coml/spidr-mmsulab

SpidR MMS-ulab

SpidR MMS-ulab is a SpidR model pretrained on the Segmented MMS ulab v2 dataset for the DiscoPhon benchmark. It was pretrained using the spidr library.

You can load it with:

from spidr.models import SpidR
from torch.hub import load_state_dict_from_url

state_dict = load_state_dict_from_url("https://huggingface.co/coml/spidr-mmsulab/resolve/main/final.pt")
model = SpidR().eval()
model.load_state_dict(state_dict)

Files:

• config.json: Model configuration.
• final.pt: Model checkpoint.

Intermediate checkpoints

The checkpoints branch contains the training checkpoints, as saved by spidr:

• The step_*.pt files contain the model weights, epoch and step.
• final.pt contains the full training state (model weights, optimizer, scheduler, gradient scaler, epoch and step), to resume training.

List them with hf models ls coml/spidr-mmsulab --revision checkpoints, and load one with:

from pathlib import Path

import torch
from huggingface_hub import snapshot_download
from spidr.models import SpidR, build_model

# Download the checkpoint with its config.json, which build_model looks for in the same folder
folder = snapshot_download(
    "coml/spidr-mmsulab",
    revision="checkpoints",
    allow_patterns=["step_50000.pt", "config.json"],
)
checkpoint = Path(folder) / "step_50000.pt"

# Build the model with the default config and load the weights
model = SpidR().eval()
model.load_state_dict(torch.load(checkpoint)["model"])

# Or build it from the downloaded config.json and load the weights in one call
model = build_model(model_type="spidr", checkpoint=checkpoint)

Citing

Please cite the DiscoPhon paper

@inproceedings{poli2026discophon,
  title     = {{DiscoPhon: Benchmarking the Unsupervised Discovery of Phoneme Inventories With Discrete Speech Units}},
  author    = {Maxime Poli and Manel Khentout and Angelo {Ortiz Tandazo} and Ewan Dunbar and Emmanuel Chemla and Emmanuel Dupoux},
  year      = {2026},
  booktitle = {{Interspeech 2026}},
  pages     = {6664--6669},
  doi       = {10.21437/Interspeech.2026-2791},
  issn      = {2958-1796},
}

along with SpidR.