccf9188ea9765352575cfc0c0ebf
+size 10930616719

class-conditional/imagenet256/eit-imagenet256-800ep.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8158e2e9266cd89701ee68ce377cb19d3449a2d37e28926d5edf4567f0678738
+size 10930616911

class-conditional/imagenet256/eit-imagenet256-80ep.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dbf1f5e761ba4dfa3d2db1b87be46676c48570a6c7ddaa280df6dee5568e6d24
+size 10930613455

class-conditional/imagenet256/eit-imagenet256-weak-20ep.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f4eeed90ccc08248024f38e4db69a3abff5534e3d3085171b0a7eac035b4ef91
+size 10930613455

class-conditional/imagenet512/eit-imagenet512-400ep.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7086879e8cab5e15abb5b8a662064bdf7536fd5efa25693899d78475e0eb1de9
+size 10937694799

class-conditional/imagenet512/eit-imagenet512-80ep.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bcebf339cf9d348a2e8c36626d4b372d9599e43afecff4faf4e4df964c65ce77
+size 10937691343

text-to-image/variant-a/eit-t2i-variant-a-512.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3bce2c93a5908faf3b1d2f56cdf03b6b6694be23aed4feffe6d8f1621e15a71c
+size 21368378916

text-to-image/variant-b/eit-t2i-variant-b-512.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b39b8c63d35bde9d17e79e918e010c564ce5d08f3554f2f8fc311e84ceec8cf1
+size 21368378916

vae/e2e-f16d32-nodecft/e2e-f16d32-nodecft-discriminator-ckpt.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c5e34db6865a2f3fdb49a8ef6f2e655c7410d0c163a4d3cc7843ed76c0b74dcf
+size 11077612

vae/e2e-f16d32-nodecft/e2e-f16d32-nodecft-latents-stats.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0dfdac01bcc40948b7d0a811ff2f10be87b6a2ab5b6b3284f1a3cf0dfa8d6544
+size 1954

vae/e2e-f16d32-nodecft/e2e-f16d32-nodecft.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4414ad26489985e950bd63b85e18c94fe974f7626aef33b11f5bc0e204743ebd
+size 279466992

vae/e2e-f16d32/e2e-f16d32-discriminator-ckpt.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:443c9d4655a7360be24f3a2324314d0488d5864ce6ecbe5afcf178a037a3db33
+size 11077768

vae/e2e-f16d32/e2e-f16d32-latents-stats.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4c383ad155dff35e2d97d512ed0338814eb3b43417b1e4994d782655177a459a
+size 1990

vae/e2e-f16d32/e2e-f16d32.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fce3025c7c675d0682f9713081853fc9ce8ea7174ce59547032795a9e9e67890
+size 279469132

vae/e2e-f8d4/e2e-f8d4-discriminator-ckpt.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:43b8c066919fde0da8a7622db17e764e40f32094229c9efc8bb7f8ee899c765e
+size 11077560

vae/e2e-f8d4/e2e-f8d4-latents-stats.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4aca6bd07302b529d3b74de9b123821cc4b08796aab8ce9fbb870515fd2023f8
+size 1686

vae/e2e-f8d4/e2e-f8d4.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ca3b576547a2d0abd32e75789d5ecf3e74bd42fee476dd7145ee75676027982c
+size 334704166

vae/e2e-flux-vae-mmdit-l2-t2i-s2/e2e-flux-vae-mmdit-l2-t2i-s2-discriminator-ckpt.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e3bf0ec6e133b29a71084b60a766ea7285ff4fab935f26b5250a02d3b4fa9c09
+size 11077924

vae/e2e-flux-vae-mmdit-l2-t2i-s2/e2e-flux-vae-mmdit-l2-t2i-s2-latents-stats.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7aea1b648cf7e48beb3c7272058ffa828dd7695fd383aa983b49d9902ba9df00
+size 1898

vae/e2e-flux-vae-mmdit-l2-t2i-s2/e2e-flux-vae-mmdit-l2-t2i-s2.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7cad9a42f3aaff413003d1b1779e102871ba9b5d37d3b28f63ae8083cd095756
+size 335377826

vae/e2e-flux-vae-sit-imagenet-s2/e2e-flux-vae-sit-imagenet-s2-discriminator-ckpt.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0c7a814513967de83b594f0a998f4fae0b887d6fd0fe5abbf8585c4e8278a58b
+size 11077924

