|
Download README.md from cachedf/recachedpo-reference-data: direct link, hf CLI and curl.
- Browser
- Download file 1.05 kB
-
https://huggingface.co/datasets/cachedf/recachedpo-reference-data/resolve/main/README.md
- Command line
-
hf download hf://datasets/cachedf/recachedpo-reference-data/README.md
-
curl -L -o README.md https://huggingface.co/datasets/cachedf/recachedpo-reference-data/resolve/main/README.md
1.05 kB
| language: | |
| - en | |
| task_categories: | |
| - text-to-image | |
| pretty_name: ReCacheDPO reference images and conditioning | |
| # ReCacheDPO reference data | |
| Generated reference images and cached text-conditioning tensors for ReCacheDPO scheduling experiments. No backbone model weights are included. | |
| The `taoquan-2.1` bank contains SD3.5 references; `sd3_shift3_train_512_20260918` contains SD3 shift=3 training references. Each bank includes its original `generation_config.json` and `prompts.txt` with the exact generation settings and prompt order. | |
| Files are partitioned into `part-XX` directories. To restore a bank, flatten its part directories, preserving filenames, and place the two metadata files alongside them. `archive_manifest.json` maps original repository-relative paths to remote paths and records byte sizes. Do not mix images or conditioning from different generation configurations. | |
| Upload is incremental: a bank is complete only once all manifest entries exist. The active local shift=1 reference banks are separate from these archived banks. | |