vaiton/
minerva

The first family of Large Language Models pretrained from scratch on Italian!

629 10 months ago

5 Tags
... /
model
eb8154f6f880 · 1.1GB
    Metadata
  • general.architecture
    llama
  • general.file_type
    Q8_0
  • llama.attention.head_count
    16
  • llama.attention.head_count_kv
    4
  • llama.attention.layer_norm_rms_epsilon
    1e-05
  • llama.block_count
    16
  • llama.context_length
    16384
  • llama.embedding_length
    2048
  • llama.feed_forward_length
    7168
  • llama.rope.dimension_count
    128
  • llama.rope.freq_base
    10000
  • llama.vocab_size
    32768
  • tokenizer.ggml.add_bos_token
    true
  • tokenizer.ggml.add_eos_token
    false
  • tokenizer.ggml.bos_token_id
    1
  • tokenizer.ggml.eos_token_id
    2
  • tokenizer.ggml.model
    llama
  • tokenizer.ggml.scores
    [0, 0, 0, 0, 0, ...]
  • tokenizer.ggml.token_type
    [2, 3, 3, 6, 6, ...]
  • tokenizer.ggml.tokens
    [<unk>, <s>, </s>, <0x00>, <0x01>, ...]
  • tokenizer.ggml.unknown_token_id
    0
    • Tensor
  • token_embd.weight
    Q8_0
    [2048, 32768]
    • blk.0
  • blk.0.attn_k.weight
    Q8_0
    [2048, 512]
  • blk.0.attn_norm.weight
    F32
    [2048]
  • blk.0.attn_output.weight
    Q8_0
    [2048, 2048]
  • blk.0.attn_q.weight
    Q8_0
    [2048, 2048]
  • blk.0.attn_v.weight
    Q8_0
    [2048, 512]
  • blk.0.ffn_down.weight
    Q8_0
    [7168, 2048]
  • blk.0.ffn_gate.weight
    Q8_0
    [2048, 7168]
  • blk.0.ffn_norm.weight
    F32
    [2048]
  • blk.0.ffn_up.weight
    Q8_0
    [2048, 7168]
    • blk.1
  • blk.1.attn_k.weight
    Q8_0
    [2048, 512]
  • blk.1.attn_norm.weight
    F32
    [2048]
  • blk.1.attn_output.weight
    Q8_0
    [2048, 2048]
  • blk.1.attn_q.weight
    Q8_0
    [2048, 2048]
  • blk.1.attn_v.weight
    Q8_0
    [2048, 512]
  • blk.1.ffn_down.weight
    Q8_0
    [7168, 2048]
  • blk.1.ffn_gate.weight
    Q8_0
    [2048, 7168]
  • blk.1.ffn_norm.weight
    F32
    [2048]
  • blk.1.ffn_up.weight
    Q8_0
    [2048, 7168]
    • blk.2
  • blk.2.attn_k.weight
    Q8_0
    [2048, 512]
  • blk.2.attn_norm.weight
    F32
    [2048]
  • blk.2.attn_output.weight
    Q8_0
    [2048, 2048]
  • blk.2.attn_q.weight
    Q8_0
    [2048, 2048]
  • blk.2.attn_v.weight
    Q8_0
    [2048, 512]
  • blk.2.ffn_down.weight
    Q8_0
    [7168, 2048]
  • blk.2.ffn_gate.weight
    Q8_0
    [2048, 7168]
  • blk.2.ffn_norm.weight
    F32
    [2048]
  • blk.2.ffn_up.weight
    Q8_0
    [2048, 7168]
    • blk.3
  • blk.3.attn_k.weight
    Q8_0
    [2048, 512]
  • blk.3.attn_norm.weight
    F32
    [2048]
  • blk.3.attn_output.weight
    Q8_0
    [2048, 2048]
  • blk.3.attn_q.weight
    Q8_0
    [2048, 2048]
  • blk.3.attn_v.weight
    Q8_0
    [2048, 512]
  • blk.3.ffn_down.weight
    Q8_0
    [7168, 2048]
  • blk.3.ffn_gate.weight
    Q8_0
    [2048, 7168]
  • blk.3.ffn_norm.weight
    F32
