Hudson / gpt2-instruct

Nvidia Megatron GPT2 345m Instruct

27 6 weeks ago

1 Tag
... /
model
7eec74653f28 · 437MB
    Metadata
  • general.architecture
    gpt2
  • general.file_type
    7
  • general.name
    megatron-gpt2-345m-evol_instruct_v2
  • general.quantization_version
    2
  • gpt2.attention.head_count
    16
  • gpt2.attention.layer_norm_epsilon
    1e-05
  • gpt2.block_count
    24
  • gpt2.context_length
    1024
  • gpt2.embedding_length
    1024
  • gpt2.feed_forward_length
    4096
  • tokenizer.ggml.bos_token_id
    50256
  • tokenizer.ggml.eos_token_id
    50256
  • tokenizer.ggml.merges
    [Ġ t, Ġ a, h e, i n, r e, ...]
  • tokenizer.ggml.model
    gpt2
  • tokenizer.ggml.pre
    gpt-2
  • tokenizer.ggml.token_type
    [1, 1, 1, 1, 1, ...]
  • tokenizer.ggml.tokens
    [!, ", #, $, %, ...]
  • tokenizer.ggml.unknown_token_id
    50256
    • Tensor
  • Name
    Type
    Shape
  • token_embd.weight
    Q8_0
    [1024, 50257]
    • blk.0
  • blk.0.attn_norm.bias
    F32
    [1024]
  • blk.0.attn_norm.weight
    F32
    [1024]
  • blk.0.attn_output.bias
    F32
    [1024]
  • blk.0.attn_output.weight
    Q8_0
    [1024, 1024]
  • blk.0.attn_qkv.bias
    F32
    [3072]
  • blk.0.attn_qkv.weight
    Q8_0
    [1024, 3072]
  • blk.0.ffn_down.bias
    F32
    [1024]
  • blk.0.ffn_down.weight
    Q8_0
    [4096, 1024]
  • blk.0.ffn_norm.bias
    F32
    [1024]
  • blk.0.ffn_norm.weight
    F32
    [1024]
  • blk.0.ffn_up.bias
    F32
    [4096]
  • blk.0.ffn_up.weight
    Q8_0
    [1024, 4096]
    • blk.1
  • blk.1.attn_norm.bias
    F32
    [1024]
  • blk.1.attn_norm.weight
    F32
    [1024]
  • blk.1.attn_output.bias
    F32
    [1024]
  • blk.1.attn_output.weight
    Q8_0
    [1024, 1024]
  • blk.1.attn_qkv.bias
    F32
    [3072]
  • blk.1.attn_qkv.weight
    Q8_0
    [1024, 3072]
  • blk.1.ffn_down.bias
    F32
    [1024]
  • blk.1.ffn_down.weight
    Q8_0
    [4096, 1024]
  • blk.1.ffn_norm.bias
    F32
    [1024]
  • blk.1.ffn_norm.weight
    F32
    [1024]
  • blk.1.ffn_up.bias
    F32
    [4096]
  • blk.1.ffn_up.weight
    Q8_0
    [1024, 4096]
    • blk.2
  • blk.2.attn_norm.bias
    F32
    [1024]
  • blk.2.attn_norm.weight
    F32
    [1024]
  • blk.2.attn_output.bias
    F32
    [1024]
  • blk.2.attn_output.weight
    Q8_0
    [1024, 1024]
  • blk.2.attn_qkv.bias
    F32
    [3072]
  • blk.2.attn_qkv.weight
    Q8_0
    [1024, 3072]
  • blk.2.ffn_down.bias
    F32
    [1024]
  • blk.2.ffn_down.weight
    Q8_0
    [4096, 1024]
  • blk.2.ffn_norm.bias
    F32
    [1024]
  • blk.2.ffn_norm.weight
    F32
    [1024]
  • blk.2.ffn_up.bias
    F32
    [4096]
  • blk.2.ffn_up.weight
    Q8_0
    [1024, 4096]
    • blk.3
  • blk.3.attn_norm.bias
    F32
    [1024]
  • blk.3.attn_norm.weight
    F32
    [1024]
  • blk.3.attn_output.bias
    F32
    [1024]
  • blk.3.attn_output.weight
    Q8_0
    [1024, 1024]
  • blk.3.attn_qkv.bias
    F32
    [3072]
  • blk.3.attn_qkv.weight
    Q8_0
    [1024, 3072]
  • blk.3.ffn_down.bias
    F32
