Dataset Viewer
Duplicate

The dataset viewer should be available soon. Please retry later.

moracles-weights

LoRA adapters of Qwen/Qwen3-8B used as target models by the moracles activation-oracle project. Each adapter is stored by content: weights/<sha256>/ holds adapter_config.json and adapter_model.safetensors, where <sha256> is the sha256 of the JSON list [[file name, file sha256, bytes], ...] of the folder's files.

index.parquet maps each model_id (the project's canonical fine-tune id, as in the moracles-data models table) to its folder, the files' sha256s and sizes, and the licence basis below.

Licences

The adapters come from three kinds of source; index.parquet's licence_basis gives each adapter's:

  • Diff Interpretation Tuning hidden-topic adapters (model ids dit-q8-*): converted from the weight differences released at diff-interpretation-tuning/loras under the MIT licence; copyright the Diff Interpretation Tuning authors.
  • Adapters trained on public instruction or domain datasets: released under the licence of their training data, named in licence_basis (MIT, Apache-2.0, CC-BY-4.0, ODC-BY, or CC-BY-SA-4.0 for StackExchange-trained adapters, which require attribution to the StackExchange contributors).
  • Adapters trained by the project on generated data: MIT.

Adapters whose training data is under a non-commercial licence, the Llama 3 licence, or a licence we could not establish are not included.

Downloads last month
5