adammnrr commited on
Commit
9597f8d
·
verified ·
1 Parent(s): 5f36588

Add SleepFM pretrained checkpoints (base + sleep-staging head), CC BY-NC 4.0

Browse files
Files changed (3) hide show
  1. README.md +61 -0
  2. model_base/best.pt +3 -0
  3. model_sleep_staging/best.pth +3 -0
README.md ADDED
@@ -0,0 +1,61 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: cc-by-nc-4.0
3
+ library_name: braindecode
4
+ tags:
5
+ - eeg
6
+ - polysomnography
7
+ - sleep
8
+ - foundation-model
9
+ - braindecode
10
+ ---
11
+
12
+ # SleepFM — pretrained checkpoints
13
+
14
+ Mirror of the official **SleepFM** checkpoints, re-hosted for stable loading from
15
+ [Braindecode](https://github.com/braindecode/braindecode).
16
+
17
+ SleepFM is a multimodal polysomnography (PSG) foundation model introduced in:
18
+
19
+ > R. Thapa et al., *"A multimodal sleep foundation model for disease prediction,"*
20
+ > **Nature Medicine** (2026). https://doi.org/10.1038/s41591-025-04133-4
21
+
22
+ ## Files
23
+
24
+ | File | Description | Used by |
25
+ |------|-------------|---------|
26
+ | `model_base/best.pt` | Pretrained channel-agnostic PSG encoder (backbone) | `SleepFM.load_pretrained_backbone`, `SleepFMStager.load_pretrained_backbone` |
27
+ | `model_sleep_staging/best.pth` | Downstream sleep-staging head | `SleepFMStager.load_pretrained_staging_head` |
28
+
29
+ These are byte-for-byte copies of the upstream artifacts; only the hosting location
30
+ changed. The layout (`model_base/`, `model_sleep_staging/`) mirrors the upstream repo.
31
+
32
+ ## Usage
33
+
34
+ ```python
35
+ import torch
36
+ from braindecode.models import SleepFMStager
37
+
38
+ base = torch.hub.load_state_dict_from_url(
39
+ "https://huggingface.co/braindecode/SleepFM/resolve/main/model_base/best.pt",
40
+ map_location="cpu",
41
+ )
42
+ staging = torch.hub.load_state_dict_from_url(
43
+ "https://huggingface.co/braindecode/SleepFM/resolve/main/model_sleep_staging/best.pth",
44
+ map_location="cpu",
45
+ )
46
+
47
+ model = SleepFMStager(n_chans=4, n_outputs=5, n_times=3840, sfreq=128)
48
+ model.load_pretrained_backbone(base)
49
+ model.load_pretrained_staging_head(staging)
50
+ model.eval()
51
+ ```
52
+
53
+ ## License & attribution
54
+
55
+ - **License: Creative Commons Attribution-NonCommercial 4.0 International (CC BY-NC 4.0).**
56
+ - Copyright (c) 2025 Rahul Thapa.
57
+ - Upstream source: https://github.com/zou-group/sleepfm-clinical
58
+
59
+ These weights are **not** covered by Braindecode's BSD-3 license and inherit the
60
+ upstream **noncommercial** terms. Re-hosted for reproducibility and stable
61
+ availability only; attribution and the CC BY-NC 4.0 restriction are preserved.
model_base/best.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ffc9fc10233ebc4d0aae71abce87070db51b8bac6e4b16a5c1b4401a5f73f799
3
+ size 19444020
model_sleep_staging/best.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a742faf4dd9a1f5ebcccc464894081480217e683974750eb3f10e6c4d1b0fde9
3
+ size 9351915