    [1024]
  • blk.3.ffn_down.weight
    Q8_0
    [4096, 1024]
  • blk.3.ffn_norm.bias
    F32
    [1024]
  • blk.3.ffn_norm.weight
    F32
    [1024]
  • blk.3.ffn_up.bias
    F32
    [4096]
  • blk.3.ffn_up.weight
    Q8_0
    [1024, 4096]
    • blk.4
  • blk.4.attn_norm.bias
    F32
    [1024]
  • blk.4.attn_norm.weight
    F32
    [1024]
  • blk.4.attn_output.bias
    F32
    [1024]
  • blk.4.attn_output.weight
    Q8_0
    [1024, 1024]
  • blk.4.attn_qkv.bias
    F32
    [3072]
  • blk.4.attn_qkv.weight
    Q8_0
    [1024, 3072]
  • blk.4.ffn_down.bias
    F32
    [1024]
  • blk.4.ffn_down.weight
    Q8_0
    [4096, 1024]
  • blk.4.ffn_norm.bias
    F32
    [1024]
  • blk.4.ffn_norm.weight
    F32
    [1024]
  • blk.4.ffn_up.bias
    F32
    [4096]
  • blk.4.ffn_up.weight
    Q8_0
    [1024, 4096]
    • blk.5
  • blk.5.attn_norm.bias
    F32
    [1024]
  • blk.5.attn_norm.weight
    F32
    [1024]
  • blk.5.attn_output.bias
    F32
    [1024]
  • blk.5.attn_output.weight
    Q8_0
    [1024, 1024]
  • blk.5.attn_qkv.bias
    F32
    [3072]
  • blk.5.attn_qkv.weight
    Q8_0
    [1024, 3072]
  • blk.5.ffn_down.bias
    F32
    [1024]
  • blk.5.ffn_down.weight
    Q8_0
    [4096, 1024]
  • blk.5.ffn_norm.bias
    F32
    [1024]
  • blk.5.ffn_norm.weight
    F32
    [1024]
  • blk.5.ffn_up.bias
    F32
    [4096]
  • blk.5.ffn_up.weight
    Q8_0
    [1024, 4096]
    • blk.6
  • blk.6.attn_norm.bias
    F32
    [1024]
  • blk.6.attn_norm.weight
    F32
    [1024]
  • blk.6.attn_output.bias
    F32
    [1024]
  • blk.6.attn_output.weight
    Q8_0
    [1024, 1024]
  • blk.6.attn_qkv.bias
    F32
    [3072]
  • blk.6.attn_qkv.weight
    Q8_0
    [1024, 3072]
  • blk.6.ffn_down.bias
    F32
    [1024]
  • blk.6.ffn_down.weight
    Q8_0
    [4096, 1024]
  • blk.6.ffn_norm.bias
    F32
    [1024]
  • blk.6.ffn_norm.weight
    F32
    [1024]
  • blk.6.ffn_up.bias
    F32
    [4096]
  • blk.6.ffn_up.weight
    Q8_0
    [1024, 4096]
    • blk.7
  • blk.7.attn_norm.bias
    F32
    [1024]
  • blk.7.attn_norm.weight
    F32
    [1024]
  • blk.7.attn_output.bias
    F32
    [1024]
  • blk.7.attn_output.weight
    Q8_0
    [1024, 1024]
  • blk.7.attn_qkv.bias
    F32
    [3072]
  • blk.7.attn_qkv.weight
    Q8_0
    [1024, 3072]
  • blk.7.ffn_down.bias
    F32
    [1024]
  • blk.7.ffn_down.weight
    Q8_0
    [4096, 1024]
  • blk.7.ffn_norm.bias
    F32
    [1024]
  • blk.7.ffn_norm.weight
    F32
    [1024]
  • blk.7.ffn_up.bias
    F32
    [4096]
  • blk.7.ffn_up.weight
    Q8_0
    [1024, 4096]
    • blk.8
  • blk.8.attn_norm.bias
    F32
    [1024]
  • blk.8.attn_norm.weight
    F32
    [1024]
  • blk.8.attn_output.bias
    F32
    [1024]
  • blk.8.attn_output.weight
    Q8_0
    [1024, 1024]
  • blk.8.attn_qkv.bias
    F32
    [3072]
  • blk.8.attn_qkv.weight
    Q8_0
    [1024, 3072]
  • blk.8.ffn_down.bias
    F32
    [1024]
  • blk.8.ffn_down.weight
    Q8_0
    [4096, 1024]
  • blk.8.ffn_norm.bias
    F32
    [1024]
  • blk.8.ffn_norm.weight
    F32
    [1024]
