mannix / qwen2-57b

Mixture-of-Experts model 57b

3,659 3 months ago

18 Tags
... /
model
6f39cba0c240 · 43GB
    Metadata
  • general.architecture
    qwen2moe
  • general.file_type
    9
  • general.finetune
    ..
  • general.languages
    [en]
  • general.license
    apache-2.0
  • general.name
    ..
  • general.quantization_version
    2
  • general.size_label
    64x8.9B
  • general.tags
    [chat, text-generation]
  • general.type
    model
  • quantize.imatrix.chunks_count
    64
  • quantize.imatrix.dataset
    /shared/opt/work_models/_imatrix/calibration_datav3.txt
  • quantize.imatrix.entries_count
    309
  • quantize.imatrix.file
    imatrix.dat
  • qwen2moe.attention.head_count
    28
  • qwen2moe.attention.head_count_kv
    4
  • qwen2moe.attention.layer_norm_rms_epsilon
    1e-06
  • qwen2moe.block_count
    28
  • qwen2moe.context_length
    32768
  • qwen2moe.embedding_length
    3584
  • qwen2moe.expert_count
    64
  • qwen2moe.expert_feed_forward_length
    2560
  • qwen2moe.expert_shared_feed_forward_length
    20480
  • qwen2moe.expert_used_count
    8
  • qwen2moe.feed_forward_length
    18944
  • qwen2moe.rope.freq_base
    1e+06
  • tokenizer.ggml.bos_token_id
    151643
  • tokenizer.ggml.eos_token_id
    151645
  • tokenizer.ggml.merges
    [Ġ Ġ, ĠĠ ĠĠ, i n, Ġ t, ĠĠĠĠ ĠĠĠĠ, ...]
  • tokenizer.ggml.model
    gpt2
  • tokenizer.ggml.padding_token_id
    151643
  • tokenizer.ggml.pre
    qwen2
  • tokenizer.ggml.token_type
    [1, 1, 1, 1, 1, ...]
  • tokenizer.ggml.tokens
    [!, ", #, $, %, ...]
    • Tensor
  • Name
    Type
    Shape
  • token_embd.weight
    Q5_1
    [3584, 151936]
    • blk.0
  • blk.0.attn_k.bias
    F32
    [512]
  • blk.0.attn_k.weight
    Q5_1
    [3584, 512]
  • blk.0.attn_norm.weight
    F32
    [3584]
  • blk.0.attn_output.weight
    Q5_1
    [3584, 3584]
  • blk.0.attn_q.bias
    F32
    [3584]
  • blk.0.attn_q.weight
    Q5_1
    [3584, 3584]
  • blk.0.attn_v.bias
    F32
    [512]
  • blk.0.attn_v.weight
    Q5_1
    [3584, 512]
  • blk.0.ffn_down_exps.weight
    Q5_1
    [2560, 3584, 64]
  • blk.0.ffn_down_shexp.weight
    Q5_1
    [20480, 3584]
  • blk.0.ffn_gate_exps.weight
    Q5_1
    [3584, 2560, 64]
  • blk.0.ffn_gate_inp.weight
    F32
    [3584, 64]
  • blk.0.ffn_gate_inp_shexp.weight
    F32
    [3584]
  • blk.0.ffn_gate_shexp.weight
    Q5_1
    [3584, 20480]
  • blk.0.ffn_norm.weight
    F32
    [3584]
  • blk.0.ffn_up_exps.weight
    Q5_1
    [3584, 2560, 64]
  • blk.0.ffn_up_shexp.weight
    Q5_1
    [3584, 20480]
    • blk.1
  • blk.1.attn_k.bias
    F32
    [512]
  • blk.1.attn_k.weight
    Q5_1
    [3584, 512]
  • blk.1.attn_norm.weight
    F32
    [3584]
  • blk.1.attn_output.weight
    Q5_1
    [3584, 3584]
  • blk.1.attn_q.bias
    F32
    [3584]
  • blk.1.attn_q.weight
    Q5_1
    [3584, 3584]
  • blk.1.attn_v.bias
    F32
    [512]
  • blk.1.attn_v.weight
    Q5_1
    [3584, 512]
  • blk.1.ffn_down_exps.weight
    Q5_1
    [2560, 3584, 64]
  • blk.1.ffn_down_shexp.weight
    Q5_1
    [20480, 3584]
  • blk.1.ffn_gate_exps.weight
    Q5_1
    [3584, 2560, 64]
  • blk.1.ffn_gate_inp.weight
    F32
    [3584, 64]
  • blk.1.ffn_gate_inp_shexp.weight
    F32
    [3584]
  • blk.1.ffn_gate_shexp.weight
    Q5_1
    [3584, 20480]
  • blk.1.ffn_norm.weight
    F32
    [3584]
  • blk.1.ffn_up_exps.weight
    Q5_1
    [3584, 2560, 64]
