🛠️ Installation

curl -LsSf https://astral.sh/uv/install.sh | sh
./install.sh

🤗 Pre-Trained Models

uv run huggingface-cli login
uv run huggingface-cli download SakanaAI/doc-to-lora --local-dir trained_d2l --include "*/"

🚀 Python API Usage

# caveat: this interface only supports non-batched inputs
for batched inference please see src/ctxtolora/modeling/hypernet.py
import torch
from ctxtolora.modelloading import gettokenizer from ctxtolora.modeling.hypernet import ModulatedPretrainedModel
model loading
checkpointpath = "trainedd2l/gemmademo/checkpoint-80000/pytorchmodel.bin"
statedict = torch.load(checkpointpath, weights_only=False)
model = ModulatedPretrainedModel.fromstatedict(
    statedict, train=False, usesequence_packing=False
)
model.reset()
tokenizer = gettokenizer(model.basemodel.nameorpath)
prepare data
doc = open("data/sakana_wiki.txt", "r").read()
chat = [{"role": "user", "content": "Tell me about Sakana AI."}]
chatids = tokenizer.applychat_template(
    chat,
    addspecialtokens=False,
    returnattentionmask=False,
    addgenerationprompt=True,
    return_tensors="pt",
).to(model.device)
calls after internalization will be influenced by internalized info
model.internalize(doc)
outputs = model.generate(inputids=chatids, maxnewtokens=512) print(tokenizer.decode(outputs[0]))
remove internalized info
model.reset()
without internalized info, the model will halucinate
outputs = model.generate(inputids=chatids, maxnewtokens=512)
print(tokenizer.decode(outputs[0]))

🎮 Interactive Demo

uv run demo/app.py

🧪 Experimental Scripts

| Experiment | Data prep | Training | Evaluation | Notes | | ------------------------------------ | ------------------------------------- | ----------------------------- | ---------------------------- | ----------------------------------------------------------------------------------------------------------------------------------- | | Main experiment | scripts/mainexp/0-downloaddata.sh | scripts/mainexp/1-train.sh | scripts/mainexp/eval/*.sh | Downloading data is fastest; regenerate only if you need fresh synthetic data. Evaluation scripts reproduce the main paper metrics. | | NIAH | scripts/niah/0-gendata.sh | scripts/niah/1-train.sh | scripts/niah/2-eval.sh | Run the scripts in order; data generation only needs to happen once |

🔬 Self-Generated Data Viewer

uv run webui/selfgenviewer.py

📚 Citation

@inproceedings{charakorn2026doctolora,
  title       ={Doc-to-Lo{RA}: Learning to Instantly Internalize Contexts},
  author      ={Rujikorn Charakorn and Edoardo Cetin and Shinnosuke Uesaka and Robert Tjarko Lange},
  booktitle   ={Forty-third International Conference on Machine Learning},
  year        ={2026},
  url         ={https://openreview.net/forum?id=iW1oBBO72S}
}