    [2048]
  • blk.3.ffn_up.weight
    Q8_0
    [2048, 7168]
    • blk.4
  • blk.4.attn_k.weight
    Q8_0
    [2048, 512]
  • blk.4.attn_norm.weight
    F32
    [2048]
  • blk.4.attn_output.weight
    Q8_0
    [2048, 2048]
  • blk.4.attn_q.weight
    Q8_0
    [2048, 2048]
  • blk.4.attn_v.weight
    Q8_0
    [2048, 512]
  • blk.4.ffn_down.weight
    Q8_0
    [7168, 2048]
  • blk.4.ffn_gate.weight
    Q8_0
    [2048, 7168]
  • blk.4.ffn_norm.weight
    F32
    [2048]
  • blk.4.ffn_up.weight
    Q8_0
    [2048, 7168]
    • blk.5
  • blk.5.attn_k.weight
    Q8_0
    [2048, 512]
  • blk.5.attn_norm.weight
    F32
    [2048]
  • blk.5.attn_output.weight
    Q8_0
    [2048, 2048]
  • blk.5.attn_q.weight
    Q8_0
    [2048, 2048]
  • blk.5.attn_v.weight
    Q8_0
    [2048, 512]
  • blk.5.ffn_down.weight
    Q8_0
    [7168, 2048]
  • blk.5.ffn_gate.weight
    Q8_0
    [2048, 7168]
  • blk.5.ffn_norm.weight
    F32
    [2048]
  • blk.5.ffn_up.weight
    Q8_0
    [2048, 7168]
    • blk.6
  • blk.6.attn_k.weight
    Q8_0
    [2048, 512]
  • blk.6.attn_norm.weight
    F32
    [2048]
  • blk.6.attn_output.weight
    Q8_0
    [2048, 2048]
  • blk.6.attn_q.weight
    Q8_0
    [2048, 2048]
  • blk.6.attn_v.weight
    Q8_0
    [2048, 512]
  • blk.6.ffn_down.weight
    Q8_0
    [7168, 2048]
  • blk.6.ffn_gate.weight
    Q8_0
    [2048, 7168]
  • blk.6.ffn_norm.weight
    F32
    [2048]
  • blk.6.ffn_up.weight
    Q8_0
    [2048, 7168]
    • blk.7
  • blk.7.attn_k.weight
    Q8_0
    [2048, 512]
  • blk.7.attn_norm.weight
    F32
    [2048]
  • blk.7.attn_output.weight
    Q8_0
    [2048, 2048]
  • blk.7.attn_q.weight
    Q8_0
    [2048, 2048]
  • blk.7.attn_v.weight
    Q8_0
    [2048, 512]
  • blk.7.ffn_down.weight
    Q8_0
    [7168, 2048]
  • blk.7.ffn_gate.weight
    Q8_0
    [2048, 7168]
  • blk.7.ffn_norm.weight
    F32
    [2048]
  • blk.7.ffn_up.weight
    Q8_0
    [2048, 7168]
    • blk.8
  • blk.8.attn_k.weight
    Q8_0
    [2048, 512]
  • blk.8.attn_norm.weight
    F32
    [2048]
  • blk.8.attn_output.weight
    Q8_0
    [2048, 2048]
  • blk.8.attn_q.weight
    Q8_0
    [2048, 2048]
  • blk.8.attn_v.weight
    Q8_0
    [2048, 512]
  • blk.8.ffn_down.weight
    Q8_0
    [7168, 2048]
  • blk.8.ffn_gate.weight
    Q8_0
    [2048, 7168]
  • blk.8.ffn_norm.weight
    F32
    [2048]
  • blk.8.ffn_up.weight
    Q8_0
    [2048, 7168]
    • blk.9
  • blk.9.attn_k.weight
    Q8_0
    [2048, 512]
  • blk.9.attn_norm.weight
    F32
    [2048]
  • blk.9.attn_output.weight
    Q8_0
    [2048, 2048]
  • blk.9.attn_q.weight
    Q8_0
    [2048, 2048]
  • blk.9.attn_v.weight
    Q8_0
    [2048, 512]
  • blk.9.ffn_down.weight
    Q8_0
    [7168, 2048]
  • blk.9.ffn_gate.weight
    Q8_0
    [2048, 7168]
  • blk.9.ffn_norm.weight
    F32
    [2048]
  • blk.9.ffn_up.weight
    Q8_0
    [2048, 7168]
    • blk.10
  • blk.10.attn_k.weight