  • blk.1.ffn_up_shexp.weight
    Q5_1
    [3584, 20480]
    • blk.2
  • blk.2.attn_k.bias
    F32
    [512]
  • blk.2.attn_k.weight
    Q5_1
    [3584, 512]
  • blk.2.attn_norm.weight
    F32
    [3584]
  • blk.2.attn_output.weight
    Q5_1
    [3584, 3584]
  • blk.2.attn_q.bias
    F32
    [3584]
  • blk.2.attn_q.weight
    Q5_1
    [3584, 3584]
  • blk.2.attn_v.bias
    F32
    [512]
  • blk.2.attn_v.weight
    Q5_1
    [3584, 512]
  • blk.2.ffn_down_exps.weight
    Q5_1
    [2560, 3584, 64]
  • blk.2.ffn_down_shexp.weight
    Q5_1
    [20480, 3584]
  • blk.2.ffn_gate_exps.weight
    Q5_1
    [3584, 2560, 64]
  • blk.2.ffn_gate_inp.weight
    F32
    [3584, 64]
  • blk.2.ffn_gate_inp_shexp.weight
    F32
    [3584]
  • blk.2.ffn_gate_shexp.weight
    Q5_1
    [3584, 20480]
  • blk.2.ffn_norm.weight
    F32
    [3584]
  • blk.2.ffn_up_exps.weight
    Q5_1
    [3584, 2560, 64]
  • blk.2.ffn_up_shexp.weight
    Q5_1
    [3584, 20480]
    • blk.3
  • blk.3.attn_k.bias
    F32
    [512]
  • blk.3.attn_k.weight
    Q5_1
    [3584, 512]
  • blk.3.attn_norm.weight
    F32
    [3584]
  • blk.3.attn_output.weight
    Q5_1
    [3584, 3584]
  • blk.3.attn_q.bias
    F32
    [3584]
  • blk.3.attn_q.weight
    Q5_1
    [3584, 3584]
  • blk.3.attn_v.bias
    F32
    [512]
  • blk.3.attn_v.weight
    Q5_1
    [3584, 512]
  • blk.3.ffn_down_exps.weight
    Q5_1
    [2560, 3584, 64]
  • blk.3.ffn_down_shexp.weight
    Q5_1
    [20480, 3584]
  • blk.3.ffn_gate_exps.weight
    Q5_1
    [3584, 2560, 64]
  • blk.3.ffn_gate_inp.weight
    F32
    [3584, 64]
  • blk.3.ffn_gate_inp_shexp.weight
    F32
    [3584]
  • blk.3.ffn_gate_shexp.weight
    Q5_1
    [3584, 20480]
  • blk.3.ffn_norm.weight
    F32
    [3584]
  • blk.3.ffn_up_exps.weight
    Q5_1
    [3584, 2560, 64]
  • blk.3.ffn_up_shexp.weight
    Q5_1
    [3584, 20480]
    • blk.4
  • blk.4.attn_k.bias
    F32
    [512]
  • blk.4.attn_k.weight
    Q5_1
    [3584, 512]
  • blk.4.attn_norm.weight
    F32
    [3584]
  • blk.4.attn_output.weight
    Q5_1
    [3584, 3584]
  • blk.4.attn_q.bias
    F32
    [3584]
  • blk.4.attn_q.weight
    Q5_1
    [3584, 3584]
  • blk.4.attn_v.bias
    F32
    [512]
  • blk.4.attn_v.weight
    Q5_1
    [3584, 512]
  • blk.4.ffn_down_exps.weight
    Q5_1
    [2560, 3584, 64]
  • blk.4.ffn_down_shexp.weight
    Q5_1
    [20480, 3584]
  • blk.4.ffn_gate_exps.weight
    Q5_1
    [3584, 2560, 64]
  • blk.4.ffn_gate_inp.weight
    F32
    [3584, 64]
  • blk.4.ffn_gate_inp_shexp.weight
    F32
    [3584]
  • blk.4.ffn_gate_shexp.weight
    Q5_1
    [3584, 20480]
  • blk.4.ffn_norm.weight
    F32
    [3584]
  • blk.4.ffn_up_exps.weight
    Q5_1
    [3584, 2560, 64]
  • blk.4.ffn_up_shexp.weight
    Q5_1
    [3584, 20480]
    • blk.5
  • blk.5.attn_k.bias
    F32
    [512]
  • blk.5.attn_k.weight
    Q5_1
    [3584, 512]
  • blk.5.attn_norm.weight
    F32
    [3584]
  • blk.5.attn_output.weight
    Q5_1
    [3584, 3584]
  • blk.5.attn_q.bias
    F32
    [3584]
  • blk.5.attn_q.weight
    Q5_1
    [3584, 3584]
  • blk.5.attn_v.bias
    F32
    [512]
  • blk.5.attn_v.weight
    Q5_1
    [3584, 512]
  • blk.5.ffn_down_exps.weight
    Q5_1
    [2560, 3584, 64]
  • blk.5.ffn_down_shexp.weight
    Q5_1
    [20480, 3584]
  • blk.5.ffn_gate_exps.weight
    Q5_1
    [3584, 2560, 64]
  • blk.5.ffn_gate_inp.weight
    F32
    [3584, 64]
  • blk.5.ffn_gate_inp_shexp.weight
