EntropyYue / chatglm3

ChatGLM3-6B 是 ChatGLM 系列最新一代的开源模型，保留了前两代模型对话流畅、部署门槛低等众多优秀特性

6b

823 2 months ago

1 Tag
... /
model
60b185bbd000 · 3.6GB
    Metadata
  • chatglm.attention.head_count
    32
  • chatglm.attention.head_count_kv
    2
  • chatglm.attention.layer_norm_rms_epsilon
    1e-05
  • chatglm.block_count
    28
  • chatglm.context_length
    131072
  • chatglm.embedding_length
    4096
  • chatglm.feed_forward_length
    13696
  • chatglm.rope.dimension_count
    64
  • chatglm.rope.freq_base
    5e+06
  • general.architecture
    chatglm
  • general.basename
    chatglm3
  • general.file_type
    2
  • general.finetune
    128k
  • general.languages
    [zh, en]
  • general.name
    Chatglm3 6b 128k
  • general.organization
    THUDM
  • general.quantization_version
    2
  • general.size_label
    6B
  • general.tags
    [glm, chatglm, thudm]
  • general.type
    model
  • tokenizer.ggml.add_bos_token
    false
  • tokenizer.ggml.eos_token_id
    2
  • tokenizer.ggml.model
    llama
  • tokenizer.ggml.padding_token_id
    0
  • tokenizer.ggml.pre
    chatglm-spm
  • tokenizer.ggml.scores
    [0, 0, 0, 0, 0, ...]
  • tokenizer.ggml.token_type
    [2, 3, 3, 6, 6, ...]
  • tokenizer.ggml.tokens
    [<unk>, <bos>, <eos>, <0x00>, <0x01>, ...]
    • Tensor
  • Name
    Type
    Shape
  • token_embd.weight
    Q4_0
    [4096, 65024]
    • blk.0
  • blk.0.attn_norm.weight
    F32
    [4096]
  • blk.0.attn_output.weight
    Q4_0
    [4096, 4096]
  • blk.0.attn_qkv.bias
    F32
    [4608]
  • blk.0.attn_qkv.weight
    Q4_0
    [4096, 4608]
  • blk.0.ffn_down.weight
    Q4_0
    [13696, 4096]
  • blk.0.ffn_norm.weight
    F32
    [4096]
  • blk.0.ffn_up.weight
    Q4_0
    [4096, 27392]
    • blk.1
  • blk.1.attn_norm.weight
    F32
    [4096]
  • blk.1.attn_output.weight
    Q4_0
    [4096, 4096]
  • blk.1.attn_qkv.bias
    F32
    [4608]
  • blk.1.attn_qkv.weight
    Q4_0
    [4096, 4608]
  • blk.1.ffn_down.weight
    Q4_0
    [13696, 4096]
  • blk.1.ffn_norm.weight
    F32
    [4096]
  • blk.1.ffn_up.weight
    Q4_0
    [4096, 27392]
    • blk.2
  • blk.2.attn_norm.weight
    F32
    [4096]
  • blk.2.attn_output.weight
    Q4_0
    [4096, 4096]
  • blk.2.attn_qkv.bias
    F32
    [4608]
  • blk.2.attn_qkv.weight
    Q4_0
    [4096, 4608]
  • blk.2.ffn_down.weight
    Q4_0
    [13696, 4096]
  • blk.2.ffn_norm.weight
    F32
    [4096]
  • blk.2.ffn_up.weight
    Q4_0
    [4096, 27392]
    • blk.3
  • blk.3.attn_norm.weight
    F32
    [4096]
  • blk.3.attn_output.weight
    Q4_0
    [4096, 4096]
  • blk.3.attn_qkv.bias
    F32
    [4608]
  • blk.3.attn_qkv.weight
    Q4_0
    [4096, 4608]
  • blk.3.ffn_down.weight
    Q4_0
    [13696, 4096]
  • blk.3.ffn_norm.weight
    F32
    [4096]
  • blk.3.ffn_up.weight
    Q4_0
    [4096, 27392]
    • blk.4
  • blk.4.attn_norm.weight
    F32
    [4096]
  • blk.4.attn_output.weight
    Q4_0
    [4096, 4096]
