cosmoglint 1.0.2

Transformer-based generative model for galaxies

CosmoGLINT: Cosmological Generative model for Line INtensity mapping with Transformer

Transformer-based models that generate galaxies.

---

Installation

This package is available on PyPI.

pip install cosmoglint

This package requires PyTorch. Please install PyTorch first following https://pytorch.org

Documentation & Source code

GitHub: https://github.com/knmoriwaki/cosmoglint

Basic usage

Load model:

import json
from cosmoglint.model.transformer import transformer_model

with open(f"args.json", "r") as f:
  option = json.load(f, object_hook=lambda d: argparse.Namespace(**d))
  
model = transformer_model(option)

with a json file (example):

{"model_name": "transformer1", "max_length": 50, "d_model": 128, "num_layers": 4, "num_heads": 8, "num_features_out": 100, "num_features_in": 4}

Predict probability:

prob = model(context, x) 

Generate new galaxies:

generated, prob = model.generate(context, x, prob_threshold=1e-5)

Input:

• context: a tensor of shape (N, C_h), containing the properties of halo.
• x: a tensor of shape (N, L, C_g), containing the properties of up to L galaxies for each of the N halos in the batch. Each feature vector of size C_g may include, for example, the halo mass, relative distance to the halo center, radial velocity, and tangential velocity. Input None to generate from scratch.
• prob_threshold (optional): when sampling, the probability below this threshold is set to zero.

Output:

• prob: a tensor of shape (N, L, C_g, d). prob[i,j,k,:] is the probability distribution over d bins for the k-th parameter of the (j+1)-th galaxy in the sequence for the i-th batch element.
• generated: a tensor of shape (N, L, C_g). generated[i,j,k] is the sampled values for each parameter of (j+1)-th galaxy in the sequence for the i-th batch element.

Shape:

• N: Batch size
• L: Sequence length
• C_h: Number of halo properties
• C_g: Number of galaxy properties predicted
• d: Number of bins for the probability distribution of each parameter

Options:

Key	Description
model_name	Name or identifier for the model configuration (default: "transformer1"). Available options are: - "transformer1": halo is prepended to the sequence - "transformer2": halo and galaxy features are embedded together
max_length	Maximum number of galaxies (sequence length) the model will process per halo.
d_model	Dimensionality of the internal feature space (i.e., size of the token embeddings and hidden layers in the transformer).
num_layers	Number of transformer decoder layers stacked in the model.
num_heads	Number of attention heads in the multi-head self-attention layers.
num_features_out	Total number of output bins for the probability distribution.
num_features_in	Number of features per galaxy (e.g., SFR, relative distance, radial/tangential velocity).

Citation

If you use CosmoGLINT in your research, please cite Moriwaki et al. 2025

@ARTICLE{CosmoGLINT,
  title = {CosmoGLINT: Cosmological Generative Model for Line Intensity Mapping with Transformer},
  author = {{Moriwaki}, Kana and {Jun}, Rui Lan and {Osato}, Ken and {Yoshida}, Naoki},
  journal = {arXiv preprints},
  year = 2025,
  month = jun,
  eid = {arXiv:2506.16843},
  doi = {10.48550/arXiv.2506.16843},
  archivePrefix = {arXiv},
  eprint = {2506.16843},
  primaryClass = {astro-ph.CO}
}