vae/e2e-flux-vae-sit-imagenet-s2/e2e-flux-vae-sit-imagenet-s2-latents-stats.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:460326d43ba3c58ec4fe7a1815990a490b687663bc3e756466190a0bad8a3655
+size 1898

vae/e2e-flux-vae-sit-imagenet-s2/e2e-flux-vae-sit-imagenet-s2.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b13483498773f5d4b50f604e1b4cbca6e1bb273b8a354dafe4546e1ecd564ddb
+size 335377826

vae/e2e-vae-flux2-sitxl2-auxkl2p5e8-s3-20k/config.json

@@ -0,0 +1,107 @@
+{
+  "_class_name": "AutoencoderKLFlux2",
+  "_diffusers_version": "0.36.0.dev0",
+  "act_fn": "silu",
+  "batch_norm_eps": 0.0001,
+  "batch_norm_momentum": 0.1,
+  "block_out_channels": [
+    128,
+    256,
+    512,
+    512
+  ],
+  "down_block_types": [
+    "DownEncoderBlock2D",
+    "DownEncoderBlock2D",
+    "DownEncoderBlock2D",
+    "DownEncoderBlock2D"
+  ],
+  "force_upcast": true,
+  "in_channels": 3,
+  "latent_channels": 32,
+  "layers_per_block": 2,
+  "mid_block_add_attention": true,
+  "norm_num_groups": 32,
+  "out_channels": 3,
+  "patch_size": [
+    2,
+    2
+  ],
+  "sample_size": 1024,
+  "scaling_factor": [
+    0.887221,
+    0.886926,
+    0.887096,
+    0.881914,
+    0.886901,
+    0.887307,
+    0.886685,
+    0.886833,
+    0.88654,
+    0.886646,
+    0.887031,
+    0.887073,
+    0.887381,
+    0.886768,
+    0.887299,
+    0.885674,
+    0.886466,
+    0.886795,
+    0.886979,
+    0.886519,
+    0.886949,
+    0.886949,
+    0.887036,
+    0.886828,
+    0.887344,
+    0.887004,
+    0.886816,
+    0.886799,
+    0.887354,
+    0.887523,
+    0.887012,
+    0.886643
+  ],
+  "shift_factor": [
+    -0.000623,
+    0.000143,
+    0.000282,
+    0.001599,
+    -0.001093,
+    -0.000477,
+    0.000855,
+    -0.00083,
+    -0.000883,
+    0.001076,
+    -0.000417,
+    0.0002,
+    0.000626,
+    -0.000649,
+    -0.000137,
+    0.001453,
+    -0.000816,
+    -0.000307,
+    -0.00029,
+    0.000662,
+    0.000484,
+    0.000993,
+    0.000574,
+    -0.000139,
+    -0.000914,
+    0.000578,
+    -0.000691,
+    -0.000323,
+    0.000619,
+    -0.000606,
+    -0.000372,
+    0.000437
+  ],
+  "up_block_types": [
+    "UpDecoderBlock2D",
+    "UpDecoderBlock2D",
+    "UpDecoderBlock2D",
+    "UpDecoderBlock2D"
+  ],
+  "use_post_quant_conv": true,
+  "use_quant_conv": true
+}

vae/e2e-vae-flux2-sitxl2-auxkl2p5e8-s3-20k/diffusion_pytorch_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1c7984e4666d9a67f1b1e8a3b8ad7bd55d7349ae06275b1d8c7befb299dd0905
+size 336213524

vae/e2e-vae-flux2-sitxl2-auxkl2p5e8-s3-20k/e2e-vae-flux2-sitxl2-auxkl2p5e8-s3-20k-discriminator-ckpt.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f313d6c9b2015d9d41121088e96d3de73032296669b04d56bd6be513679b76f5
+size 11078312

vae/e2e-vae-flux2-sitxl2-auxkl2p5e8-s3-20k/e2e-vae-flux2-sitxl2-auxkl2p5e8-s3-20k-latents-stats.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4eddef2d704296f7db1ffda94e41dbf65a5c84934dac16e4a46880b34c5933df
+size 2086

