free02/ glm5.1:part3

149 1 month ago

... /
model
e7ebd9f6df8c · 50GB
    Metadata
  • split.count
    6
  • split.no
    2
  • split.tensors.count
    1809
    • Tensor
    blk.18
  • blk.18.ffn_gate_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.18.ffn_gate_inp.weight
    F32
    [6144, 256]
  • blk.18.ffn_gate_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.18.ffn_norm.weight
    F32
    [6144]
  • blk.18.ffn_up_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.18.ffn_up_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.18.indexer.attn_k.weight
    Q8_0
    [6144, 128]
  • blk.18.indexer.attn_q_b.weight
    Q8_0
    [2048, 4096]
  • blk.18.indexer.k_norm.bias
    F32
    [128]
  • blk.18.indexer.k_norm.weight
    F32
    [128]
  • blk.18.indexer.proj.weight
    F32
    [6144, 32]
    • blk.19
  • blk.19.attn_k_b.weight
    Q8_0
    [192, 512, 64]
  • blk.19.attn_kv_a_mqa.weight
    Q8_0
    [6144, 576]
  • blk.19.attn_kv_a_norm.weight
    F32
    [512]
  • blk.19.attn_norm.weight
    F32
    [6144]
  • blk.19.attn_output.weight
    Q5_K
    [16384, 6144]
  • blk.19.attn_q_a.weight
    Q5_K
    [6144, 2048]
  • blk.19.attn_q_a_norm.weight
    F32
    [2048]
  • blk.19.attn_q_b.weight
    Q8_0
    [2048, 16384]
  • blk.19.attn_v_b.weight
    Q8_0
    [512, 256, 64]
  • blk.19.exp_probs_b.bias
    F32
    [256]
  • blk.19.ffn_down_exps.weight
    IQ3_XXS
    [2048, 6144, 256]
  • blk.19.ffn_down_shexp.weight
    Q6_K
    [2048, 6144]
  • blk.19.ffn_gate_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.19.ffn_gate_inp.weight
    F32
    [6144, 256]
  • blk.19.ffn_gate_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.19.ffn_norm.weight
    F32
    [6144]
  • blk.19.ffn_up_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.19.ffn_up_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.19.indexer.attn_k.weight
    Q8_0
    [6144, 128]
  • blk.19.indexer.attn_q_b.weight
    Q8_0
    [2048, 4096]
  • blk.19.indexer.k_norm.bias
    F32
    [128]
  • blk.19.indexer.k_norm.weight
    F32
    [128]
  • blk.19.indexer.proj.weight
    F32
    [6144, 32]
    • blk.20
  • blk.20.attn_k_b.weight
    Q8_0
    [192, 512, 64]
  • blk.20.attn_kv_a_mqa.weight
    Q8_0
    [6144, 576]
  • blk.20.attn_kv_a_norm.weight
    F32
    [512]
  • blk.20.attn_norm.weight
    F32
    [6144]
  • blk.20.attn_output.weight
    Q5_K
    [16384, 6144]
  • blk.20.attn_q_a.weight
    Q5_K
    [6144, 2048]
  • blk.20.attn_q_a_norm.weight
    F32
    [2048]
  • blk.20.attn_q_b.weight
    Q8_0
    [2048, 16384]
  • blk.20.attn_v_b.weight
    Q8_0
    [512, 256, 64]
  • blk.20.exp_probs_b.bias
    F32
    [256]
  • blk.20.ffn_down_exps.weight
    IQ3_XXS
    [2048, 6144, 256]
  • blk.20.ffn_down_shexp.weight
    Q6_K
    [2048, 6144]
  • blk.20.ffn_gate_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.20.ffn_gate_inp.weight
    F32
    [6144, 256]
  • blk.20.ffn_gate_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.20.ffn_norm.weight
    F32
    [6144]
  • blk.20.ffn_up_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.20.ffn_up_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.20.indexer.attn_k.weight
