cnmoro/
smol_llama-101m-gqa-python:q8_0

6 1 year ago

... /
model
f9528e5a32b6 · 108MB
    Metadata
  • general.architecture
    llama
  • general.file_type
    Q8_0
  • llama.attention.head_count
    24
  • llama.attention.head_count_kv
    8
  • llama.attention.layer_norm_rms_epsilon
    1e-05
  • llama.block_count
    6
  • llama.context_length
    1024
  • llama.embedding_length
    768
  • llama.feed_forward_length
    3072
  • llama.rope.dimension_count
    32
  • llama.rope.freq_base
    10000
  • 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>, ...]
    • Tensor
  • token_embd.weight
    Q8_0
    [768, 32128]
    • blk.0
  • blk.0.attn_k.weight
    Q8_0
    [768, 256]
  • blk.0.attn_norm.weight
    F32
    [768]
  • blk.0.attn_output.weight
    Q8_0
    [768, 768]
  • blk.0.attn_q.weight
    Q8_0
    [768, 768]
  • blk.0.attn_v.weight
    Q8_0
    [768, 256]
  • blk.0.ffn_down.weight
    Q8_0
    [3072, 768]
  • blk.0.ffn_gate.weight
    Q8_0
    [768, 3072]
  • blk.0.ffn_norm.weight
    F32
    [768]
  • blk.0.ffn_up.weight
    Q8_0
    [768, 3072]
    • blk.1
  • blk.1.attn_k.weight
    Q8_0
    [768, 256]
  • blk.1.attn_norm.weight
    F32
    [768]
  • blk.1.attn_output.weight
    Q8_0
    [768, 768]
  • blk.1.attn_q.weight
    Q8_0
    [768, 768]
  • blk.1.attn_v.weight
    Q8_0
    [768, 256]
  • blk.1.ffn_down.weight
    Q8_0
    [3072, 768]
  • blk.1.ffn_gate.weight
    Q8_0
    [768, 3072]
  • blk.1.ffn_norm.weight
    F32
    [768]
  • blk.1.ffn_up.weight
    Q8_0
    [768, 3072]
    • blk.2
  • blk.2.attn_k.weight
    Q8_0
    [768, 256]
  • blk.2.attn_norm.weight
    F32
    [768]
  • blk.2.attn_output.weight
    Q8_0
    [768, 768]
  • blk.2.attn_q.weight
    Q8_0
    [768, 768]
  • blk.2.attn_v.weight
    Q8_0
    [768, 256]
  • blk.2.ffn_down.weight
    Q8_0
    [3072, 768]
  • blk.2.ffn_gate.weight
    Q8_0
    [768, 3072]
  • blk.2.ffn_norm.weight
    F32
    [768]
  • blk.2.ffn_up.weight
    Q8_0
    [768, 3072]
    • blk.3
  • blk.3.attn_k.weight
    Q8_0
    [768, 256]
  • blk.3.attn_norm.weight
    F32
    [768]
  • blk.3.attn_output.weight
    Q8_0
    [768, 768]
  • blk.3.attn_q.weight
    Q8_0
    [768, 768]
  • blk.3.attn_v.weight
    Q8_0
    [768, 256]
  • blk.3.ffn_down.weight
    Q8_0
    [3072, 768]
  • blk.3.ffn_gate.weight
    Q8_0
    [768, 3072]
  • blk.3.ffn_norm.weight
    F32
    [768]
  • blk.3.ffn_up.weight
    Q8_0
    [768, 3072]
    • blk.4
  • blk.4.attn_k.weight
    Q8_0
    [768, 256]
  • blk.4.attn_norm.weight
    F32
    [768]
  • blk.4.attn_output.weight
    Q8_0
    [768, 768]
  • blk.4.attn_q.weight
    Q8_0
    [768, 768]
  • blk.4.attn_v.weight
    Q8_0
    [768, 256]
  • blk.4.ffn_down.weight
    Q8_0
    [3072, 768]
  • blk.4.ffn_gate.weight
    Q8_0
    [768, 3072]
  • blk.4.ffn_norm.weight
    F32
    [768]
  • blk.4.ffn_up.weight
    Q8_0
    [768, 3072]
    • blk.5
  • blk.5.attn_k.weight
    Q8_0
    [768, 256]
  • blk.5.attn_norm.weight
    F32
    [768]
  • blk.5.attn_output.weight
    Q8_0
    [768, 768]
  • blk.5.attn_q.weight
    Q8_0
    [768, 768]
  • blk.5.attn_v.weight
    Q8_0
    [768, 256]
  • blk.5.ffn_down.weight
    Q8_0
    [3072, 768]
  • blk.5.ffn_gate.weight
    Q8_0
    [768, 3072]
  • blk.5.ffn_norm.weight
    F32
    [768]
  • blk.5.ffn_up.weight
    Q8_0
    [768, 3072]
  • output.weight
    Q8_0
    [768, 32128]
  • output_norm.weight
    F32
    [768]