{
    "parallelism": "CP",
    "attention_layers": [
        {
            "q": "q_scaled",
            "gather_kv": true,
            "gather_q": false,
            "replace": null,
            "polygraphy_class": "AttentionLayerHint"
        }
    ],
    "dist_collectives": {
        "group_size": 0,
        "root": -1,
        "nb_rank": 2,
        "reduce_op": "max",
        "groups": [],
        "polygraphy_class": "DistCollective"
    },
    "inputs": [
        {
            "name": "input",
            "seq_len_idx": 0,
            "rank": 3,
            "polygraphy_class": "ShardTensor"
        }
    ],
    "outputs": [
        {
            "name": "output",
            "seq_len_idx": 0,
            "rank": 3,
            "polygraphy_class": "ShardTensor"
        }
    ],
    "k_seq_len_idx": 3,
    "v_seq_len_idx": 2,
    "kv_rank": 4,
    "polygraphy_class": "ShardHints"
}