evilfreelancer/
rugpt3

Language model for Russian

125m 356m 760m

178 5 months ago

19 Tags
... /
model
65574de167a6 · 182MB
    Metadata
  • general.architecture
    gpt2
  • general.file_type
    7
  • general.name
    Rugpt3Small_Based_On_Gpt2
  • general.organization
    Ai Forever
  • general.quantization_version
    2
  • general.size_label
    164M
  • general.type
    model
  • gpt2.attention.head_count
    12
  • gpt2.attention.layer_norm_epsilon
    1e-05
  • gpt2.block_count
    12
  • gpt2.context_length
    2048
  • gpt2.embedding_length
    768
  • gpt2.feed_forward_length
    3072
  • tokenizer.ggml.add_bos_token
    false
  • tokenizer.ggml.bos_token_id
    1
  • tokenizer.ggml.eos_token_id
    2
  • tokenizer.ggml.mask_token_id
    4
  • tokenizer.ggml.merges
    [Ġ Ð, Ð ¾, Ð µ, Ð °, Ð ¸, ...]
  • tokenizer.ggml.model
    gpt2
  • tokenizer.ggml.padding_token_id
    0
  • tokenizer.ggml.token_type
    [3, 3, 3, 3, 3, ...]
  • tokenizer.ggml.tokens
    [<pad>, <s>, </s>, <unk>, <mask>, ...]
  • tokenizer.ggml.unknown_token_id
    3
    • Tensor
  • token_embd.weight
    Q8_0
    [768, 50264]
    • blk.0
  • blk.0.attn_norm.bias
    F32
    [768]
  • blk.0.attn_norm.weight
    F32
    [768]
  • blk.0.attn_output.bias
    F32
    [768]
  • blk.0.attn_output.weight
    Q8_0
    [768, 768]
  • blk.0.attn_qkv.bias
    F32
    [2304]
  • blk.0.attn_qkv.weight
    Q8_0
    [768, 2304]
  • blk.0.ffn_down.bias
    F32
    [768]
  • blk.0.ffn_down.weight
    Q8_0
    [3072, 768]
  • blk.0.ffn_norm.bias
    F32
    [768]
  • blk.0.ffn_norm.weight
    F32
    [768]
  • blk.0.ffn_up.bias
    F32
    [3072]
  • blk.0.ffn_up.weight
    Q8_0
    [768, 3072]
    • blk.1
  • blk.1.attn_norm.bias
    F32
    [768]
  • blk.1.attn_norm.weight
    F32
    [768]
  • blk.1.attn_output.bias
    F32
    [768]
  • blk.1.attn_output.weight
    Q8_0
    [768, 768]
  • blk.1.attn_qkv.bias
    F32
    [2304]
  • blk.1.attn_qkv.weight
    Q8_0
    [768, 2304]
  • blk.1.ffn_down.bias
    F32
    [768]
  • blk.1.ffn_down.weight
    Q8_0
    [3072, 768]
  • blk.1.ffn_norm.bias
    F32
    [768]
  • blk.1.ffn_norm.weight
    F32
    [768]
  • blk.1.ffn_up.bias
    F32
    [3072]
  • blk.1.ffn_up.weight
    Q8_0
    [768, 3072]
    • blk.2
  • blk.2.attn_norm.bias
    F32
    [768]
  • blk.2.attn_norm.weight
    F32
    [768]
  • blk.2.attn_output.bias
    F32
    [768]
  • blk.2.attn_output.weight
    Q8_0
    [768, 768]
  • blk.2.attn_qkv.bias
    F32
    [2304]
  • blk.2.attn_qkv.weight
    Q8_0
    [768, 2304]
  • blk.2.ffn_down.bias
    F32
    [768]
  • blk.2.ffn_down.weight
    Q8_0
    [3072, 768]
  • blk.2.ffn_norm.bias
    F32
    [768]