    Q8_0
    [6144, 128]
  • blk.20.indexer.attn_q_b.weight
    Q8_0
    [2048, 4096]
  • blk.20.indexer.k_norm.bias
    F32
    [128]
  • blk.20.indexer.k_norm.weight
    F32
    [128]
  • blk.20.indexer.proj.weight
    F32
    [6144, 32]
    • blk.21
  • blk.21.attn_k_b.weight
    Q8_0
    [192, 512, 64]
  • blk.21.attn_kv_a_mqa.weight
    Q8_0
    [6144, 576]
  • blk.21.attn_kv_a_norm.weight
    F32
    [512]
  • blk.21.attn_norm.weight
    F32
    [6144]
  • blk.21.attn_output.weight
    Q5_K
    [16384, 6144]
  • blk.21.attn_q_a.weight
    Q5_K
    [6144, 2048]
  • blk.21.attn_q_a_norm.weight
    F32
    [2048]
  • blk.21.attn_q_b.weight
    Q8_0
    [2048, 16384]
  • blk.21.attn_v_b.weight
    Q8_0
    [512, 256, 64]
  • blk.21.exp_probs_b.bias
    F32
    [256]
  • blk.21.ffn_down_exps.weight
    IQ3_XXS
    [2048, 6144, 256]
  • blk.21.ffn_down_shexp.weight
    Q6_K
    [2048, 6144]
  • blk.21.ffn_gate_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.21.ffn_gate_inp.weight
    F32
    [6144, 256]
  • blk.21.ffn_gate_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.21.ffn_norm.weight
    F32
    [6144]
  • blk.21.ffn_up_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.21.ffn_up_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.21.indexer.attn_k.weight
    Q8_0
    [6144, 128]
  • blk.21.indexer.attn_q_b.weight
    Q8_0
    [2048, 4096]
  • blk.21.indexer.k_norm.bias
    F32
    [128]
  • blk.21.indexer.k_norm.weight
    F32
    [128]
  • blk.21.indexer.proj.weight
    F32
    [6144, 32]
    • blk.22
  • blk.22.attn_k_b.weight
    Q8_0
    [192, 512, 64]
  • blk.22.attn_kv_a_mqa.weight
    Q8_0
    [6144, 576]
  • blk.22.attn_kv_a_norm.weight
    F32
    [512]
  • blk.22.attn_norm.weight
    F32
    [6144]
  • blk.22.attn_output.weight
    Q5_K
    [16384, 6144]
  • blk.22.attn_q_a.weight
    Q5_K
    [6144, 2048]
  • blk.22.attn_q_a_norm.weight
    F32
    [2048]
  • blk.22.attn_q_b.weight
    Q8_0
    [2048, 16384]
  • blk.22.attn_v_b.weight
    Q8_0
    [512, 256, 64]
  • blk.22.exp_probs_b.bias
    F32
    [256]
  • blk.22.ffn_down_exps.weight
    IQ3_XXS
    [2048, 6144, 256]
  • blk.22.ffn_down_shexp.weight
    Q6_K
    [2048, 6144]
  • blk.22.ffn_gate_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.22.ffn_gate_inp.weight
    F32
    [6144, 256]
  • blk.22.ffn_gate_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.22.ffn_norm.weight
    F32
    [6144]
  • blk.22.ffn_up_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.22.ffn_up_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.22.indexer.attn_k.weight
    Q8_0
    [6144, 128]
  • blk.22.indexer.attn_q_b.weight
    Q8_0
    [2048, 4096]
  • blk.22.indexer.k_norm.bias
    F32
    [128]
  • blk.22.indexer.k_norm.weight
    F32
    [128]
  • blk.22.indexer.proj.weight
    F32
    [6144, 32]
    • blk.23
  • blk.23.attn_k_b.weight
    Q8_0
    [192, 512, 64]
  • blk.23.attn_kv_a_mqa.weight
    Q8_0
    [6144, 576]
  • blk.23.attn_kv_a_norm.weight
    F32
    [512]
  • blk.23.attn_norm.weight
    F32
    [6144]
  • blk.23.attn_output.weight
    Q5_K
