1_7b/xnnpack/config.json

{
  "$schema": "https://huggingface.co/software-mansion/react-native-executorch-spec/resolve/main/config.schema.json",
  "model": "smollm2",
  "family": "smollm",
  "capabilities": [
    "text-generation"
  ],
  "backend": "xnnpack",
  "license": "apache-2.0",
  "variants": [
    {
      "file": "smollm2_1_7b_xnnpack_bf16.pte",
      "precision": "bf16",
      "quantized": false,
      "default": true,
      "methods": {
        "enable_dynamic_shape": true,
        "forward": {
          "inputs": [
            {
              "shape": [
                1,
                {
                  "min": 1,
                  "max": 2048
                }
              ],
              "dtype": "int64"
            },
            {
              "shape": [
                1
              ],
              "dtype": "int64"
            }
          ],
          "outputs": [
            {
              "shape": [
                1,
                49152
              ],
              "dtype": "bfloat16"
            }
          ]
        },
        "get_bos_id": 1,
        "get_eos_ids": [
          2
        ],
        "get_max_context_len": 2048,
        "get_max_seq_len": 2048,
        "get_n_layers": null,
        "get_vocab_size": 49152,
        "use_kv_cache": true,
        "use_sdpa_with_kv_cache": true
      }
    },
    {
      "file": "smollm2_1_7b_xnnpack_8da4w.pte",
      "precision": "8da4w",
      "quantized": true,
      "default": true,
      "methods": {
        "enable_dynamic_shape": true,
        "forward": {
          "inputs": [
            {
              "shape": [
                1,
                {
                  "min": 1,
                  "max": 2048
                }
              ],
              "dtype": "int64"
            },
            {
              "shape": [
                1
              ],
              "dtype": "int64"
            }
          ],
          "outputs": [
            {
              "shape": [
                1,
                49152
              ],
              "dtype": "float32"
            }
          ]
        },
        "get_bos_id": 1,
        "get_eos_ids": [
          2
        ],
        "get_max_context_len": 2048,
        "get_max_seq_len": 2048,
        "get_n_layers": null,
        "get_vocab_size": 49152,
        "use_kv_cache": true,
        "use_sdpa_with_kv_cache": true
      }
    }
  ],
  "size": "1_7b"
}