  • blk.2.ffn_norm.weight
    F32
    [768]
  • blk.2.ffn_up.bias
    F32
    [3072]
  • blk.2.ffn_up.weight
    Q8_0
    [768, 3072]
    • blk.3
  • blk.3.attn_norm.bias
    F32
    [768]
  • blk.3.attn_norm.weight
    F32
    [768]
  • blk.3.attn_output.bias
    F32
    [768]
  • blk.3.attn_output.weight
    Q8_0
    [768, 768]
  • blk.3.attn_qkv.bias
    F32
    [2304]
  • blk.3.attn_qkv.weight
    Q8_0
    [768, 2304]
  • blk.3.ffn_down.bias
    F32
    [768]
  • blk.3.ffn_down.weight
    Q8_0
    [3072, 768]
  • blk.3.ffn_norm.bias
    F32
    [768]
  • blk.3.ffn_norm.weight
    F32
    [768]
  • blk.3.ffn_up.bias
    F32
    [3072]
  • blk.3.ffn_up.weight
    Q8_0
    [768, 3072]
    • blk.4
  • blk.4.attn_norm.bias
    F32
    [768]
  • blk.4.attn_norm.weight
    F32
    [768]
  • blk.4.attn_output.bias
    F32
    [768]
  • blk.4.attn_output.weight
    Q8_0
    [768, 768]
  • blk.4.attn_qkv.bias
    F32
    [2304]
  • blk.4.attn_qkv.weight
    Q8_0
    [768, 2304]
  • blk.4.ffn_down.bias
    F32
    [768]
  • blk.4.ffn_down.weight
    Q8_0
    [3072, 768]
  • blk.4.ffn_norm.bias
    F32
    [768]
  • blk.4.ffn_norm.weight
    F32
    [768]
  • blk.4.ffn_up.bias
    F32
    [3072]
  • blk.4.ffn_up.weight
    Q8_0
    [768, 3072]
    • blk.5
  • blk.5.attn_norm.bias
    F32
    [768]
  • blk.5.attn_norm.weight
    F32
    [768]
  • blk.5.attn_output.bias
    F32
    [768]
  • blk.5.attn_output.weight
    Q8_0
    [768, 768]
  • blk.5.attn_qkv.bias
    F32
    [2304]
  • blk.5.attn_qkv.weight
    Q8_0
    [768, 2304]
  • blk.5.ffn_down.bias
    F32
    [768]
  • blk.5.ffn_down.weight
    Q8_0
    [3072, 768]
  • blk.5.ffn_norm.bias
    F32
    [768]
  • blk.5.ffn_norm.weight
    F32
    [768]
  • blk.5.ffn_up.bias
    F32
    [3072]
  • blk.5.ffn_up.weight
    Q8_0
    [768, 3072]
    • blk.6
  • blk.6.attn_norm.bias
    F32
    [768]
  • blk.6.attn_norm.weight
    F32
    [768]
  • blk.6.attn_output.bias
    F32
    [768]
  • blk.6.attn_output.weight
    Q8_0
    [768, 768]
  • blk.6.attn_qkv.bias
    F32
    [2304]
  • blk.6.attn_qkv.weight
    Q8_0
    [768, 2304]
  • blk.6.ffn_down.bias
    F32
    [768]
  • blk.6.ffn_down.weight
    Q8_0
    [3072, 768]
  • blk.6.ffn_norm.bias
    F32
    [768]
  • blk.6.ffn_norm.weight
    F32
    [768]
  • blk.6.ffn_up.bias
    F32
    [3072]
  • blk.6.ffn_up.weight
    Q8_0
    [768, 3072]
    • blk.7
  • blk.7.attn_norm.bias
    F32
    [768]
  • blk.7.attn_norm.weight
    F32
    [768]
  • blk.7.attn_output.bias
    F32
    [768]
  • blk.7.attn_output.weight
    Q8_0
    [768, 768]
  • blk.7.attn_qkv.bias
    F32
    [2304]
  • blk.7.attn_qkv.weight
    Q8_0
    [768, 2304]
  • blk.7.ffn_down.bias
