Sakana AIDoc-to-LoRA
Doc-to-LoRA for Gemma 2 2B
This trained hypernetwork creates adapters for google/gemma-2-2b-it. It is an auxiliary model, not a replacement copy of the base language model.
Research release
A document becomes a trained-model update
The hypernetwork compresses document information into a LoRA adapter. The target model can then answer without receiving the full document again.Doc-to-LoRA model method and distinction from Text-to-LoRA
Use the matching target checkpoint
Practical guidance. Load the named target model and its tokenizer. An adapter for another model may have incompatible layer shapes or behavior. Keep the adapter generator's terms separate from the target model's terms.
The Gemma demo has a separate saved checkpoint
The gemma_demo directory contains the checkpoint used by the demonstration. It remains grouped with the Gemma-targeted model.Doc-to-LoRA released hypernetwork files
Specifications & access
| Detail | Value and conditions |
|---|---|
| Status | Research release |
| Access | Downloadable |
| Model identifiers | SakanaAI/doc-to-lora/gemma_2b_d2l |
| Target language model | google/gemma-2-2b-itDoc-to-LoRA for Gemma 2 2B released training arguments |
| Released directory | gemma_2b_d2lDoc-to-LoRA released hypernetwork files |
| Weight terms | No explicit checkpoint license was identified in the reviewed weight repository.Doc-to-LoRA released hypernetwork files |
| Code license | MITDoc-to-LoRA code license |