Add FIM-PP paper checkpoints (Table 1 fine-tuned, Table 2 zero-shot and fine-tuned)
d642f39 verified metadata
tags:
- time-series
- temporal-point-processes
- hawkes-processes
- scientific-ml
license: mit
FIM-PP paper checkpoints
This repository holds the checkpoints behind the real-data results of
"In-Context Learning of Temporal Point Processes with Foundation Inference Models" (ICLR 2026),
other than the pretrained model itself, which is FIM4Science/fim-pp.
| Subfolder | Paper result | Description |
|---|---|---|
table1-finetuned/{taxi,taobao,stackoverflow,amazon,retweet} |
Table 1 and appendix N = 5/10, FIM-PP (f) | FIM-PP fine-tuned on the CDiff train split of each dataset |
table2-zero-shot |
Table 2, FIM-PP (zs) | Epoch 38 of the pretraining run. FIM4Science/fim-pp is epoch 36 and was used for Table 1 |
table2-finetuned/{taxi,taobao} |
Table 2, FIM-PP (f) | table2-zero-shot fine-tuned on the EasyTPP train split of each dataset |
Each subfolder contains config.json, model-checkpoint.pth and model.safetensors.
Usage
Install OpenFIM and pass <repo>/<subfolder> as the checkpoint:
from fim.models.hawkes import FIMHawkes, FIMHawkesConfig
FIMHawkesConfig.register_for_auto_class()
model = FIMHawkes.load_model("FIM4Science/fim-pp-paper-checkpoints/table1-finetuned/taxi")
The evaluation scripts, configs and a description of the evaluation protocol are in
scripts/hawkes/REPRODUCE_FIM_PP.md.
Citation
@inproceedings{fim_pp,
title={In-Context Learning of Temporal Point Processes with Foundation Inference Models},
author={David Berghaus and Patrick Seifner and Kostadin Cvejoski and Cesar Ojeda and Ramses J. Sanchez},
booktitle={The Fourteenth International Conference on Learning Representations},
year={2026},
url={https://openreview.net/forum?id=h9HwUAODFP}
}