    F32
    [3584]
  • blk.5.ffn_gate_shexp.weight
    Q5_1
    [3584, 20480]
  • blk.5.ffn_norm.weight
    F32
    [3584]
  • blk.5.ffn_up_exps.weight
    Q5_1
    [3584, 2560, 64]
  • blk.5.ffn_up_shexp.weight
    Q5_1
    [3584, 20480]
    • blk.6
  • blk.6.attn_k.bias
    F32
    [512]
  • blk.6.attn_k.weight
    Q5_1
    [3584, 512]
  • blk.6.attn_norm.weight
    F32
    [3584]
  • blk.6.attn_output.weight
    Q5_1
    [3584, 3584]
  • blk.6.attn_q.bias
    F32
    [3584]
  • blk.6.attn_q.weight
    Q5_1
    [3584, 3584]
  • blk.6.attn_v.bias
    F32
    [512]
  • blk.6.attn_v.weight
    Q5_1
    [3584, 512]
  • blk.6.ffn_down_exps.weight
    Q5_1
    [2560, 3584, 64]
  • blk.6.ffn_down_shexp.weight
    Q5_1
    [20480, 3584]
  • blk.6.ffn_gate_exps.weight
    Q5_1
    [3584, 2560, 64]
  • blk.6.ffn_gate_inp.weight
    F32
    [3584, 64]
  • blk.6.ffn_gate_inp_shexp.weight
    F32
    [3584]
  • blk.6.ffn_gate_shexp.weight
    Q5_1
    [3584, 20480]
  • blk.6.ffn_norm.weight
    F32
    [3584]
  • blk.6.ffn_up_exps.weight
    Q5_1
    [3584, 2560, 64]
  • blk.6.ffn_up_shexp.weight
    Q5_1
    [3584, 20480]
    • blk.7
  • blk.7.attn_k.bias
    F32
    [512]
  • blk.7.attn_k.weight
    Q5_1
    [3584, 512]
  • blk.7.attn_norm.weight
    F32
    [3584]
  • blk.7.attn_output.weight
    Q5_1
    [3584, 3584]
  • blk.7.attn_q.bias
    F32
    [3584]
  • blk.7.attn_q.weight
    Q5_1
    [3584, 3584]
  • blk.7.attn_v.bias
    F32
    [512]
  • blk.7.attn_v.weight
    Q5_1
    [3584, 512]
  • blk.7.ffn_down_exps.weight
    Q5_1
    [2560, 3584, 64]
  • blk.7.ffn_down_shexp.weight
    Q5_1
    [20480, 3584]
  • blk.7.ffn_gate_exps.weight
    Q5_1
    [3584, 2560, 64]
  • blk.7.ffn_gate_inp.weight
    F32
    [3584, 64]
  • blk.7.ffn_gate_inp_shexp.weight
    F32
    [3584]
  • blk.7.ffn_gate_shexp.weight
    Q5_1
    [3584, 20480]
  • blk.7.ffn_norm.weight
    F32
    [3584]
  • blk.7.ffn_up_exps.weight
    Q5_1
    [3584, 2560, 64]
  • blk.7.ffn_up_shexp.weight
    Q5_1
    [3584, 20480]
    • blk.8
  • blk.8.attn_k.bias
    F32
    [512]
  • blk.8.attn_k.weight
    Q5_1
    [3584, 512]
  • blk.8.attn_norm.weight
    F32
    [3584]
  • blk.8.attn_output.weight
    Q5_1
    [3584, 3584]
  • blk.8.attn_q.bias
    F32
    [3584]
  • blk.8.attn_q.weight
    Q5_1
    [3584, 3584]
  • blk.8.attn_v.bias
    F32
    [512]
  • blk.8.attn_v.weight
    Q5_1
    [3584, 512]
  • blk.8.ffn_down_exps.weight
    Q5_1
    [2560, 3584, 64]
  • blk.8.ffn_down_shexp.weight
    Q5_1
    [20480, 3584]
  • blk.8.ffn_gate_exps.weight
    Q5_1
    [3584, 2560, 64]
  • blk.8.ffn_gate_inp.weight
    F32
    [3584, 64]
  • blk.8.ffn_gate_inp_shexp.weight
    F32
    [3584]
  • blk.8.ffn_gate_shexp.weight
    Q5_1
    [3584, 20480]
  • blk.8.ffn_norm.weight
    F32
    [3584]
  • blk.8.ffn_up_exps.weight
    Q5_1
    [3584, 2560, 64]
  • blk.8.ffn_up_shexp.weight
    Q5_1
    [3584, 20480]
    • blk.9
  • blk.9.attn_k.bias
    F32
    [512]
  • blk.9.attn_k.weight
    Q5_1
    [3584, 512]
  • blk.9.attn_norm.weight
    F32
    [3584]
  • blk.9.attn_output.weight
    Q5_1
    [3584, 3584]
  • blk.9.attn_q.bias
    F32
    [3584]
  • blk.9.attn_q.weight
    Q5_1
    [3584, 3584]
  • blk.9.attn_v.bias
    F32
    [512]
  • blk.9.attn_v.weight
    Q5_1
    [3584, 512]
  • blk.9.ffn_down_exps.weight
    Q5_1
    [2560, 3584, 64]
  • blk.9.ffn_down_shexp.weight