vae/e2e-vae-flux2-sitxl2-auxkl2p5e8-s3-20k/e2e-vae-flux2-sitxl2-auxkl2p5e8-s3-20k-stage2-bn-latents-stats.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9f9756e19a68ea7613119733891e4d5cadecd91aa4dcf4c9141be49157db03cb
+size 2086

vae/e2e-vae-flux2-sitxl2-auxkl2p5e8-s3-20k/e2e-vae-flux2-sitxl2-auxkl2p5e8-s3-20k.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:41718728433592d1c4a9271e5870b3540f7d7ccc60e39b20f2ba7dbc9edc6be2
+size 336289088

vae/e2e-vae-flux2-sitxl2-auxkl2p5e8-s3-20k/e2e-vae-flux2-sitxl2-auxkl2p5e8-s3-20k.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1c7984e4666d9a67f1b1e8a3b8ad7bd55d7349ae06275b1d8c7befb299dd0905
+size 336213524

vae/e2e-vae-flux2-sitxl2-auxkl2p5e8-s3-20k/vae_params.json

@@ -0,0 +1,82 @@
+{
+  "name": "e2e-vae-flux2-sitxl2-auxkl2p5e8-s3-20k",
+  "vae_arch": "flux2",
+  "latent_channels": 32,
+  "downsample_factor": 8,
+  "weights_safetensors": "e2e-vae-flux2-sitxl2-auxkl2p5e8-s3-20k.safetensors",
+  "weights_pt": "e2e-vae-flux2-sitxl2-auxkl2p5e8-s3-20k.pt",
+  "latents_stats_pt": "e2e-vae-flux2-sitxl2-auxkl2p5e8-s3-20k-latents-stats.pt",
+  "scaling_factor": [
+    0.887221,
+    0.886926,
+    0.887096,
+    0.881914,
+    0.886901,
+    0.887307,
+    0.886685,
+    0.886833,
+    0.88654,
+    0.886646,
+    0.887031,
+    0.887073,
+    0.887381,
+    0.886768,
+    0.887299,
+    0.885674,
+    0.886466,
+    0.886795,
+    0.886979,
+    0.886519,
+    0.886949,
+    0.886949,
+    0.887036,
+    0.886828,
+    0.887344,
+    0.887004,
+    0.886816,
+    0.886799,
+    0.887354,
+    0.887523,
+    0.887012,
+    0.886643
+  ],
+  "shift_factor": [
+    -0.000623,
+    0.000143,
+    0.000282,
+    0.001599,
+    -0.001093,
+    -0.000477,
+    0.000855,
+    -0.00083,
+    -0.000883,
+    0.001076,
+    -0.000417,
+    0.0002,
+    0.000626,
+    -0.000649,
+    -0.000137,
+    0.001453,
+    -0.000816,
+    -0.000307,
+    -0.00029,
+    0.000662,
+    0.000484,
+    0.000993,
+    0.000574,
+    -0.000139,
+    -0.000914,
+    0.000578,
+    -0.000691,
+    -0.000323,
+    0.000619,
+    -0.000606,
+    -0.000372,
+    0.000437
+  ],
+  "cli_args": [
+    "--vae_pth=vae/e2e-vae-flux2-sitxl2-auxkl2p5e8-s3-20k/e2e-vae-flux2-sitxl2-auxkl2p5e8-s3-20k.safetensors",
+    "--model.vae.scaling_factor=[0.887221,0.886926,0.887096,0.881914,0.886901,0.887307,0.886685,0.886833,0.886540,0.886646,0.887031,0.887073,0.887381,0.886768,0.887299,0.885674,0.886466,0.886795,0.886979,0.886519,0.886949,0.886949,0.887036,0.886828,0.887344,0.887004,0.886816,0.886799,0.887354,0.887523,0.887012,0.886643]",
+    "--model.vae.shift_factor=[-0.000623,0.000143,0.000282,0.001599,-0.001093,-0.000477,0.000855,-0.000830,-0.000883,0.001076,-0.000417,0.000200,0.000626,-0.000649,-0.000137,0.001453,-0.000816,-0.000307,-0.000290,0.000662,0.000484,0.000993,0.000574,-0.000139,-0.000914,0.000578,-0.000691,-0.000323,0.000619,-0.000606,-0.000372,0.000437]"
+  ]
+}