  • blk.8.ffn_up.bias
    F32
    [4096]
  • blk.8.ffn_up.weight
    Q8_0
    [1024, 4096]
    • blk.9
  • blk.9.attn_norm.bias
    F32
    [1024]
  • blk.9.attn_norm.weight
    F32
    [1024]
  • blk.9.attn_output.bias
    F32
    [1024]
  • blk.9.attn_output.weight
    Q8_0
    [1024, 1024]
  • blk.9.attn_qkv.bias
    F32
    [3072]
  • blk.9.attn_qkv.weight
    Q8_0
    [1024, 3072]
  • blk.9.ffn_down.bias
    F32
    [1024]
  • blk.9.ffn_down.weight
    Q8_0
    [4096, 1024]
  • blk.9.ffn_norm.bias
    F32
    [1024]
  • blk.9.ffn_norm.weight
    F32
    [1024]
  • blk.9.ffn_up.bias
    F32
    [4096]
  • blk.9.ffn_up.weight
    Q8_0
    [1024, 4096]
    • blk.10
  • blk.10.attn_norm.bias
    F32
    [1024]
  • blk.10.attn_norm.weight
    F32
    [1024]
  • blk.10.attn_output.bias
    F32
    [1024]
  • blk.10.attn_output.weight
    Q8_0
    [1024, 1024]
  • blk.10.attn_qkv.bias
    F32
    [3072]
  • blk.10.attn_qkv.weight
    Q8_0
    [1024, 3072]
  • blk.10.ffn_down.bias
    F32
    [1024]
  • blk.10.ffn_down.weight
    Q8_0
    [4096, 1024]
  • blk.10.ffn_norm.bias
    F32
    [1024]
  • blk.10.ffn_norm.weight
    F32
    [1024]
  • blk.10.ffn_up.bias
    F32
    [4096]
  • blk.10.ffn_up.weight
    Q8_0
    [1024, 4096]
    • blk.11
  • blk.11.attn_norm.bias
    F32
    [1024]
  • blk.11.attn_norm.weight
    F32
    [1024]
  • blk.11.attn_output.bias
    F32
    [1024]
  • blk.11.attn_output.weight
    Q8_0
    [1024, 1024]
  • blk.11.attn_qkv.bias
    F32
    [3072]
  • blk.11.attn_qkv.weight
    Q8_0
    [1024, 3072]
  • blk.11.ffn_down.bias
    F32
    [1024]
  • blk.11.ffn_down.weight
    Q8_0
    [4096, 1024]
  • blk.11.ffn_norm.bias
    F32
    [1024]
  • blk.11.ffn_norm.weight
    F32
    [1024]
  • blk.11.ffn_up.bias
    F32
    [4096]
  • blk.11.ffn_up.weight
    Q8_0
    [1024, 4096]
    • blk.12
  • blk.12.attn_norm.bias
    F32
    [1024]
  • blk.12.attn_norm.weight
    F32
    [1024]
  • blk.12.attn_output.bias
    F32
    [1024]
  • blk.12.attn_output.weight
    Q8_0
    [1024, 1024]
  • blk.12.attn_qkv.bias
    F32
    [3072]
  • blk.12.attn_qkv.weight
    Q8_0
    [1024, 3072]
  • blk.12.ffn_down.bias
    F32
    [1024]
  • blk.12.ffn_down.weight
    Q8_0
    [4096, 1024]
  • blk.12.ffn_norm.bias
    F32
    [1024]
  • blk.12.ffn_norm.weight
    F32
    [1024]
  • blk.12.ffn_up.bias
    F32
    [4096]
  • blk.12.ffn_up.weight
    Q8_0
    [1024, 4096]
    • blk.13
  • blk.13.attn_norm.bias
    F32
    [1024]
  • blk.13.attn_norm.weight
    F32
    [1024]
  • blk.13.attn_output.bias
    F32
    [1024]
  • blk.13.attn_output.weight
    Q8_0
    [1024, 1024]
  • blk.13.attn_qkv.bias
    F32
    [3072]
  • blk.13.attn_qkv.weight
    Q8_0
    [1024, 3072]
  • blk.13.ffn_down.bias
    F32
    [1024]
  • blk.13.ffn_down.weight
    Q8_0
    [4096, 1024]
  • blk.13.ffn_norm.bias
    F32
    [1024]
  • blk.13.ffn_norm.weight
    F32
    [1024]
  • blk.13.ffn_up.bias
    F32
    [4096]
  • blk.13.ffn_up.weight
    Q8_0
    [1024, 4096]
    • blk.14