    Q5_1
    [20480, 3584]
  • blk.9.ffn_gate_exps.weight
    Q5_1
    [3584, 2560, 64]
  • blk.9.ffn_gate_inp.weight
    F32
    [3584, 64]
  • blk.9.ffn_gate_inp_shexp.weight
    F32
    [3584]
  • blk.9.ffn_gate_shexp.weight
    Q5_1
    [3584, 20480]
  • blk.9.ffn_norm.weight
    F32
    [3584]
  • blk.9.ffn_up_exps.weight
    Q5_1
    [3584, 2560, 64]
  • blk.9.ffn_up_shexp.weight
    Q5_1
    [3584, 20480]
    • blk.10
  • blk.10.attn_k.bias
    F32
    [512]
  • blk.10.attn_k.weight
    Q5_1
    [3584, 512]
  • blk.10.attn_norm.weight
    F32
    [3584]
  • blk.10.attn_output.weight
    Q5_1
    [3584, 3584]
  • blk.10.attn_q.bias
    F32
    [3584]
  • blk.10.attn_q.weight
    Q5_1
    [3584, 3584]
  • blk.10.attn_v.bias
    F32
    [512]
  • blk.10.attn_v.weight
    Q5_1
    [3584, 512]
  • blk.10.ffn_down_exps.weight
    Q5_1
    [2560, 3584, 64]
  • blk.10.ffn_down_shexp.weight
    Q5_1
    [20480, 3584]
  • blk.10.ffn_gate_exps.weight
    Q5_1
    [3584, 2560, 64]
  • blk.10.ffn_gate_inp.weight
    F32
    [3584, 64]
  • blk.10.ffn_gate_inp_shexp.weight
    F32
    [3584]
  • blk.10.ffn_gate_shexp.weight
    Q5_1
    [3584, 20480]
  • blk.10.ffn_norm.weight
    F32
    [3584]
  • blk.10.ffn_up_exps.weight
    Q5_1
    [3584, 2560, 64]
  • blk.10.ffn_up_shexp.weight
    Q5_1
    [3584, 20480]
    • blk.11
  • blk.11.attn_k.bias
    F32
    [512]
  • blk.11.attn_k.weight
    Q5_1
    [3584, 512]
  • blk.11.attn_norm.weight
    F32
    [3584]
  • blk.11.attn_output.weight
    Q5_1
    [3584, 3584]
  • blk.11.attn_q.bias
    F32
    [3584]
  • blk.11.attn_q.weight
    Q5_1
    [3584, 3584]
  • blk.11.attn_v.bias
    F32
    [512]
  • blk.11.attn_v.weight
    Q5_1
    [3584, 512]
  • blk.11.ffn_down_exps.weight
    Q5_1
    [2560, 3584, 64]
  • blk.11.ffn_down_shexp.weight
    Q5_1
    [20480, 3584]
  • blk.11.ffn_gate_exps.weight
    Q5_1
    [3584, 2560, 64]
  • blk.11.ffn_gate_inp.weight
    F32
    [3584, 64]
  • blk.11.ffn_gate_inp_shexp.weight
    F32
    [3584]
  • blk.11.ffn_gate_shexp.weight
    Q5_1
    [3584, 20480]
  • blk.11.ffn_norm.weight
    F32
    [3584]
  • blk.11.ffn_up_exps.weight
    Q5_1
    [3584, 2560, 64]
  • blk.11.ffn_up_shexp.weight
    Q5_1
    [3584, 20480]
    • blk.12
  • blk.12.attn_k.bias
    F32
    [512]
  • blk.12.attn_k.weight
    Q5_1
    [3584, 512]
  • blk.12.attn_norm.weight
    F32
    [3584]
  • blk.12.attn_output.weight
    Q5_1
    [3584, 3584]
  • blk.12.attn_q.bias
    F32
    [3584]
  • blk.12.attn_q.weight
    Q5_1
    [3584, 3584]
  • blk.12.attn_v.bias
    F32
    [512]
  • blk.12.attn_v.weight
    Q5_1
    [3584, 512]
  • blk.12.ffn_down_exps.weight
    Q5_1
    [2560, 3584, 64]
  • blk.12.ffn_down_shexp.weight
    Q5_1
    [20480, 3584]
  • blk.12.ffn_gate_exps.weight
    Q5_1
    [3584, 2560, 64]
  • blk.12.ffn_gate_inp.weight
    F32
    [3584, 64]
  • blk.12.ffn_gate_inp_shexp.weight
    F32
    [3584]
  • blk.12.ffn_gate_shexp.weight
    Q5_1
    [3584, 20480]
  • blk.12.ffn_norm.weight
    F32
    [3584]
  • blk.12.ffn_up_exps.weight
    Q5_1
    [3584, 2560, 64]
  • blk.12.ffn_up_shexp.weight
    Q5_1
    [3584, 20480]
    • blk.13
  • blk.13.attn_k.bias
    F32
    [512]
  • blk.13.attn_k.weight
    Q5_1
    [3584, 512]
  • blk.13.attn_norm.weight
    F32
    [3584]
  • blk.13.attn_output.weight
    Q5_1
    [3584, 3584]
  • blk.13.attn_q.bias
