{
  "base_model": "Qwen/Qwen2.5-VL-3B-Instruct",
  "layer": 12,
  "variant": "diversified_v1",
  "pretraining": {
    "dataset": "cc3m",
    "n_samples": 1242328,
    "format": "natural_caption_chat"
  },
  "sae": {
    "sparsity": "topk",
    "k": 32,
    "width_mult": 8,
    "d_model": 4096,
    "d_sae": 32768
  },
  "training": {
    "epochs": 50,
    "batch_size": 4096,
    "lr": 0.0003,
    "weight_decay": 0.0
  }
}