2 8 months ago

c256a05cc341 · 113MB
    Metadata
  • general.architecture
    llama
  • general.file_type
    F16
  • llama.attention.head_count
    32
  • llama.attention.head_count_kv
    32
  • llama.attention.layer_norm_rms_epsilon
    1e-05
  • llama.block_count
    4
  • llama.context_length
    4096
  • llama.embedding_length
    1024
  • llama.feed_forward_length
    11008
  • llama.rope.dimension_count
    32
  • llama.rope.freq_base
    10000
  • llama.vocab_size
    1000
  • tokenizer.ggml.merges
    [Ġ Ġ, Ġ ĠĠĠ, ĠĠ ĠĠ, ĠĠĠ Ġ, i n, ...]
  • tokenizer.ggml.model
    gpt2
  • tokenizer.ggml.pre
    llama-bpe
  • tokenizer.ggml.scores
    [0, 1, 2, 3, 4, ...]
  • tokenizer.ggml.token_type
    [1, 1, 1, 1, 1, ...]
  • tokenizer.ggml.tokens
    [!, ", #, $, %, ...]
  • Tensor
  • token_embd.weight
    F16
    [1024, 1000]
  • layers.0.attn_norm.bias
    F32
    [1024]
  • layers.0.attn_norm.weight
    F32
    [1024]
  • layers.0.attn_output.bias
    F32
    [1024]
  • layers.0.attn_output.weight
    F16
    [1024, 1024]
  • layers.0.attn_q.bias
    F32
    [3072]
  • layers.0.attn_q.weight
    F16
    [1024, 3072]
  • layers.0.ffn_down.bias
    F32
    [1024]
  • layers.0.ffn_down.weight
    F16
    [4096, 1024]
  • layers.0.ffn_gate.bias
    F32
    [4096]
  • layers.0.ffn_gate.weight
    F16
    [1024, 4096]
  • layers.0.ffn_norm.bias
    F32
    [1024]
  • layers.0.ffn_norm.weight
    F32
    [1024]
  • layers.1.attn_norm.bias
    F32
    [1024]
  • layers.1.attn_norm.weight
    F32
    [1024]
  • layers.1.attn_output.bias
    F32
    [1024]
  • layers.1.attn_output.weight
    F16
    [1024, 1024]
  • layers.1.attn_q.bias
    F32
    [3072]
  • layers.1.attn_q.weight
    F16
    [1024, 3072]
  • layers.1.ffn_down.bias
    F32
    [1024]
  • layers.1.ffn_down.weight
    F16
    [4096, 1024]
  • layers.1.ffn_gate.bias
    F32
    [4096]
  • layers.1.ffn_gate.weight
    F16
    [1024, 4096]
  • layers.1.ffn_norm.bias
    F32
    [1024]
  • layers.1.ffn_norm.weight
    F32
    [1024]
  • layers.2.attn_norm.bias
    F32
    [1024]
  • layers.2.attn_norm.weight
    F32
    [1024]
  • layers.2.attn_output.bias
    F32
    [1024]
  • layers.2.attn_output.weight
    F16
    [1024, 1024]
  • layers.2.attn_q.bias
    F32
    [3072]
  • layers.2.attn_q.weight
    F16
    [1024, 3072]
  • layers.2.ffn_down.bias
    F32
    [1024]
  • layers.2.ffn_down.weight
    F16
    [4096, 1024]
  • layers.2.ffn_gate.bias
    F32
    [4096]
  • layers.2.ffn_gate.weight
    F16
    [1024, 4096]
  • layers.2.ffn_norm.bias
    F32
    [1024]
  • layers.2.ffn_norm.weight
    F32
    [1024]
  • layers.3.attn_norm.bias
    F32
    [1024]
  • layers.3.attn_norm.weight
    F32
    [1024]
  • layers.3.attn_output.bias
    F32
    [1024]
  • layers.3.attn_output.weight
    F16
    [1024, 1024]
  • layers.3.attn_q.bias
    F32
    [3072]
  • layers.3.attn_q.weight
    F16
    [1024, 3072]
  • layers.3.ffn_down.bias
    F32
    [1024]
  • layers.3.ffn_down.weight
    F16
    [4096, 1024]
  • layers.3.ffn_gate.bias
    F32
    [4096]
  • layers.3.ffn_gate.weight
    F16
    [1024, 4096]
  • layers.3.ffn_norm.bias
    F32
    [1024]
  • layers.3.ffn_norm.weight
    F32
    [1024]
  • output.bias
    F32
    [1000]
  • output.weight
    F16
    [1024, 1000]