PDM-3-HT FLUX.2 VAE Latents for ImageNet-256 Training

PDM-3-HT FLUX.2 VAE latents for the ImageNet-256 training set. This repository contains latent cache shards as a public research artifact for VAE-backend experiments. The dataset was last updated on 2026-05-28 by LAXMAYDAY.

Use Cases

• Training or fine-tuning generative image models based on pre-computed VAE latents.
• Conducting research on latent space properties and manipulations for the ImageNet dataset.
• Benchmarking the performance of different VAE backends in image synthesis pipelines.
• Developing compression or representation learning techniques using high-dimensional latent codes.

Strengths

• Dataset is explicitly intended as a public research artifact for VAE-backend experiments.
• Source data is the well-known ImageNet-1k training set, accessed via ILSVRC.
• Last update timestamp is precisely documented as 2026-05-28 03:23:24.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.

Provenance

Source

ImageNet-1k train via ILSVRC/imagenet-1k.

Collection Method

Images were preprocessed and encoded through a VAE; specific preprocessing details are referenced in the full description.

Freshness

Last updated 2026-05-28 03:23:24.