    F32
    [3584]
  • blk.13.attn_q.weight
    Q5_1
    [3584, 3584]
  • blk.13.attn_v.bias
    F32
    [512]
  • blk.13.attn_v.weight
    Q5_1
    [3584, 512]
  • blk.13.ffn_down_exps.weight
    Q5_1
    [2560, 3584, 64]
  • blk.13.ffn_down_shexp.weight
    Q5_1
    [20480, 3584]
  • blk.13.ffn_gate_exps.weight
    Q5_1
    [3584, 2560, 64]
  • blk.13.ffn_gate_inp.weight
    F32
    [3584, 64]
  • blk.13.ffn_gate_inp_shexp.weight
    F32
    [3584]
  • blk.13.ffn_gate_shexp.weight
    Q5_1
    [3584, 20480]
  • blk.13.ffn_norm.weight
    F32
    [3584]
  • blk.13.ffn_up_exps.weight
    Q5_1
    [3584, 2560, 64]
  • blk.13.ffn_up_shexp.weight
    Q5_1
    [3584, 20480]
    • blk.14
  • blk.14.attn_k.bias
    F32
    [512]
  • blk.14.attn_k.weight
    Q5_1
    [3584, 512]
  • blk.14.attn_norm.weight
    F32
    [3584]
  • blk.14.attn_output.weight
    Q5_1
    [3584, 3584]
  • blk.14.attn_q.bias
    F32
    [3584]
  • blk.14.attn_q.weight
    Q5_1
    [3584, 3584]
  • blk.14.attn_v.bias
    F32
    [512]
  • blk.14.attn_v.weight
    Q5_1
    [3584, 512]
  • blk.14.ffn_down_exps.weight
    Q5_1
    [2560, 3584, 64]
  • blk.14.ffn_down_shexp.weight
    Q5_1
    [20480, 3584]
  • blk.14.ffn_gate_exps.weight
    Q5_1
    [3584, 2560, 64]
  • blk.14.ffn_gate_inp.weight
    F32
    [3584, 64]
  • blk.14.ffn_gate_inp_shexp.weight
    F32
    [3584]
  • blk.14.ffn_gate_shexp.weight
    Q5_1
    [3584, 20480]
  • blk.14.ffn_norm.weight
    F32
    [3584]
  • blk.14.ffn_up_exps.weight
    Q5_1
    [3584, 2560, 64]
  • blk.14.ffn_up_shexp.weight
    Q5_1
    [3584, 20480]
    • blk.15
  • blk.15.attn_k.bias
    F32
    [512]
  • blk.15.attn_k.weight
    Q5_1
    [3584, 512]
  • blk.15.attn_norm.weight
    F32
    [3584]
  • blk.15.attn_output.weight
    Q5_1
    [3584, 3584]
  • blk.15.attn_q.bias
    F32
    [3584]
  • blk.15.attn_q.weight
    Q5_1
    [3584, 3584]
  • blk.15.attn_v.bias
    F32
    [512]
  • blk.15.attn_v.weight
    Q5_1
    [3584, 512]
  • blk.15.ffn_down_exps.weight
    Q5_1
    [2560, 3584, 64]
  • blk.15.ffn_down_shexp.weight
    Q5_1
    [20480, 3584]
  • blk.15.ffn_gate_exps.weight
    Q5_1
    [3584, 2560, 64]
  • blk.15.ffn_gate_inp.weight
    F32
    [3584, 64]
  • blk.15.ffn_gate_inp_shexp.weight
    F32
    [3584]
  • blk.15.ffn_gate_shexp.weight
    Q5_1
    [3584, 20480]
  • blk.15.ffn_norm.weight
    F32
    [3584]
  • blk.15.ffn_up_exps.weight
    Q5_1
    [3584, 2560, 64]
  • blk.15.ffn_up_shexp.weight
    Q5_1
    [3584, 20480]
    • blk.16
  • blk.16.attn_k.bias
    F32
    [512]
  • blk.16.attn_k.weight
    Q5_1
    [3584, 512]
  • blk.16.attn_norm.weight
    F32
    [3584]
  • blk.16.attn_output.weight
    Q5_1
    [3584, 3584]
  • blk.16.attn_q.bias
    F32
    [3584]
  • blk.16.attn_q.weight
    Q5_1
    [3584, 3584]
  • blk.16.attn_v.bias
    F32
    [512]
  • blk.16.attn_v.weight
    Q5_1
    [3584, 512]
  • blk.16.ffn_down_exps.weight
    Q5_1
    [2560, 3584, 64]
  • blk.16.ffn_down_shexp.weight
    Q5_1
    [20480, 3584]
  • blk.16.ffn_gate_exps.weight
    Q5_1
    [3584, 2560, 64]
  • blk.16.ffn_gate_inp.weight
    F32
    [3584, 64]
  • blk.16.ffn_gate_inp_shexp.weight
    F32
    [3584]
  • blk.16.ffn_gate_shexp.weight
    Q5_1
    [3584, 20480]
  • blk.16.ffn_norm.weight
    F32
    [3584]
  • blk.16.ffn_up_exps.weight
    Q5_1
    [3584, 2560, 64]
  • blk.16.ffn_up_shexp.weight
