Views
No views yet
ParoleLM/socialjax-harvest-frame-aligned-512.ParoleLM/socialjax-harvest-tokenizer-5125,000[B,T,16,22]16 frames2x28851232270.7 MBtokenizer.pt contains the exact model_state_dict used for the published dataset, without optimizer state.
config.json and the relevant source files are included for reproducible loading.1import json
2import torch
3
4from src.architectures.genie.tokenizers.video_tokeniser import VideoTokenizer
5
6config = json.load(open("config.json"))
7model = VideoTokenizer(**config["model"], window=config["data"]["window"])
8checkpoint = torch.load("tokenizer.pt", map_location="cpu")
9model.load_state_dict(checkpoint["model_state_dict"])
10model.eval()