{
  "format": "flm-browser-v2",
  "name": "FLM WikiText",
  "dataset": "WikiText-2 raw",
  "model_id": "flm-central-1024-wikitext2-s42",
  "neurons": 1024,
  "pools": 128,
  "features": 96,
  "embedding": 96,
  "vocabulary": 4096,
  "bos": 0,
  "eos": 1,
  "arrays": {
    "embedding": {
      "offset": 0,
      "length": 393216,
      "shape": [
        4096,
        96
      ],
      "dtype": "float32"
    },
    "input_weight": {
      "offset": 1572864,
      "length": 98304,
      "shape": [
        1024,
        96
      ],
      "dtype": "float32"
    },
    "input_bias": {
      "offset": 1966080,
      "length": 1024,
      "shape": [
        1024
      ],
      "dtype": "float32"
    },
    "offsets": {
      "offset": 1970176,
      "length": 1025,
      "shape": [
        1025
      ],
      "dtype": "uint32"
    },
    "sources": {
      "offset": 1974276,
      "length": 76130,
      "shape": [
        76130
      ],
      "dtype": "uint32"
    },
    "weights": {
      "offset": 2278796,
      "length": 76130,
      "shape": [
        76130
      ],
      "dtype": "float32"
    },
    "alpha": {
      "offset": 2583316,
      "length": 1024,
      "shape": [
        1024
      ],
      "dtype": "float32"
    },
    "beta": {
      "offset": 2587412,
      "length": 1024,
      "shape": [
        1024
      ],
      "dtype": "float32"
    },
    "pool": {
      "offset": 2591508,
      "length": 1024,
      "shape": [
        1024
      ],
      "dtype": "uint32"
    },
    "pool_sizes": {
      "offset": 2595604,
      "length": 128,
      "shape": [
        128
      ],
      "dtype": "float32"
    },
    "norm_weight": {
      "offset": 2596116,
      "length": 256,
      "shape": [
        256
      ],
      "dtype": "float32"
    },
    "norm_bias": {
      "offset": 2597140,
      "length": 256,
      "shape": [
        256
      ],
      "dtype": "float32"
    },
    "projection_weight": {
      "offset": 2598164,
      "length": 24576,
      "shape": [
        96,
        256
      ],
      "dtype": "float32"
    },
    "projection_bias": {
      "offset": 2696468,
      "length": 96,
      "shape": [
        96
      ],
      "dtype": "float32"
    },
    "readout_bias": {
      "offset": 2696852,
      "length": 4096,
      "shape": [
        4096
      ],
      "dtype": "float32"
    }
  },
  "variant": "flm",
  "checkpoint_step": 6000,
  "checkpoint_sha256": "fb136f3f70e1a0e1d78f116f112ce4141a88180573713043b4ad5ae04add2446",
  "weights_sha256": "9b8fbd448987149292725051832d1729313cced533404f7992fd6440b0d695a9",
  "weights_bytes": 2713236,
  "anatomy_sha256": "92f2d5eef9ba4f880e2ee077cc27022ad8440d865c89816b3b1110f0a6b6dedb",
  "tokenizer_sha256": "ef7f54d89966a08485006e0f64eaf2b0b99217b65ad79f96979f12adbd9ca49d",
  "browser_tokenizer_sha256": "920757426393a61f53017759cb4037345681c99d03ef744cc81daf57a922b3e2",
  "source_graph_sha256": "a2f35369a5ed4981b8ef75c92df285f7df576a020582f119a1372bfbedbaebee",
  "trained_parameters": 600003,
  "retained_edges": 76130,
  "source_neurons": 166700,
  "training": {
    "seed": 42,
    "protocol": {
      "steps": 6000,
      "batch": 16,
      "sequence": 96,
      "warmup": 16,
      "learning_rate": 0.002,
      "final_learning_rate": 0.0002,
      "lr_warmup_updates": 100,
      "weight_decay": 0.01,
      "eval_tokens": 32768,
      "threads": 4,
      "tokenizer_sha256": "ef7f54d89966a08485006e0f64eaf2b0b99217b65ad79f96979f12adbd9ca49d",
      "train_cache_sha256": "dccfcab2728e223fc1620a00afba24feafe6d47373ec7619632d3e5b80b89789",
      "validation_cache_sha256": "a9fe379a68775dc5808338d69d3061c0dfbf2c1e7ae64b8a1f1f55d57e4e6075",
      "graph_sha256": "a2f35369a5ed4981b8ef75c92df285f7df576a020582f119a1372bfbedbaebee"
    },
    "graph_sha256": "a2f35369a5ed4981b8ef75c92df285f7df576a020582f119a1372bfbedbaebee",
    "tokenizer_sha256": "ef7f54d89966a08485006e0f64eaf2b0b99217b65ad79f96979f12adbd9ca49d",
    "data": "data\\processed\\wikitext2-bpe",
    "tokenizer": "data\\tokenizers\\wikitext2-4096\\tokenizer.json",
    "model_version": "flm-lexical-v2",
    "source_commit": "14cedb6caaa14468f1aa55f53f214ae6ff1d36b5",
    "python_torch": "2.8.0+cpu",
    "parameter_card": {
      "config": {
        "neurons": 1024,
        "pools": 128,
        "embedding": 96,
        "variant": "flm",
        "backend": "auto",
        "vocabulary": 4096,
        "tied_readout": true
      },
      "trainable_parameters": 600003,
      "parameter_groups": {
        "edge_log_gain": 76130,
        "alpha_logit": 1024,
        "beta_logit": 1024,
        "recurrent_logit": 1,
        "output_bias": 4096,
        "embedding.weight": 393216,
        "input.weight": 98304,
        "input.bias": 1024,
        "norm.weight": 256,
        "norm.bias": 256,
        "readout.weight": 24576,
        "readout.bias": 96
      },
      "edges": 76130,
      "vocabulary": 4096
    },
    "test_set_used_for_training": false,
    "exposure": {
      "presented_tokens": 9216000,
      "presented_bytes": 32661129,
      "scored_bytes": 27224178
    }
  },
  "norm_epsilon": 1e-05,
  "description": "A compact fly-wired next-token predictor trained from scratch on WikiText-2 raw.",
  "capability": "Experimental text completion; not instruction tuned or a reliable question-answering system.",
  "license": "MIT original code; CC BY 4.0 anatomy; WikiText source attribution and license notes in the dataset card"
}