    Q5_1
    [3584, 20480]
    • blk.17
  • blk.17.attn_k.bias
    F32
    [512]
  • blk.17.attn_k.weight
    Q5_1
    [3584, 512]
  • blk.17.attn_norm.weight
    F32
    [3584]
  • blk.17.attn_output.weight
    Q5_1
    [3584, 3584]
  • blk.17.attn_q.bias
    F32
    [3584]
  • blk.17.attn_q.weight
    Q5_1
    [3584, 3584]
  • blk.17.attn_v.bias
    F32
    [512]
  • blk.17.attn_v.weight
    Q5_1
    [3584, 512]
  • blk.17.ffn_down_exps.weight
    Q5_1
    [2560, 3584, 64]
  • blk.17.ffn_down_shexp.weight
    Q5_1
    [20480, 3584]
  • blk.17.ffn_gate_exps.weight
    Q5_1
    [3584, 2560, 64]
  • blk.17.ffn_gate_inp.weight
    F32
    [3584, 64]
  • blk.17.ffn_gate_inp_shexp.weight
    F32
    [3584]
  • blk.17.ffn_gate_shexp.weight
    Q5_1
    [3584, 20480]
  • blk.17.ffn_norm.weight
    F32
    [3584]
  • blk.17.ffn_up_exps.weight
    Q5_1
    [3584, 2560, 64]
  • blk.17.ffn_up_shexp.weight
    Q5_1
    [3584, 20480]
    • blk.18
  • blk.18.attn_k.bias
    F32
    [512]
  • blk.18.attn_k.weight
    Q5_1
    [3584, 512]
  • blk.18.attn_norm.weight
    F32
    [3584]
  • blk.18.attn_output.weight
    Q5_1
    [3584, 3584]
  • blk.18.attn_q.bias
    F32
    [3584]
  • blk.18.attn_q.weight
    Q5_1
    [3584, 3584]
  • blk.18.attn_v.bias
    F32
    [512]
  • blk.18.attn_v.weight
    Q5_1
    [3584, 512]
  • blk.18.ffn_down_exps.weight
    Q5_1
    [2560, 3584, 64]
  • blk.18.ffn_down_shexp.weight
    Q5_1
    [20480, 3584]
  • blk.18.ffn_gate_exps.weight
    Q5_1
    [3584, 2560, 64]
  • blk.18.ffn_gate_inp.weight
    F32
    [3584, 64]
  • blk.18.ffn_gate_inp_shexp.weight
    F32
    [3584]
  • blk.18.ffn_gate_shexp.weight
    Q5_1
    [3584, 20480]
  • blk.18.ffn_norm.weight
    F32
    [3584]
  • blk.18.ffn_up_exps.weight
    Q5_1
    [3584, 2560, 64]
  • blk.18.ffn_up_shexp.weight
    Q5_1
    [3584, 20480]
    • blk.19
  • blk.19.attn_k.bias
    F32
    [512]
  • blk.19.attn_k.weight
    Q5_1
    [3584, 512]
  • blk.19.attn_norm.weight
    F32
    [3584]
  • blk.19.attn_output.weight
    Q5_1
    [3584, 3584]
  • blk.19.attn_q.bias
    F32
    [3584]
  • blk.19.attn_q.weight
    Q5_1
    [3584, 3584]
  • blk.19.attn_v.bias
    F32
    [512]
  • blk.19.attn_v.weight
    Q5_1
    [3584, 512]
  • blk.19.ffn_down_exps.weight
    Q5_1
    [2560, 3584, 64]
  • blk.19.ffn_down_shexp.weight
    Q5_1
    [20480, 3584]
  • blk.19.ffn_gate_exps.weight
    Q5_1
    [3584, 2560, 64]
  • blk.19.ffn_gate_inp.weight
    F32
    [3584, 64]
  • blk.19.ffn_gate_inp_shexp.weight
    F32
    [3584]
  • blk.19.ffn_gate_shexp.weight
    Q5_1
    [3584, 20480]
  • blk.19.ffn_norm.weight
    F32
    [3584]
  • blk.19.ffn_up_exps.weight
    Q5_1
    [3584, 2560, 64]
  • blk.19.ffn_up_shexp.weight
    Q5_1
    [3584, 20480]
    • blk.20
  • blk.20.attn_k.bias
    F32
    [512]
  • blk.20.attn_k.weight
    Q5_1
    [3584, 512]
  • blk.20.attn_norm.weight
    F32
    [3584]
  • blk.20.attn_output.weight
    Q5_1
    [3584, 3584]
  • blk.20.attn_q.bias
    F32
    [3584]
  • blk.20.attn_q.weight
    Q5_1
    [3584, 3584]
  • blk.20.attn_v.bias
    F32
    [512]
  • blk.20.attn_v.weight
    Q5_1
    [3584, 512]
  • blk.20.ffn_down_exps.weight
    Q5_1
    [2560, 3584, 64]
  • blk.20.ffn_down_shexp.weight
    Q5_1
    [20480, 3584]
  • blk.20.ffn_gate_exps.weight
    Q5_1
    [3584, 2560, 64]
  • blk.20.ffn_gate_inp.weight