    [16384, 6144]
  • blk.23.attn_q_a.weight
    Q5_K
    [6144, 2048]
  • blk.23.attn_q_a_norm.weight
    F32
    [2048]
  • blk.23.attn_q_b.weight
    Q8_0
    [2048, 16384]
  • blk.23.attn_v_b.weight
    Q8_0
    [512, 256, 64]
  • blk.23.exp_probs_b.bias
    F32
    [256]
  • blk.23.ffn_down_exps.weight
    IQ3_XXS
    [2048, 6144, 256]
  • blk.23.ffn_down_shexp.weight
    Q6_K
    [2048, 6144]
  • blk.23.ffn_gate_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.23.ffn_gate_inp.weight
    F32
    [6144, 256]
  • blk.23.ffn_gate_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.23.ffn_norm.weight
    F32
    [6144]
  • blk.23.ffn_up_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.23.ffn_up_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.23.indexer.attn_k.weight
    Q8_0
    [6144, 128]
  • blk.23.indexer.attn_q_b.weight
    Q8_0
    [2048, 4096]
  • blk.23.indexer.k_norm.bias
    F32
    [128]
  • blk.23.indexer.k_norm.weight
    F32
    [128]
  • blk.23.indexer.proj.weight
    F32
    [6144, 32]
    • blk.24
  • blk.24.attn_k_b.weight
    Q8_0
    [192, 512, 64]
  • blk.24.attn_kv_a_mqa.weight
    Q8_0
    [6144, 576]
  • blk.24.attn_kv_a_norm.weight
    F32
    [512]
  • blk.24.attn_norm.weight
    F32
    [6144]
  • blk.24.attn_output.weight
    Q5_K
    [16384, 6144]
  • blk.24.attn_q_a.weight
    Q5_K
    [6144, 2048]
  • blk.24.attn_q_a_norm.weight
    F32
    [2048]
  • blk.24.attn_q_b.weight
    Q8_0
    [2048, 16384]
  • blk.24.attn_v_b.weight
    Q8_0
    [512, 256, 64]
  • blk.24.exp_probs_b.bias
    F32
    [256]
  • blk.24.ffn_down_exps.weight
    IQ3_XXS
    [2048, 6144, 256]
  • blk.24.ffn_down_shexp.weight
    Q6_K
    [2048, 6144]
  • blk.24.ffn_gate_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.24.ffn_gate_inp.weight
    F32
    [6144, 256]
  • blk.24.ffn_gate_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.24.ffn_norm.weight
    F32
    [6144]
  • blk.24.ffn_up_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.24.ffn_up_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.24.indexer.attn_k.weight
    Q8_0
    [6144, 128]
  • blk.24.indexer.attn_q_b.weight
    Q8_0
    [2048, 4096]
  • blk.24.indexer.k_norm.bias
    F32
    [128]
  • blk.24.indexer.k_norm.weight
    F32
    [128]
  • blk.24.indexer.proj.weight
    F32
    [6144, 32]
    • blk.25
  • blk.25.attn_k_b.weight
    Q8_0
    [192, 512, 64]
  • blk.25.attn_kv_a_mqa.weight
    Q8_0
    [6144, 576]
  • blk.25.attn_kv_a_norm.weight
    F32
    [512]
  • blk.25.attn_norm.weight
    F32
    [6144]
  • blk.25.attn_output.weight
    Q5_K
    [16384, 6144]
  • blk.25.attn_q_a.weight
    Q5_K
    [6144, 2048]
  • blk.25.attn_q_a_norm.weight
    F32
    [2048]
  • blk.25.attn_q_b.weight
    Q8_0
    [2048, 16384]
  • blk.25.attn_v_b.weight
    Q8_0
    [512, 256, 64]
  • blk.25.exp_probs_b.bias
    F32
    [256]
  • blk.25.ffn_down_exps.weight
    IQ3_XXS
    [2048, 6144, 256]
  • blk.25.ffn_down_shexp.weight
    Q6_K
    [2048, 6144]
  • blk.25.ffn_gate_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.25.ffn_gate_inp.weight
    F32
    [6144, 256]