  • blk.4.attn_qkv.bias
    F32
    [4608]
  • blk.4.attn_qkv.weight
    Q4_0
    [4096, 4608]
  • blk.4.ffn_down.weight
    Q4_0
    [13696, 4096]
  • blk.4.ffn_norm.weight
    F32
    [4096]
  • blk.4.ffn_up.weight
    Q4_0
    [4096, 27392]
    • blk.5
  • blk.5.attn_norm.weight
    F32
    [4096]
  • blk.5.attn_output.weight
    Q4_0
    [4096, 4096]
  • blk.5.attn_qkv.bias
    F32
    [4608]
  • blk.5.attn_qkv.weight
    Q4_0
    [4096, 4608]
  • blk.5.ffn_down.weight
    Q4_0
    [13696, 4096]
  • blk.5.ffn_norm.weight
    F32
    [4096]
  • blk.5.ffn_up.weight
    Q4_0
    [4096, 27392]
    • blk.6
  • blk.6.attn_norm.weight
    F32
    [4096]
  • blk.6.attn_output.weight
    Q4_0
    [4096, 4096]
  • blk.6.attn_qkv.bias
    F32
    [4608]
  • blk.6.attn_qkv.weight
    Q4_0
    [4096, 4608]
  • blk.6.ffn_down.weight
    Q4_0
    [13696, 4096]
  • blk.6.ffn_norm.weight
    F32
    [4096]
  • blk.6.ffn_up.weight
    Q4_0
    [4096, 27392]
    • blk.7
  • blk.7.attn_norm.weight
    F32
    [4096]
  • blk.7.attn_output.weight
    Q4_0
    [4096, 4096]
  • blk.7.attn_qkv.bias
    F32
    [4608]
  • blk.7.attn_qkv.weight
    Q4_0
    [4096, 4608]
  • blk.7.ffn_down.weight
    Q4_0
    [13696, 4096]
  • blk.7.ffn_norm.weight
    F32
    [4096]
  • blk.7.ffn_up.weight
    Q4_0
    [4096, 27392]
    • blk.8
  • blk.8.attn_norm.weight
    F32
    [4096]
  • blk.8.attn_output.weight
    Q4_0
    [4096, 4096]
  • blk.8.attn_qkv.bias
    F32
    [4608]
  • blk.8.attn_qkv.weight
    Q4_0
    [4096, 4608]
  • blk.8.ffn_down.weight
    Q4_0
    [13696, 4096]
  • blk.8.ffn_norm.weight
    F32
    [4096]
  • blk.8.ffn_up.weight
    Q4_0
    [4096, 27392]
    • blk.9
  • blk.9.attn_norm.weight
    F32
    [4096]
  • blk.9.attn_output.weight
    Q4_0
    [4096, 4096]
  • blk.9.attn_qkv.bias
    F32
    [4608]
  • blk.9.attn_qkv.weight
    Q4_0
    [4096, 4608]
  • blk.9.ffn_down.weight
    Q4_0
    [13696, 4096]
  • blk.9.ffn_norm.weight
    F32
    [4096]
  • blk.9.ffn_up.weight
    Q4_0
    [4096, 27392]
    • blk.10
  • blk.10.attn_norm.weight
    F32
    [4096]
  • blk.10.attn_output.weight
    Q4_0
    [4096, 4096]
  • blk.10.attn_qkv.bias
    F32
    [4608]
  • blk.10.attn_qkv.weight
    Q4_0
    [4096, 4608]
  • blk.10.ffn_down.weight
    Q4_0
    [13696, 4096]
  • blk.10.ffn_norm.weight
    F32
    [4096]
  • blk.10.ffn_up.weight
    Q4_0
    [4096, 27392]
    • blk.11
  • blk.11.attn_norm.weight
    F32
    [4096]
  • blk.11.attn_output.weight
    Q4_0
    [4096, 4096]
  • blk.11.attn_qkv.bias
    F32
    [4608]
  • blk.11.attn_qkv.weight
    Q4_0
    [4096, 4608]
  • blk.11.ffn_down.weight
    Q4_0
    [13696, 4096]
  • blk.11.ffn_norm.weight
    F32
    [4096]
  • blk.11.ffn_up.weight
    Q4_0
    [4096, 27392]
    • blk.12
  • blk.12.attn_norm.weight
    F32
    [4096]
  • blk.12.attn_output.weight
    Q4_0
    [4096, 4096]
  • blk.12.attn_qkv.bias
