Hudson/ nanochat-40M:latest

3 1 week ago

... /
model
281789a55ba3 · 43MB
    Metadata
  • general.architecture
    llama
  • general.file_type
    Q8_0
  • llama.attention.head_count
    8
  • 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
    10
  • llama.context_length
    1024
  • llama.embedding_length
    512
  • llama.feed_forward_length
    1536
  • llama.rope.dimension_count
    64
  • llama.rope.freq_base
    10000
  • llama.vocab_size
    16384
  • tokenizer.ggml.add_bos_token
    false
  • tokenizer.ggml.add_eos_token
    false
  • tokenizer.ggml.bos_token_id
    0
  • tokenizer.ggml.eos_token_id
    2
  • tokenizer.ggml.merges
    [Ġ t, Ġ a, i n, h e, r e, ...]
  • tokenizer.ggml.model
    gpt2
  • tokenizer.ggml.padding_token_id
    10
  • tokenizer.ggml.pre
    gpt-2
  • tokenizer.ggml.token_type
    [3, 3, 3, 3, 3, ...]
  • tokenizer.ggml.tokens
    [<|endoftext|>, <|im_start|>, <|im_end|>, <think>, </think>, ...]
    • Tensor
  • token_embd.weight
    Q8_0
    [512, 16384]
    • blk.0
  • blk.0.attn_k.weight
    Q8_0
    [512, 256]
  • blk.0.attn_norm.weight
    F32
    [512]
  • blk.0.attn_output.weight
    Q8_0
    [512, 512]
  • blk.0.attn_q.weight
    Q8_0
    [512, 512]
  • blk.0.attn_v.weight
    Q8_0
    [512, 256]
  • blk.0.ffn_down.weight
    Q8_0
    [1536, 512]
  • blk.0.ffn_gate.weight
    Q8_0
    [512, 1536]
  • blk.0.ffn_norm.weight
    F32
    [512]
  • blk.0.ffn_up.weight
    Q8_0
    [512, 1536]
    • blk.1
  • blk.1.attn_k.weight
    Q8_0
    [512, 256]
  • blk.1.attn_norm.weight
    F32
    [512]
  • blk.1.attn_output.weight
    Q8_0
    [512, 512]
  • blk.1.attn_q.weight
    Q8_0
    [512, 512]
  • blk.1.attn_v.weight
    Q8_0
    [512, 256]
  • blk.1.ffn_down.weight
    Q8_0
    [1536, 512]
  • blk.1.ffn_gate.weight
    Q8_0
    [512, 1536]
  • blk.1.ffn_norm.weight
    F32
    [512]
  • blk.1.ffn_up.weight
    Q8_0
    [512, 1536]
    • blk.2
  • blk.2.attn_k.weight
    Q8_0
    [512, 256]
  • blk.2.attn_norm.weight
    F32
    [512]
  • blk.2.attn_output.weight
    Q8_0
    [512, 512]
  • blk.2.attn_q.weight
    Q8_0
    [512, 512]
  • blk.2.attn_v.weight
    Q8_0
    [512, 256]
  • blk.2.ffn_down.weight
    Q8_0
    [1536, 512]
  • blk.2.ffn_gate.weight
    Q8_0
    [512, 1536]
  • blk.2.ffn_norm.weight
    F32
    [512]
  • blk.2.ffn_up.weight
    Q8_0
    [512, 1536]
    • blk.3
  • blk.3.attn_k.weight
    Q8_0
    [512, 256]
  • blk.3.attn_norm.weight
    F32
    [512]
  • blk.3.attn_output.weight
    Q8_0
    [512, 512]
  • blk.3.attn_q.weight
    Q8_0
    [512, 512]
  • blk.3.attn_v.weight
    Q8_0
    [512, 256]
  • blk.3.ffn_down.weight
    Q8_0
    [1536, 512]
  • blk.3.ffn_gate.weight
    Q8_0
    [512, 1536]
  • blk.3.ffn_norm.weight
    F32
    [512]