  • blk.25.ffn_gate_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.25.ffn_norm.weight
    F32
    [6144]
  • blk.25.ffn_up_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.25.ffn_up_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.25.indexer.attn_k.weight
    Q8_0
    [6144, 128]
  • blk.25.indexer.attn_q_b.weight
    Q8_0
    [2048, 4096]
  • blk.25.indexer.k_norm.bias
    F32
    [128]
  • blk.25.indexer.k_norm.weight
    F32
    [128]
  • blk.25.indexer.proj.weight
    F32
    [6144, 32]
    • blk.26
  • blk.26.attn_k_b.weight
    Q8_0
    [192, 512, 64]
  • blk.26.attn_kv_a_mqa.weight
    Q8_0
    [6144, 576]
  • blk.26.attn_kv_a_norm.weight
    F32
    [512]
  • blk.26.attn_norm.weight
    F32
    [6144]
  • blk.26.attn_output.weight
    Q5_K
    [16384, 6144]
  • blk.26.attn_q_a.weight
    Q5_K
    [6144, 2048]
  • blk.26.attn_q_a_norm.weight
    F32
    [2048]
  • blk.26.attn_q_b.weight
    Q8_0
    [2048, 16384]
  • blk.26.attn_v_b.weight
    Q8_0
    [512, 256, 64]
  • blk.26.exp_probs_b.bias
    F32
    [256]
  • blk.26.ffn_down_exps.weight
    IQ3_XXS
    [2048, 6144, 256]
  • blk.26.ffn_down_shexp.weight
    Q6_K
    [2048, 6144]
  • blk.26.ffn_gate_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.26.ffn_gate_inp.weight
    F32
    [6144, 256]
  • blk.26.ffn_gate_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.26.ffn_norm.weight
    F32
    [6144]
  • blk.26.ffn_up_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.26.ffn_up_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.26.indexer.attn_k.weight
    Q8_0
    [6144, 128]
  • blk.26.indexer.attn_q_b.weight
    Q8_0
    [2048, 4096]
  • blk.26.indexer.k_norm.bias
    F32
    [128]
  • blk.26.indexer.k_norm.weight
    F32
    [128]
  • blk.26.indexer.proj.weight
    F32
    [6144, 32]
    • blk.27
  • blk.27.attn_k_b.weight
    Q8_0
    [192, 512, 64]
  • blk.27.attn_kv_a_mqa.weight
    Q8_0
    [6144, 576]
  • blk.27.attn_kv_a_norm.weight
    F32
    [512]
  • blk.27.attn_norm.weight
    F32
    [6144]
  • blk.27.attn_output.weight
    Q5_K
    [16384, 6144]
  • blk.27.attn_q_a.weight
    Q5_K
    [6144, 2048]
  • blk.27.attn_q_a_norm.weight
    F32
    [2048]
  • blk.27.attn_q_b.weight
    Q8_0
    [2048, 16384]
  • blk.27.attn_v_b.weight
    Q8_0
    [512, 256, 64]
  • blk.27.exp_probs_b.bias
    F32
    [256]
  • blk.27.ffn_down_exps.weight
    IQ3_XXS
    [2048, 6144, 256]
  • blk.27.ffn_down_shexp.weight
    Q6_K
    [2048, 6144]
  • blk.27.ffn_gate_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.27.ffn_gate_inp.weight
    F32
    [6144, 256]
  • blk.27.ffn_gate_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.27.ffn_norm.weight
    F32
    [6144]
  • blk.27.ffn_up_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.27.ffn_up_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.27.indexer.attn_k.weight
    Q8_0
    [6144, 128]
  • blk.27.indexer.attn_q_b.weight
    Q8_0
    [2048, 4096]
  • blk.27.indexer.k_norm.bias
    F32
    [128]
  • blk.27.indexer.k_norm.weight
    F32
    [128]
  • blk.27.indexer.proj.weight
    F32
    [6144, 32]
    • blk.28
  • blk.28.attn_k_b.weight
    Q8_0
