first release

Author: mxmpl

README.md

@@ -1,8 +1,7 @@
 # SpidR: Learning Fast and Stable Linguistic Units for Spoken Language Models Without Supervision
 
-[📜 [Paper](https://arxiv.org)] [📖 [BibTeX](https://github.com/facebookresearch/spidr/tree/main?tab=readme-ov-file#citation)]
-
-This repository contains the checkpoints and training code for the self-supervised speech models from https://arxiv.org.
+This repository contains the checkpoints and training code for the self-supervised speech models
+in the SpidR paper (coming soon!).
 
 ## Overview
 

hubconf.py

@@ -3,7 +3,7 @@
 
 from pathlib import Path
 
-from torch.hub import _add_to_sys_path
+from torch.hub import _add_to_sys_path  # noqa: PLC2701
 
 dependencies = ["torch", "numpy"]
 

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "flit_core.buildapi"
 
 [project]
 name = "spidr"
-version = "1.0.0"
+version = "0.1.0"
 description = "Learning Fast and Stable Linguistic Units for Spoken Language Models Without Supervision"
 readme = "README.md"
 license = "CC-BY-NC-4.0"

src/spidr/models/utils.py

@@ -183,19 +183,19 @@ def state_dict_from_dinosr_fairseq_checkpoint(checkpoint: dict[str, Any]) -> Ord
 def spidr_base(*, pretrained: bool = True, check_hash: bool = False, progress: bool = True) -> SpidR:
     model = SpidR(SpidRConfig())
     if pretrained:
-        url = ""
-        state_dict = load_state_dict_from_url(url, check_hash=check_hash, progress=progress)
-        model.load_state_dict(state_dict)
+        url = "https://dl.fbaipublicfiles.com/shared/devai/assets/models/spidr_base.pt"
+        checkpoint = load_state_dict_from_url(url, check_hash=check_hash, progress=progress, map_location="cpu")
+        model.load_state_dict(checkpoint["model"])
     model.eval()
     return model
 
 
 def dinosr_base_reproduced(*, pretrained: bool = True, check_hash: bool = False, progress: bool = True) -> DinoSR:
     model = DinoSR(DinoSRConfig())
     if pretrained:
-        url = ""
-        state_dict = load_state_dict_from_url(url, check_hash=check_hash, progress=progress)
-        model.load_state_dict(state_dict)
+        url = "https://dl.fbaipublicfiles.com/shared/devai/assets/models/dinosr_base_reproduced.pt"
+        checkpoint = load_state_dict_from_url(url, check_hash=check_hash, progress=progress, map_location="cpu")
+        model.load_state_dict(checkpoint["model"])
     model.eval()
     return model