by SakanaAI
Hypernetworks that update LLMs to remember factual information
# Add to your Claude Code skills
git clone https://github.com/SakanaAI/doc-to-loracurl -LsSf https://astral.sh/uv/install.sh | sh
./install.sh
uv run huggingface-cli login
uv run huggingface-cli download SakanaAI/doc-to-lora --local-dir trained_d2l --include "*/"
# caveat: this interface only supports non-batched inputs
# for batched inference please see `src/ctx_to_lora/modeling/hypernet.py`
import torch
from ctx_to_lora.model_loading import get_tokenizer
from ctx_to_lora.modeling.hypernet import ModulatedPretrainedModel
# model loading
checkpoint_path = "trained_d2l/gemma_demo/checkpoint-80000/pytorch_model.bin"
state_dict = torch.load(checkpoint_path, weights_only=False)
model = ModulatedPretrainedModel.from_state_dict(
state_dict, train=False, use_sequence_packing=False
)
model.reset()
tokenizer = get_tokenizer(model.base_model.name_or_path)
# prepare data
doc = open("data/sakana_wiki.txt", "r").read()
chat = [{"role": "user", "content": "Tell me about Sakana AI."}]
chat_ids = tokenizer.apply_chat_template(
chat,
add_special_tokens=False,
return_attention_mask=False,
add_generation_prompt=True,
return_tensors="pt",
).to(model.device)
# calls after internalization will be influenced by internalized info
model.internalize(doc)
outputs = model.generate(input_ids=chat_ids, max_new_tokens=512)
print(tokenizer.decode(outputs[0]))
# remove internalized info
# model.reset()
# without internalized info, the model will halucinate
# outputs = model.generate(input_ids=chat_ids, max_new_tokens=512)
# print(tokenizer.decode(outputs[0]))
uv run demo/app.py
To run any of the following scripts, use uv run $PATH_TO_SCRIPT from the root of this project.
| Experiment | Data prep | Training | Evaluation | Notes | | ------------------------------------ | ------------------------------------- | ----------------------------- | ---------------------------- | ----------------------------------------------------------------------------------------------------------------------------------- | | | | | | Downloading data is fastest; regenerate only if you need fresh synthetic data. Evaluation scripts reproduce the main paper metrics. | | | | | | Run the scripts in order; data generation only needs to happen once |
No comments yet. Be the first to share your thoughts!
scripts/main_exp/0-download_data.shscripts/main_exp/1-train.shscripts/main_exp/eval/*.shscripts/niah/0-gen_data.shscripts/niah/1-train.shscripts/niah/2-eval.shAfter downloading/generating the data, we can see samples of the data using this script.
uv run webui/self_gen_viewer.py
See more info at webui/SELF_GEN_VIEWER.md.
@techreport{sakana2025doc-to-lora,
title = {{Doc-to-LoRA: Learning to Instantly Internalize Contexts}},
author = {Rujikorn Charakorn and Edoardo Cetin and Shinnosuke Uesaka and Robert Tjarko Lange},
institution = {Sakana AI},
year = {2026},
month = {Febuary},
note = {Technical Report}
}