    [192, 512, 64]
  • blk.28.attn_kv_a_mqa.weight
    Q8_0
    [6144, 576]
  • blk.28.attn_kv_a_norm.weight
    F32
    [512]
  • blk.28.attn_norm.weight
    F32
    [6144]
  • blk.28.attn_output.weight
    Q5_K
    [16384, 6144]
  • blk.28.attn_q_a.weight
    Q5_K
    [6144, 2048]
  • blk.28.attn_q_a_norm.weight
    F32
    [2048]
  • blk.28.attn_q_b.weight
    Q8_0
    [2048, 16384]
  • blk.28.attn_v_b.weight
    Q8_0
    [512, 256, 64]
  • blk.28.exp_probs_b.bias
    F32
    [256]
  • blk.28.ffn_down_exps.weight
    IQ3_XXS
    [2048, 6144, 256]
  • blk.28.ffn_down_shexp.weight
    Q6_K
    [2048, 6144]
  • blk.28.ffn_gate_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.28.ffn_gate_inp.weight
    F32
    [6144, 256]
  • blk.28.ffn_gate_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.28.ffn_norm.weight
    F32
    [6144]
  • blk.28.ffn_up_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.28.ffn_up_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.28.indexer.attn_k.weight
    Q8_0
    [6144, 128]
  • blk.28.indexer.attn_q_b.weight
    Q8_0
    [2048, 4096]
  • blk.28.indexer.k_norm.bias
    F32
    [128]
  • blk.28.indexer.k_norm.weight
    F32
    [128]
  • blk.28.indexer.proj.weight
    F32
    [6144, 32]
    • blk.29
  • blk.29.attn_k_b.weight
    Q8_0
    [192, 512, 64]
  • blk.29.attn_kv_a_mqa.weight
    Q8_0
    [6144, 576]
  • blk.29.attn_kv_a_norm.weight
    F32
    [512]
  • blk.29.attn_norm.weight
    F32
    [6144]
  • blk.29.attn_output.weight
    Q5_K
    [16384, 6144]
  • blk.29.attn_q_a.weight
    Q5_K
    [6144, 2048]
  • blk.29.attn_q_a_norm.weight
    F32
    [2048]
  • blk.29.attn_q_b.weight
    Q8_0
    [2048, 16384]
  • blk.29.attn_v_b.weight
    Q8_0
    [512, 256, 64]
  • blk.29.exp_probs_b.bias
    F32
    [256]
  • blk.29.ffn_down_exps.weight
    IQ3_XXS
    [2048, 6144, 256]
  • blk.29.ffn_down_shexp.weight
    Q6_K
    [2048, 6144]
  • blk.29.ffn_gate_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.29.ffn_gate_inp.weight
    F32
    [6144, 256]
  • blk.29.ffn_gate_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.29.ffn_norm.weight
    F32
    [6144]
  • blk.29.ffn_up_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.29.ffn_up_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.29.indexer.attn_k.weight
    Q8_0
    [6144, 128]
  • blk.29.indexer.attn_q_b.weight
    Q8_0
    [2048, 4096]
  • blk.29.indexer.k_norm.bias
    F32
    [128]
  • blk.29.indexer.k_norm.weight
    F32
    [128]
  • blk.29.indexer.proj.weight
    F32
    [6144, 32]
    • blk.30
  • blk.30.attn_k_b.weight
    Q8_0
    [192, 512, 64]
  • blk.30.attn_kv_a_mqa.weight
    Q8_0
    [6144, 576]
  • blk.30.attn_kv_a_norm.weight
    F32
    [512]
  • blk.30.attn_norm.weight
    F32
    [6144]
  • blk.30.attn_output.weight
    Q5_K
    [16384, 6144]
  • blk.30.attn_q_a.weight
    Q5_K
    [6144, 2048]
  • blk.30.attn_q_a_norm.weight
    F32
    [2048]
  • blk.30.attn_q_b.weight
    Q8_0
    [2048, 16384]
  • blk.30.attn_v_b.weight
    Q8_0
    [512, 256, 64]
  • blk.30.exp_probs_b.bias
    F32
    [256]
  • blk.30.ffn_down_exps.weight
    IQ3_XXS
