config.json

{
    "model_type": "hstu",
    "architectures": [
        "HSTUForCausalLM"
    ],
    "num_hidden_layers": 12,
    "use_random_model": true,
    "max_seq_len": 8832,
    "hidden_size": 4096,
    "vocab_size": 101024,
    "torch_dtype": "float32",
    "hstu_config": {
        "hidden_size": 4096,
        "num_layers": 12,
        "num_attention_heads": 16,
        "head_dim": 256,
        "norm_epsilon": 1e-05,
        "dtype": "float32",
        "residual": true,
        "is_causal": true,
        "has_ffn": false,
        "dropout_ratio": 0
    },
    "task_config": {
        "embedding_configs": [
            {
                "feature_names": ["action_weights"],
                "table_name": "act",
                "vocab_size": 1024,
                "dim": 4096,
                "use_dynamicemb": false
            },
            {
                "feature_names": ["video_id"],
                "table_name": "item",
                "vocab_size": 100000,
                "dim": 4096,
                "use_dynamicemb": true
            }
        ]
    },
    "item_feature_name": "video_id",
    "action_feature_name": "action_weights"
}