  • blk.14.attn_norm.bias
    F32
    [1024]
  • blk.14.attn_norm.weight
    F32
    [1024]
  • blk.14.attn_output.bias
    F32
    [1024]
  • blk.14.attn_output.weight
    Q8_0
    [1024, 1024]
  • blk.14.attn_qkv.bias
    F32
    [3072]
  • blk.14.attn_qkv.weight
    Q8_0
    [1024, 3072]
  • blk.14.ffn_down.bias
    F32
    [1024]
  • blk.14.ffn_down.weight
    Q8_0
    [4096, 1024]
  • blk.14.ffn_norm.bias
    F32
    [1024]
  • blk.14.ffn_norm.weight
    F32
    [1024]
  • blk.14.ffn_up.bias
    F32
    [4096]
  • blk.14.ffn_up.weight
    Q8_0
    [1024, 4096]
    • blk.15
  • blk.15.attn_norm.bias
    F32
    [1024]
  • blk.15.attn_norm.weight
    F32
    [1024]
  • blk.15.attn_output.bias
    F32
    [1024]
  • blk.15.attn_output.weight
    Q8_0
    [1024, 1024]
  • blk.15.attn_qkv.bias
    F32
    [3072]
  • blk.15.attn_qkv.weight
    Q8_0
    [1024, 3072]
  • blk.15.ffn_down.bias
    F32
    [1024]
  • blk.15.ffn_down.weight
    Q8_0
    [4096, 1024]
  • blk.15.ffn_norm.bias
    F32
    [1024]
  • blk.15.ffn_norm.weight
    F32
    [1024]
  • blk.15.ffn_up.bias
    F32
    [4096]
  • blk.15.ffn_up.weight
    Q8_0
    [1024, 4096]
    • blk.16
  • blk.16.attn_norm.bias
    F32
    [1024]
  • blk.16.attn_norm.weight
    F32
    [1024]
  • blk.16.attn_output.bias
    F32
    [1024]
  • blk.16.attn_output.weight
    Q8_0
    [1024, 1024]
  • blk.16.attn_qkv.bias
    F32
    [3072]
  • blk.16.attn_qkv.weight
    Q8_0
    [1024, 3072]
  • blk.16.ffn_down.bias
    F32
    [1024]
  • blk.16.ffn_down.weight
    Q8_0
    [4096, 1024]
  • blk.16.ffn_norm.bias
    F32
    [1024]
  • blk.16.ffn_norm.weight
    F32
    [1024]
  • blk.16.ffn_up.bias
    F32
    [4096]
  • blk.16.ffn_up.weight
    Q8_0
    [1024, 4096]
    • blk.17
  • blk.17.attn_norm.bias
    F32
    [1024]
  • blk.17.attn_norm.weight
    F32
    [1024]
  • blk.17.attn_output.bias
    F32
    [1024]
  • blk.17.attn_output.weight
    Q8_0
    [1024, 1024]
  • blk.17.attn_qkv.bias
    F32
    [3072]
  • blk.17.attn_qkv.weight
    Q8_0
    [1024, 3072]
  • blk.17.ffn_down.bias
    F32
    [1024]
  • blk.17.ffn_down.weight
    Q8_0
    [4096, 1024]
  • blk.17.ffn_norm.bias
    F32
    [1024]
  • blk.17.ffn_norm.weight
    F32
    [1024]
  • blk.17.ffn_up.bias
    F32
    [4096]
  • blk.17.ffn_up.weight
    Q8_0
    [1024, 4096]
    • blk.18
  • blk.18.attn_norm.bias
    F32
    [1024]
  • blk.18.attn_norm.weight
    F32
    [1024]
  • blk.18.attn_output.bias
    F32
    [1024]
  • blk.18.attn_output.weight
    Q8_0
    [1024, 1024]
  • blk.18.attn_qkv.bias
    F32
    [3072]
  • blk.18.attn_qkv.weight
    Q8_0
    [1024, 3072]
  • blk.18.ffn_down.bias
    F32
    [1024]
  • blk.18.ffn_down.weight
    Q8_0
    [4096, 1024]
  • blk.18.ffn_norm.bias
    F32
    [1024]
  • blk.18.ffn_norm.weight
    F32
    [1024]
  • blk.18.ffn_up.bias
    F32
    [4096]
  • blk.18.ffn_up.weight
    Q8_0
    [1024, 4096]
    • blk.19
  • blk.19.attn_norm.bias
    F32
    [1024]
  • blk.19.attn_norm.weight
    F32
