no_topk

Weight-sparse transformer trained with the procedure from Gao et al. (2025).

Model Details

• Layers: 2
• Model Dimension: 3072
• Context Length: 512
• Head Dimension: 16
• Vocabulary Size: 4096

Sparsity

• Weight Sparsity: True
• Target L0 Fraction: 0.01
• Activation Sparsity: False

Training

• Dataset: jacobcd52/simplestories-tokenized
• Tokenizer: SimpleStories/SimpleStories-1.25M
• Total Tokens: 1,000,000,000

Training Run

• W&B Run: https://wandb.ai/asher577/nodesparse_adv_pretraining/runs/c055yvq9

Usage

import torch
from huggingface_hub import hf_hub_download

# Download model
model_path = hf_hub_download(repo_id="asher577/no_topk", filename="pytorch_model.bin")
config_path = hf_hub_download(repo_id="asher577/no_topk", filename="config.json")

# Load (requires the SparseGPT model class from this repo)
state_dict = torch.load(model_path)