    F32
    [4608]
  • blk.12.attn_qkv.weight
    Q4_0
    [4096, 4608]
  • blk.12.ffn_down.weight
    Q4_0
    [13696, 4096]
  • blk.12.ffn_norm.weight
    F32
    [4096]
  • blk.12.ffn_up.weight
    Q4_0
    [4096, 27392]
    • blk.13
  • blk.13.attn_norm.weight
    F32
    [4096]
  • blk.13.attn_output.weight
    Q4_0
    [4096, 4096]
  • blk.13.attn_qkv.bias
    F32
    [4608]
  • blk.13.attn_qkv.weight
    Q4_0
    [4096, 4608]
  • blk.13.ffn_down.weight
    Q4_0
    [13696, 4096]
  • blk.13.ffn_norm.weight
    F32
    [4096]
  • blk.13.ffn_up.weight
    Q4_0
    [4096, 27392]
    • blk.14
  • blk.14.attn_norm.weight
    F32
    [4096]
  • blk.14.attn_output.weight
    Q4_0
    [4096, 4096]
  • blk.14.attn_qkv.bias
    F32
    [4608]
  • blk.14.attn_qkv.weight
    Q4_0
    [4096, 4608]
  • blk.14.ffn_down.weight
    Q4_0
    [13696, 4096]
  • blk.14.ffn_norm.weight
    F32
    [4096]
  • blk.14.ffn_up.weight
    Q4_0
    [4096, 27392]
    • blk.15
  • blk.15.attn_norm.weight
    F32
    [4096]
  • blk.15.attn_output.weight
    Q4_0
    [4096, 4096]
  • blk.15.attn_qkv.bias
    F32
    [4608]
  • blk.15.attn_qkv.weight
    Q4_0
    [4096, 4608]
  • blk.15.ffn_down.weight
    Q4_0
    [13696, 4096]
  • blk.15.ffn_norm.weight
    F32
    [4096]
  • blk.15.ffn_up.weight
    Q4_0
    [4096, 27392]
    • blk.16
  • blk.16.attn_norm.weight
    F32
    [4096]
  • blk.16.attn_output.weight
    Q4_0
    [4096, 4096]
  • blk.16.attn_qkv.bias
    F32
    [4608]
  • blk.16.attn_qkv.weight
    Q4_0
    [4096, 4608]
  • blk.16.ffn_down.weight
    Q4_0
    [13696, 4096]
  • blk.16.ffn_norm.weight
    F32
    [4096]
  • blk.16.ffn_up.weight
    Q4_0
    [4096, 27392]
    • blk.17
  • blk.17.attn_norm.weight
    F32
    [4096]
  • blk.17.attn_output.weight
    Q4_0
    [4096, 4096]
  • blk.17.attn_qkv.bias
    F32
    [4608]
  • blk.17.attn_qkv.weight
    Q4_0
    [4096, 4608]
  • blk.17.ffn_down.weight
    Q4_0
    [13696, 4096]
  • blk.17.ffn_norm.weight
    F32
    [4096]
  • blk.17.ffn_up.weight
    Q4_0
    [4096, 27392]
    • blk.18
  • blk.18.attn_norm.weight
    F32
    [4096]
  • blk.18.attn_output.weight
    Q4_0
    [4096, 4096]
  • blk.18.attn_qkv.bias
    F32
    [4608]
  • blk.18.attn_qkv.weight
    Q4_0
    [4096, 4608]
  • blk.18.ffn_down.weight
    Q4_0
    [13696, 4096]
  • blk.18.ffn_norm.weight
    F32
    [4096]
  • blk.18.ffn_up.weight
    Q4_0
    [4096, 27392]
    • blk.19
  • blk.19.attn_norm.weight
    F32
    [4096]
  • blk.19.attn_output.weight
    Q4_0
    [4096, 4096]
  • blk.19.attn_qkv.bias
    F32
    [4608]
  • blk.19.attn_qkv.weight
    Q4_0
    [4096, 4608]
  • blk.19.ffn_down.weight
    Q4_0
    [13696, 4096]
  • blk.19.ffn_norm.weight
    F32
    [4096]
  • blk.19.ffn_up.weight
    Q4_0
    [4096, 27392]
    • blk.20
  • blk.20.attn_norm.weight
    F32
    [4096]
  • blk.20.attn_output.weight
    Q4_0
    [4096, 4096]