    F32
    [3584, 64]
  • blk.20.ffn_gate_inp_shexp.weight
    F32
    [3584]
  • blk.20.ffn_gate_shexp.weight
    Q5_1
    [3584, 20480]
  • blk.20.ffn_norm.weight
    F32
    [3584]
  • blk.20.ffn_up_exps.weight
    Q5_1
    [3584, 2560, 64]
  • blk.20.ffn_up_shexp.weight
    Q5_1
    [3584, 20480]
    • blk.21
  • blk.21.attn_k.bias
    F32
    [512]
  • blk.21.attn_k.weight
    Q5_1
    [3584, 512]
  • blk.21.attn_norm.weight
    F32
    [3584]
  • blk.21.attn_output.weight
    Q5_1
    [3584, 3584]
  • blk.21.attn_q.bias
    F32
    [3584]
  • blk.21.attn_q.weight
    Q5_1
    [3584, 3584]
  • blk.21.attn_v.bias
    F32
    [512]
  • blk.21.attn_v.weight
    Q5_1
    [3584, 512]
  • blk.21.ffn_down_exps.weight
    Q5_1
    [2560, 3584, 64]
  • blk.21.ffn_down_shexp.weight
    Q5_1
    [20480, 3584]
  • blk.21.ffn_gate_exps.weight
    Q5_1
    [3584, 2560, 64]
  • blk.21.ffn_gate_inp.weight
    F32
    [3584, 64]
  • blk.21.ffn_gate_inp_shexp.weight
    F32
    [3584]
  • blk.21.ffn_gate_shexp.weight
    Q5_1
    [3584, 20480]
  • blk.21.ffn_norm.weight
    F32
    [3584]
  • blk.21.ffn_up_exps.weight
    Q5_1
    [3584, 2560, 64]
  • blk.21.ffn_up_shexp.weight
    Q5_1
    [3584, 20480]
    • blk.22
  • blk.22.attn_k.bias
    F32
    [512]
  • blk.22.attn_k.weight
    Q5_1
    [3584, 512]
  • blk.22.attn_norm.weight
    F32
    [3584]
  • blk.22.attn_output.weight
    Q5_1
    [3584, 3584]
  • blk.22.attn_q.bias
    F32
    [3584]
  • blk.22.attn_q.weight
    Q5_1
    [3584, 3584]
  • blk.22.attn_v.bias
    F32
    [512]
  • blk.22.attn_v.weight
    Q5_1
    [3584, 512]
  • blk.22.ffn_down_exps.weight
    Q5_1
    [2560, 3584, 64]
  • blk.22.ffn_down_shexp.weight
    Q5_1
    [20480, 3584]
  • blk.22.ffn_gate_exps.weight
    Q5_1
    [3584, 2560, 64]
  • blk.22.ffn_gate_inp.weight
    F32
    [3584, 64]
  • blk.22.ffn_gate_inp_shexp.weight
    F32
    [3584]
  • blk.22.ffn_gate_shexp.weight
    Q5_1
    [3584, 20480]
  • blk.22.ffn_norm.weight
    F32
    [3584]
  • blk.22.ffn_up_exps.weight
    Q5_1
    [3584, 2560, 64]
  • blk.22.ffn_up_shexp.weight
    Q5_1
    [3584, 20480]
    • blk.23
  • blk.23.attn_k.bias
    F32
    [512]
  • blk.23.attn_k.weight
    Q5_1
    [3584, 512]
  • blk.23.attn_norm.weight
    F32
    [3584]
  • blk.23.attn_output.weight
    Q5_1
    [3584, 3584]
  • blk.23.attn_q.bias
    F32
    [3584]
  • blk.23.attn_q.weight
    Q5_1
    [3584, 3584]
  • blk.23.attn_v.bias
    F32
    [512]
  • blk.23.attn_v.weight
    Q5_1
    [3584, 512]
  • blk.23.ffn_down_exps.weight
    Q5_1
    [2560, 3584, 64]
  • blk.23.ffn_down_shexp.weight
    Q5_1
    [20480, 3584]
  • blk.23.ffn_gate_exps.weight
    Q5_1
    [3584, 2560, 64]
  • blk.23.ffn_gate_inp.weight
    F32
    [3584, 64]
  • blk.23.ffn_gate_inp_shexp.weight
    F32
    [3584]
  • blk.23.ffn_gate_shexp.weight
    Q5_1
    [3584, 20480]
  • blk.23.ffn_norm.weight
    F32
    [3584]
  • blk.23.ffn_up_exps.weight
    Q5_1
    [3584, 2560, 64]
  • blk.23.ffn_up_shexp.weight
    Q5_1
    [3584, 20480]
    • blk.24
  • blk.24.attn_k.bias
    F32
    [512]
  • blk.24.attn_k.weight
    Q5_1
    [3584, 512]
  • blk.24.attn_norm.weight
    F32
    [3584]
  • blk.24.attn_output.weight
    Q5_1
    [3584, 3584]
  • blk.24.attn_q.bias
    F32
    [3584]
  • blk.24.attn_q.weight
    Q5_1
    [3584, 3584]
  • blk.24.attn_v.bias
    F32
    [512]