    [1024]
  • blk.19.attn_output.bias
    F32
    [1024]
  • blk.19.attn_output.weight
    Q8_0
    [1024, 1024]
  • blk.19.attn_qkv.bias
    F32
    [3072]
  • blk.19.attn_qkv.weight
    Q8_0
    [1024, 3072]
  • blk.19.ffn_down.bias
    F32
    [1024]
  • blk.19.ffn_down.weight
    Q8_0
    [4096, 1024]
  • blk.19.ffn_norm.bias
    F32
    [1024]
  • blk.19.ffn_norm.weight
    F32
    [1024]
  • blk.19.ffn_up.bias
    F32
    [4096]
  • blk.19.ffn_up.weight
    Q8_0
    [1024, 4096]
    • blk.20
  • blk.20.attn_norm.bias
    F32
    [1024]
  • blk.20.attn_norm.weight
    F32
    [1024]
  • blk.20.attn_output.bias
    F32
    [1024]
  • blk.20.attn_output.weight
    Q8_0
    [1024, 1024]
  • blk.20.attn_qkv.bias
    F32
    [3072]
  • blk.20.attn_qkv.weight
    Q8_0
    [1024, 3072]
  • blk.20.ffn_down.bias
    F32
    [1024]
  • blk.20.ffn_down.weight
    Q8_0
    [4096, 1024]
  • blk.20.ffn_norm.bias
    F32
    [1024]
  • blk.20.ffn_norm.weight
    F32
    [1024]
  • blk.20.ffn_up.bias
    F32
    [4096]
  • blk.20.ffn_up.weight
    Q8_0
    [1024, 4096]
    • blk.21
  • blk.21.attn_norm.bias
    F32
    [1024]
  • blk.21.attn_norm.weight
    F32
    [1024]
  • blk.21.attn_output.bias
    F32
    [1024]
  • blk.21.attn_output.weight
    Q8_0
    [1024, 1024]
  • blk.21.attn_qkv.bias
    F32
    [3072]
  • blk.21.attn_qkv.weight
    Q8_0
    [1024, 3072]
  • blk.21.ffn_down.bias
    F32
    [1024]
  • blk.21.ffn_down.weight
    Q8_0
    [4096, 1024]
  • blk.21.ffn_norm.bias
    F32
    [1024]
  • blk.21.ffn_norm.weight
    F32
    [1024]
  • blk.21.ffn_up.bias
    F32
    [4096]
  • blk.21.ffn_up.weight
    Q8_0
    [1024, 4096]
    • blk.22
  • blk.22.attn_norm.bias
    F32
    [1024]
  • blk.22.attn_norm.weight
    F32
    [1024]
  • blk.22.attn_output.bias
    F32
    [1024]
  • blk.22.attn_output.weight
    Q8_0
    [1024, 1024]
  • blk.22.attn_qkv.bias
    F32
    [3072]
  • blk.22.attn_qkv.weight
    Q8_0
    [1024, 3072]
  • blk.22.ffn_down.bias
    F32
    [1024]
  • blk.22.ffn_down.weight
    Q8_0
    [4096, 1024]
  • blk.22.ffn_norm.bias
    F32
    [1024]
  • blk.22.ffn_norm.weight
    F32
    [1024]
  • blk.22.ffn_up.bias
    F32
    [4096]
  • blk.22.ffn_up.weight
    Q8_0
    [1024, 4096]
    • blk.23
  • blk.23.attn_norm.bias
    F32
    [1024]
  • blk.23.attn_norm.weight
    F32
    [1024]
  • blk.23.attn_output.bias
    F32
    [1024]
  • blk.23.attn_output.weight
    Q8_0
    [1024, 1024]
  • blk.23.attn_qkv.bias
    F32
    [3072]
  • blk.23.attn_qkv.weight
    Q8_0
    [1024, 3072]
  • blk.23.ffn_down.bias
    F32
    [1024]
  • blk.23.ffn_down.weight
    Q8_0
    [4096, 1024]
  • blk.23.ffn_norm.bias
    F32
    [1024]
  • blk.23.ffn_norm.weight
    F32
    [1024]
  • blk.23.ffn_up.bias
    F32
    [4096]
  • blk.23.ffn_up.weight
    Q8_0
    [1024, 4096]
  • output.weight
    Q8_0
    [1024, 50257]
  • output_norm.bias
    F32
    [1024]
  • position_embd.weight
    F32
    [1024, 1024]
  • output_norm.weight
    F32
    [1024]