  • blk.20.attn_qkv.bias
    F32
    [4608]
  • blk.20.attn_qkv.weight
    Q4_0
    [4096, 4608]
  • blk.20.ffn_down.weight
    Q4_0
    [13696, 4096]
  • blk.20.ffn_norm.weight
    F32
    [4096]
  • blk.20.ffn_up.weight
    Q4_0
    [4096, 27392]
    • blk.21
  • blk.21.attn_norm.weight
    F32
    [4096]
  • blk.21.attn_output.weight
    Q4_0
    [4096, 4096]
  • blk.21.attn_qkv.bias
    F32
    [4608]
  • blk.21.attn_qkv.weight
    Q4_0
    [4096, 4608]
  • blk.21.ffn_down.weight
    Q4_0
    [13696, 4096]
  • blk.21.ffn_norm.weight
    F32
    [4096]
  • blk.21.ffn_up.weight
    Q4_0
    [4096, 27392]
    • blk.22
  • blk.22.attn_norm.weight
    F32
    [4096]
  • blk.22.attn_output.weight
    Q4_0
    [4096, 4096]
  • blk.22.attn_qkv.bias
    F32
    [4608]
  • blk.22.attn_qkv.weight
    Q4_0
    [4096, 4608]
  • blk.22.ffn_down.weight
    Q4_0
    [13696, 4096]
  • blk.22.ffn_norm.weight
    F32
    [4096]
  • blk.22.ffn_up.weight
    Q4_0
    [4096, 27392]
    • blk.23
  • blk.23.attn_norm.weight
    F32
    [4096]
  • blk.23.attn_output.weight
    Q4_0
    [4096, 4096]
  • blk.23.attn_qkv.bias
    F32
    [4608]
  • blk.23.attn_qkv.weight
    Q4_0
    [4096, 4608]
  • blk.23.ffn_down.weight
    Q4_0
    [13696, 4096]
  • blk.23.ffn_norm.weight
    F32
    [4096]
  • blk.23.ffn_up.weight
    Q4_0
    [4096, 27392]
    • blk.24
  • blk.24.attn_norm.weight
    F32
    [4096]
  • blk.24.attn_output.weight
    Q4_0
    [4096, 4096]
  • blk.24.attn_qkv.bias
    F32
    [4608]
  • blk.24.attn_qkv.weight
    Q4_0
    [4096, 4608]
  • blk.24.ffn_down.weight
    Q4_0
    [13696, 4096]
  • blk.24.ffn_norm.weight
    F32
    [4096]
  • blk.24.ffn_up.weight
    Q4_0
    [4096, 27392]
    • blk.25
  • blk.25.attn_norm.weight
    F32
    [4096]
  • blk.25.attn_output.weight
    Q4_0
    [4096, 4096]
  • blk.25.attn_qkv.bias
    F32
    [4608]
  • blk.25.attn_qkv.weight
    Q4_0
    [4096, 4608]
  • blk.25.ffn_down.weight
    Q4_0
    [13696, 4096]
  • blk.25.ffn_norm.weight
    F32
    [4096]
  • blk.25.ffn_up.weight
    Q4_0
    [4096, 27392]
    • blk.26
  • blk.26.attn_norm.weight
    F32
    [4096]
  • blk.26.attn_output.weight
    Q4_0
    [4096, 4096]
  • blk.26.attn_qkv.bias
    F32
    [4608]
  • blk.26.attn_qkv.weight
    Q4_0
    [4096, 4608]
  • blk.26.ffn_down.weight
    Q4_0
    [13696, 4096]
  • blk.26.ffn_norm.weight
    F32
    [4096]
  • blk.26.ffn_up.weight
    Q4_0
    [4096, 27392]
    • blk.27
  • blk.27.attn_norm.weight
    F32
    [4096]
  • blk.27.attn_output.weight
    Q4_0
    [4096, 4096]
  • blk.27.attn_qkv.bias
    F32
    [4608]
  • blk.27.attn_qkv.weight
    Q4_0
    [4096, 4608]
  • blk.27.ffn_down.weight
    Q4_0
    [13696, 4096]
  • blk.27.ffn_norm.weight
    F32
    [4096]
  • blk.27.ffn_up.weight
    Q4_0
    [4096, 27392]
  • output.weight
    Q6_K
    [4096, 65024]
  • output_norm.weight
    F32
    [4096]