  • blk.24.attn_v.weight
    Q5_1
    [3584, 512]
  • blk.24.ffn_down_exps.weight
    Q5_1
    [2560, 3584, 64]
  • blk.24.ffn_down_shexp.weight
    Q5_1
    [20480, 3584]
  • blk.24.ffn_gate_exps.weight
    Q5_1
    [3584, 2560, 64]
  • blk.24.ffn_gate_inp.weight
    F32
    [3584, 64]
  • blk.24.ffn_gate_inp_shexp.weight
    F32
    [3584]
  • blk.24.ffn_gate_shexp.weight
    Q5_1
    [3584, 20480]
  • blk.24.ffn_norm.weight
    F32
    [3584]
  • blk.24.ffn_up_exps.weight
    Q5_1
    [3584, 2560, 64]
  • blk.24.ffn_up_shexp.weight
    Q5_1
    [3584, 20480]
    • blk.25
  • blk.25.attn_k.bias
    F32
    [512]
  • blk.25.attn_k.weight
    Q5_1
    [3584, 512]
  • blk.25.attn_norm.weight
    F32
    [3584]
  • blk.25.attn_output.weight
    Q5_1
    [3584, 3584]
  • blk.25.attn_q.bias
    F32
    [3584]
  • blk.25.attn_q.weight
    Q5_1
    [3584, 3584]
  • blk.25.attn_v.bias
    F32
    [512]
  • blk.25.attn_v.weight
    Q5_1
    [3584, 512]
  • blk.25.ffn_down_exps.weight
    Q5_1
    [2560, 3584, 64]
  • blk.25.ffn_down_shexp.weight
    Q5_1
    [20480, 3584]
  • blk.25.ffn_gate_exps.weight
    Q5_1
    [3584, 2560, 64]
  • blk.25.ffn_gate_inp.weight
    F32
    [3584, 64]
  • blk.25.ffn_gate_inp_shexp.weight
    F32
    [3584]
  • blk.25.ffn_gate_shexp.weight
    Q5_1
    [3584, 20480]
  • blk.25.ffn_norm.weight
    F32
    [3584]
  • blk.25.ffn_up_exps.weight
    Q5_1
    [3584, 2560, 64]
  • blk.25.ffn_up_shexp.weight
    Q5_1
    [3584, 20480]
    • blk.26
  • blk.26.attn_k.bias
    F32
    [512]
  • blk.26.attn_k.weight
    Q5_1
    [3584, 512]
  • blk.26.attn_norm.weight
    F32
    [3584]
  • blk.26.attn_output.weight
    Q5_1
    [3584, 3584]
  • blk.26.attn_q.bias
    F32
    [3584]
  • blk.26.attn_q.weight
    Q5_1
    [3584, 3584]
  • blk.26.attn_v.bias
    F32
    [512]
  • blk.26.attn_v.weight
    Q5_1
    [3584, 512]
  • blk.26.ffn_down_exps.weight
    Q5_1
    [2560, 3584, 64]
  • blk.26.ffn_down_shexp.weight
    Q5_1
    [20480, 3584]
  • blk.26.ffn_gate_exps.weight
    Q5_1
    [3584, 2560, 64]
  • blk.26.ffn_gate_inp.weight
    F32
    [3584, 64]
  • blk.26.ffn_gate_inp_shexp.weight
    F32
    [3584]
  • blk.26.ffn_gate_shexp.weight
    Q5_1
    [3584, 20480]
  • blk.26.ffn_norm.weight
    F32
    [3584]
  • blk.26.ffn_up_exps.weight
    Q5_1
    [3584, 2560, 64]
  • blk.26.ffn_up_shexp.weight
    Q5_1
    [3584, 20480]
    • blk.27
  • blk.27.attn_k.bias
    F32
    [512]
  • blk.27.attn_k.weight
    Q5_1
    [3584, 512]
  • blk.27.attn_norm.weight
    F32
    [3584]
  • blk.27.attn_output.weight
    Q5_1
    [3584, 3584]
  • blk.27.attn_q.bias
    F32
    [3584]
  • blk.27.attn_q.weight
    Q5_1
    [3584, 3584]
  • blk.27.attn_v.bias
    F32
    [512]
  • blk.27.attn_v.weight
    Q5_1
    [3584, 512]
  • blk.27.ffn_down_exps.weight
    Q5_1
    [2560, 3584, 64]
  • blk.27.ffn_down_shexp.weight
    Q5_1
    [20480, 3584]
  • blk.27.ffn_gate_exps.weight
    Q5_1
    [3584, 2560, 64]
  • blk.27.ffn_gate_inp.weight
    F32
    [3584, 64]
  • blk.27.ffn_gate_inp_shexp.weight
    F32
    [3584]
  • blk.27.ffn_gate_shexp.weight
    Q5_1
    [3584, 20480]
  • blk.27.ffn_norm.weight
    F32
    [3584]
  • blk.27.ffn_up_exps.weight
    Q5_1
    [3584, 2560, 64]
  • blk.27.ffn_up_shexp.weight
    Q5_1
    [3584, 20480]
  • output.weight
    Q6_K
    [3584, 151936]
  • output_norm.weight
    F32
    [3584]