    F32
    [768]
  • blk.7.ffn_down.weight
    Q8_0
    [3072, 768]
  • blk.7.ffn_norm.bias
    F32
    [768]
  • blk.7.ffn_norm.weight
    F32
    [768]
  • blk.7.ffn_up.bias
    F32
    [3072]
  • blk.7.ffn_up.weight
    Q8_0
    [768, 3072]
    • blk.8
  • blk.8.attn_norm.bias
    F32
    [768]
  • blk.8.attn_norm.weight
    F32
    [768]
  • blk.8.attn_output.bias
    F32
    [768]
  • blk.8.attn_output.weight
    Q8_0
    [768, 768]
  • blk.8.attn_qkv.bias
    F32
    [2304]
  • blk.8.attn_qkv.weight
    Q8_0
    [768, 2304]
  • blk.8.ffn_down.bias
    F32
    [768]
  • blk.8.ffn_down.weight
    Q8_0
    [3072, 768]
  • blk.8.ffn_norm.bias
    F32
    [768]
  • blk.8.ffn_norm.weight
    F32
    [768]
  • blk.8.ffn_up.bias
    F32
    [3072]
  • blk.8.ffn_up.weight
    Q8_0
    [768, 3072]
    • blk.9
  • blk.9.attn_norm.bias
    F32
    [768]
  • blk.9.attn_norm.weight
    F32
    [768]
  • blk.9.attn_output.bias
    F32
    [768]
  • blk.9.attn_output.weight
    Q8_0
    [768, 768]
  • blk.9.attn_qkv.bias
    F32
    [2304]
  • blk.9.attn_qkv.weight
    Q8_0
    [768, 2304]
  • blk.9.ffn_down.bias
    F32
    [768]
  • blk.9.ffn_down.weight
    Q8_0
    [3072, 768]
  • blk.9.ffn_norm.bias
    F32
    [768]
  • blk.9.ffn_norm.weight
    F32
    [768]
  • blk.9.ffn_up.bias
    F32
    [3072]
  • blk.9.ffn_up.weight
    Q8_0
    [768, 3072]
    • blk.10
  • blk.10.attn_norm.bias
    F32
    [768]
  • blk.10.attn_norm.weight
    F32
    [768]
  • blk.10.attn_output.bias
    F32
    [768]
  • blk.10.attn_output.weight
    Q8_0
    [768, 768]
  • blk.10.attn_qkv.bias
    F32
    [2304]
  • blk.10.attn_qkv.weight
    Q8_0
    [768, 2304]
  • blk.10.ffn_down.bias
    F32
    [768]
  • blk.10.ffn_down.weight
    Q8_0
    [3072, 768]
  • blk.10.ffn_norm.bias
    F32
    [768]
  • blk.10.ffn_norm.weight
    F32
    [768]
  • blk.10.ffn_up.bias
    F32
    [3072]
  • blk.10.ffn_up.weight
    Q8_0
    [768, 3072]
    • blk.11
  • blk.11.attn_norm.bias
    F32
    [768]
  • blk.11.attn_norm.weight
    F32
    [768]
  • blk.11.attn_output.bias
    F32
    [768]
  • blk.11.attn_output.weight
    Q8_0
    [768, 768]
  • blk.11.attn_qkv.bias
    F32
    [2304]
  • blk.11.attn_qkv.weight
    Q8_0
    [768, 2304]
  • blk.11.ffn_down.bias
    F32
    [768]
  • blk.11.ffn_down.weight
    Q8_0
    [3072, 768]
  • blk.11.ffn_norm.bias
    F32
    [768]
  • blk.11.ffn_norm.weight
    F32
    [768]
  • blk.11.ffn_up.bias
    F32
    [3072]
  • blk.11.ffn_up.weight
    Q8_0
    [768, 3072]
  • output.weight
    Q8_0
    [768, 50264]
  • output_norm.bias
    F32
    [768]
  • position_embd.weight
    F32
    [768, 2048]
  • output_norm.weight
    F32
    [768]