{
  "model_type": "sparse-autoencoder",
  "base_model": "google/t5gemma-2-270m-270m",
  "pipeline_tag": "text2text-generation",
  "architecture": "T5 Encoder-Decoder",
  "colab_link": "https://colab.research.google.com/github/mindchain/t5gemma2-sae-all-layers/blob/main/t5gemma2-sae-all-layers.ipynb",
  "tasks": [
    "mechanistic-interpretability",
    "activation-steering",
    "feature-visualization",
    "model-interventions"
  ],
  "languages": ["en"],
  "tags": [
    "sae",
    "sparse-autoencoder",
    "t5gemma",
    "t5gemma2",
    "google-t5gemma",
    "mechanistic-interpretability",
    "activation-steering",
    "steering",
    "feature-visualization",
    "neuronpedia",
    "gemma-scope",
    "sae-lens",
    "transformer-lens",
    "llm-interpretability",
    "explainable-ai",
    "xai",
    "model-steering",
    "feature-engineering",
    "representation-learning",
    "dictionary-learning",
    "layer-interpretation",
    "hidden-state-analysis",
    "nlp",
    "natural-language-processing",
    "text-to-text",
    "language-model",
    "llm",
    "large-language-model",
    "encoder-decoder",
    "transformer"
  ],
  "datasets": [],
  "metrics": {
    "mse": 0.105,
    "cosine_similarity": 0.80,
    "l0_sparsity": 0.33,
    "final_loss": 0.0014
  },
  "training": {
    "epochs": 5,
    "batch_size": 2,
    "learning_rate": 0.0001,
    "optimizer": "AdamW",
    "l1_coefficient": 0.01,
    "d_sae": 4096,
    "d_in": 640
  }
}