  • blk.3.ffn_up.weight
    Q8_0
    [512, 1536]
    • blk.4
  • blk.4.attn_k.weight
    Q8_0
    [512, 256]
  • blk.4.attn_norm.weight
    F32
    [512]
  • blk.4.attn_output.weight
    Q8_0
    [512, 512]
  • blk.4.attn_q.weight
    Q8_0
    [512, 512]
  • blk.4.attn_v.weight
    Q8_0
    [512, 256]
  • blk.4.ffn_down.weight
    Q8_0
    [1536, 512]
  • blk.4.ffn_gate.weight
    Q8_0
    [512, 1536]
  • blk.4.ffn_norm.weight
    F32
    [512]
  • blk.4.ffn_up.weight
    Q8_0
    [512, 1536]
    • blk.5
  • blk.5.attn_k.weight
    Q8_0
    [512, 256]
  • blk.5.attn_norm.weight
    F32
    [512]
  • blk.5.attn_output.weight
    Q8_0
    [512, 512]
  • blk.5.attn_q.weight
    Q8_0
    [512, 512]
  • blk.5.attn_v.weight
    Q8_0
    [512, 256]
  • blk.5.ffn_down.weight
    Q8_0
    [1536, 512]
  • blk.5.ffn_gate.weight
    Q8_0
    [512, 1536]
  • blk.5.ffn_norm.weight
    F32
    [512]
  • blk.5.ffn_up.weight
    Q8_0
    [512, 1536]
    • blk.6
  • blk.6.attn_k.weight
    Q8_0
    [512, 256]
  • blk.6.attn_norm.weight
    F32
    [512]
  • blk.6.attn_output.weight
    Q8_0
    [512, 512]
  • blk.6.attn_q.weight
    Q8_0
    [512, 512]
  • blk.6.attn_v.weight
    Q8_0
    [512, 256]
  • blk.6.ffn_down.weight
    Q8_0
    [1536, 512]
  • blk.6.ffn_gate.weight
    Q8_0
    [512, 1536]
  • blk.6.ffn_norm.weight
    F32
    [512]
  • blk.6.ffn_up.weight
    Q8_0
    [512, 1536]
    • blk.7
  • blk.7.attn_k.weight
    Q8_0
    [512, 256]
  • blk.7.attn_norm.weight
    F32
    [512]
  • blk.7.attn_output.weight
    Q8_0
    [512, 512]
  • blk.7.attn_q.weight
    Q8_0
    [512, 512]
  • blk.7.attn_v.weight
    Q8_0
    [512, 256]
  • blk.7.ffn_down.weight
    Q8_0
    [1536, 512]
  • blk.7.ffn_gate.weight
    Q8_0
    [512, 1536]
  • blk.7.ffn_norm.weight
    F32
    [512]
  • blk.7.ffn_up.weight
    Q8_0
    [512, 1536]
    • blk.8
  • blk.8.attn_k.weight
    Q8_0
    [512, 256]
  • blk.8.attn_norm.weight
    F32
    [512]
  • blk.8.attn_output.weight
    Q8_0
    [512, 512]
  • blk.8.attn_q.weight
    Q8_0
    [512, 512]
  • blk.8.attn_v.weight
    Q8_0
    [512, 256]
  • blk.8.ffn_down.weight
    Q8_0
    [1536, 512]
  • blk.8.ffn_gate.weight
    Q8_0
    [512, 1536]
  • blk.8.ffn_norm.weight
    F32
    [512]
  • blk.8.ffn_up.weight
    Q8_0
    [512, 1536]
    • blk.9
  • blk.9.attn_k.weight
    Q8_0
    [512, 256]
  • blk.9.attn_norm.weight
    F32
    [512]
  • blk.9.attn_output.weight
    Q8_0
    [512, 512]
  • blk.9.attn_q.weight
    Q8_0
    [512, 512]
  • blk.9.attn_v.weight
    Q8_0
    [512, 256]
  • blk.9.ffn_down.weight
    Q8_0
    [1536, 512]
  • blk.9.ffn_gate.weight
    Q8_0
    [512, 1536]
  • blk.9.ffn_norm.weight
    F32
    [512]
  • blk.9.ffn_up.weight
    Q8_0
    [512, 1536]
  • output_norm.weight
    F32
    [512]