    Q8_0
    [2048, 512]
  • blk.10.attn_norm.weight
    F32
    [2048]
  • blk.10.attn_output.weight
    Q8_0
    [2048, 2048]
  • blk.10.attn_q.weight
    Q8_0
    [2048, 2048]
  • blk.10.attn_v.weight
    Q8_0
    [2048, 512]
  • blk.10.ffn_down.weight
    Q8_0
    [7168, 2048]
  • blk.10.ffn_gate.weight
    Q8_0
    [2048, 7168]
  • blk.10.ffn_norm.weight
    F32
    [2048]
  • blk.10.ffn_up.weight
    Q8_0
    [2048, 7168]
    • blk.11
  • blk.11.attn_k.weight
    Q8_0
    [2048, 512]
  • blk.11.attn_norm.weight
    F32
    [2048]
  • blk.11.attn_output.weight
    Q8_0
    [2048, 2048]
  • blk.11.attn_q.weight
    Q8_0
    [2048, 2048]
  • blk.11.attn_v.weight
    Q8_0
    [2048, 512]
  • blk.11.ffn_down.weight
    Q8_0
    [7168, 2048]
  • blk.11.ffn_gate.weight
    Q8_0
    [2048, 7168]
  • blk.11.ffn_norm.weight
    F32
    [2048]
  • blk.11.ffn_up.weight
    Q8_0
    [2048, 7168]
    • blk.12
  • blk.12.attn_k.weight
    Q8_0
    [2048, 512]
  • blk.12.attn_norm.weight
    F32
    [2048]
  • blk.12.attn_output.weight
    Q8_0
    [2048, 2048]
  • blk.12.attn_q.weight
    Q8_0
    [2048, 2048]
  • blk.12.attn_v.weight
    Q8_0
    [2048, 512]
  • blk.12.ffn_down.weight
    Q8_0
    [7168, 2048]
  • blk.12.ffn_gate.weight
    Q8_0
    [2048, 7168]
  • blk.12.ffn_norm.weight
    F32
    [2048]
  • blk.12.ffn_up.weight
    Q8_0
    [2048, 7168]
    • blk.13
  • blk.13.attn_k.weight
    Q8_0
    [2048, 512]
  • blk.13.attn_norm.weight
    F32
    [2048]
  • blk.13.attn_output.weight
    Q8_0
    [2048, 2048]
  • blk.13.attn_q.weight
    Q8_0
    [2048, 2048]
  • blk.13.attn_v.weight
    Q8_0
    [2048, 512]
  • blk.13.ffn_down.weight
    Q8_0
    [7168, 2048]
  • blk.13.ffn_gate.weight
    Q8_0
    [2048, 7168]
  • blk.13.ffn_norm.weight
    F32
    [2048]
  • blk.13.ffn_up.weight
    Q8_0
    [2048, 7168]
    • blk.14
  • blk.14.attn_k.weight
    Q8_0
    [2048, 512]
  • blk.14.attn_norm.weight
    F32
    [2048]
  • blk.14.attn_output.weight
    Q8_0
    [2048, 2048]
  • blk.14.attn_q.weight
    Q8_0
    [2048, 2048]
  • blk.14.attn_v.weight
    Q8_0
    [2048, 512]
  • blk.14.ffn_down.weight
    Q8_0
    [7168, 2048]
  • blk.14.ffn_gate.weight
    Q8_0
    [2048, 7168]
  • blk.14.ffn_norm.weight
    F32
    [2048]
  • blk.14.ffn_up.weight
    Q8_0
    [2048, 7168]
    • blk.15
  • blk.15.attn_k.weight
    Q8_0
    [2048, 512]
  • blk.15.attn_norm.weight
    F32
    [2048]
  • blk.15.attn_output.weight
    Q8_0
    [2048, 2048]
  • blk.15.attn_q.weight
    Q8_0
    [2048, 2048]
  • blk.15.attn_v.weight
    Q8_0
    [2048, 512]
  • blk.15.ffn_down.weight
    Q8_0
    [7168, 2048]
  • blk.15.ffn_gate.weight
    Q8_0
    [2048, 7168]
  • blk.15.ffn_norm.weight
    F32
    [2048]
  • blk.15.ffn_up.weight
    Q8_0
    [2048, 7168]
  • output.weight
    Q8_0
    [2048, 32768]
  • output_norm.weight
    F32
    [2048]