navthings/ lilstory:latest

yesterday

a lil story model trained on tinystories, 8m parameters

... /
model
0f86a1ef2ec3 · 17MB
    Metadata
  • general.architecture
    llama
  • general.file_type
    F16
  • llama.attention.head_count
    4
  • llama.attention.head_count_kv
    4
  • llama.attention.key_length
    64
  • llama.attention.layer_norm_rms_epsilon
    1e-05
  • llama.attention.value_length
    64
  • llama.block_count
    4
  • llama.context_length
    256
  • llama.embedding_length
    256
  • llama.feed_forward_length
    1024
  • llama.rope.dimension_count
    64
  • llama.rope.freq_base
    10000
  • llama.vocab_size
    8000
  • tokenizer.ggml.bos_token_id
    1
  • tokenizer.ggml.eos_token_id
    2
  • tokenizer.ggml.merges
    [h e, Ġ t, Ġ a, Ġ s, Ġ w, ...]
  • tokenizer.ggml.model
    gpt2
  • tokenizer.ggml.padding_token_id
    3
  • tokenizer.ggml.pre
    gpt-2
  • tokenizer.ggml.token_type
    [3, 3, 3, 3, 1, ...]
  • tokenizer.ggml.tokens
    [[UNK], [BOS], [EOS], [PAD], !, ...]
  • tokenizer.ggml.unknown_token_id
    0
    • Tensor
  • token_embd.weight
    F16
    [256, 8000]
    • blk.0
  • blk.0.attn_k.weight
    F16
    [256, 256]
  • blk.0.attn_norm.weight
    F32
    [256]
  • blk.0.attn_output.weight
    F16
    [256, 256]
  • blk.0.attn_q.weight
    F16
    [256, 256]
  • blk.0.attn_v.weight
    F16
    [256, 256]
  • blk.0.ffn_down.weight
    F16
    [1024, 256]
  • blk.0.ffn_gate.weight
    F16
    [256, 1024]
  • blk.0.ffn_norm.weight
    F32
    [256]
  • blk.0.ffn_up.weight
    F16
    [256, 1024]
    • blk.1
  • blk.1.attn_k.weight
    F16
    [256, 256]
  • blk.1.attn_norm.weight
    F32
    [256]
  • blk.1.attn_output.weight
    F16
    [256, 256]
  • blk.1.attn_q.weight
    F16
    [256, 256]
  • blk.1.attn_v.weight
    F16
    [256, 256]
  • blk.1.ffn_down.weight
    F16
    [1024, 256]
  • blk.1.ffn_gate.weight
    F16
    [256, 1024]
  • blk.1.ffn_norm.weight
    F32
    [256]
  • blk.1.ffn_up.weight
    F16
    [256, 1024]
    • blk.2
  • blk.2.attn_k.weight
    F16
    [256, 256]
  • blk.2.attn_norm.weight
    F32
    [256]
  • blk.2.attn_output.weight
    F16
    [256, 256]
  • blk.2.attn_q.weight
    F16
    [256, 256]
  • blk.2.attn_v.weight
    F16
    [256, 256]
  • blk.2.ffn_down.weight
    F16
    [1024, 256]
  • blk.2.ffn_gate.weight
    F16
    [256, 1024]
  • blk.2.ffn_norm.weight
    F32
    [256]
  • blk.2.ffn_up.weight
    F16
    [256, 1024]
    • blk.3
  • blk.3.attn_k.weight
    F16
    [256, 256]
  • blk.3.attn_norm.weight
    F32
    [256]
  • blk.3.attn_output.weight
    F16
    [256, 256]
  • blk.3.attn_q.weight
    F16
    [256, 256]
  • blk.3.attn_v.weight
    F16
    [256, 256]
  • blk.3.ffn_down.weight
    F16
    [1024, 256]
  • blk.3.ffn_gate.weight
    F16
    [256, 1024]
  • blk.3.ffn_norm.weight
    F32
    [256]
  • blk.3.ffn_up.weight
    F16
    [256, 1024]
  • output.weight
    F16
    [256, 8000]
  • output_norm.weight
    F32
    [256]