ss_d4096_f0.0039

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

Model Details

• Layers: 4
• Model Dimension: 4096
• Context Length: 512
• Head Dimension: 16
• Vocabulary Size: 4096

Sparsity

• Weight Sparsity: True
• Target L0 Fraction: 0.0039
• Activation Sparsity: True

Training

• Dataset: SimpleStories/SimpleStories
• Tokenizer: SimpleStories/SimpleStories-1.25M
• Total Tokens: 2,000,000,000

Training Run

• W&B Run: https://wandb.ai/training-saes/my_sparsity/runs/6z96ccch

Usage

import torch
from huggingface_hub import hf_hub_download

# Download model
model_path = hf_hub_download(repo_id="jacobcd52/ss_d4096_f0.0039", filename="pytorch_model.bin")
config_path = hf_hub_download(repo_id="jacobcd52/ss_d4096_f0.0039", filename="config.json")

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