    [2048, 6144, 256]
  • blk.30.ffn_down_shexp.weight
    Q6_K
    [2048, 6144]
  • blk.30.ffn_gate_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.30.ffn_gate_inp.weight
    F32
    [6144, 256]
  • blk.30.ffn_gate_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.30.ffn_norm.weight
    F32
    [6144]
  • blk.30.ffn_up_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.30.ffn_up_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.30.indexer.attn_k.weight
    Q8_0
    [6144, 128]
  • blk.30.indexer.attn_q_b.weight
    Q8_0
    [2048, 4096]
  • blk.30.indexer.k_norm.bias
    F32
    [128]
  • blk.30.indexer.k_norm.weight
    F32
    [128]
  • blk.30.indexer.proj.weight
    F32
    [6144, 32]
    • blk.31
  • blk.31.attn_k_b.weight
    Q8_0
    [192, 512, 64]
  • blk.31.attn_kv_a_mqa.weight
    Q8_0
    [6144, 576]
  • blk.31.attn_kv_a_norm.weight
    F32
    [512]
  • blk.31.attn_norm.weight
    F32
    [6144]
  • blk.31.attn_output.weight
    Q5_K
    [16384, 6144]
  • blk.31.attn_q_a.weight
    Q5_K
    [6144, 2048]
  • blk.31.attn_q_a_norm.weight
    F32
    [2048]
  • blk.31.attn_q_b.weight
    Q8_0
    [2048, 16384]
  • blk.31.attn_v_b.weight
    Q8_0
    [512, 256, 64]
  • blk.31.exp_probs_b.bias
    F32
    [256]
  • blk.31.ffn_down_exps.weight
    IQ3_XXS
    [2048, 6144, 256]
  • blk.31.ffn_down_shexp.weight
    Q6_K
    [2048, 6144]
  • blk.31.ffn_gate_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.31.ffn_gate_inp.weight
    F32
    [6144, 256]
  • blk.31.ffn_gate_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.31.ffn_norm.weight
    F32
    [6144]
  • blk.31.ffn_up_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.31.ffn_up_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.31.indexer.attn_k.weight
    Q8_0
    [6144, 128]
  • blk.31.indexer.attn_q_b.weight
    Q8_0
    [2048, 4096]
  • blk.31.indexer.k_norm.bias
    F32
    [128]
  • blk.31.indexer.k_norm.weight
    F32
    [128]
  • blk.31.indexer.proj.weight
    F32
    [6144, 32]
    • blk.32
  • blk.32.attn_k_b.weight
    Q8_0
    [192, 512, 64]
  • blk.32.attn_kv_a_mqa.weight
    Q8_0
    [6144, 576]
  • blk.32.attn_kv_a_norm.weight
    F32
    [512]
  • blk.32.attn_norm.weight
    F32
    [6144]
  • blk.32.attn_output.weight
    Q5_K
    [16384, 6144]
  • blk.32.attn_q_a.weight
    Q5_K
    [6144, 2048]
  • blk.32.attn_q_a_norm.weight
    F32
    [2048]
  • blk.32.attn_q_b.weight
    Q8_0
    [2048, 16384]
  • blk.32.attn_v_b.weight
    Q8_0
    [512, 256, 64]
  • blk.32.exp_probs_b.bias
    F32
    [256]
  • blk.32.ffn_down_exps.weight
    IQ3_XXS
    [2048, 6144, 256]
  • blk.32.ffn_down_shexp.weight
    Q6_K
    [2048, 6144]
  • blk.32.ffn_gate_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.32.ffn_gate_inp.weight
    F32
    [6144, 256]
  • blk.32.ffn_gate_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.32.ffn_norm.weight
    F32
    [6144]
  • blk.32.ffn_up_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.32.ffn_up_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.32.indexer.attn_k.weight
    Q8_0
    [6144, 128]
  • blk.32.indexer.attn_q_b.weight
    Q8_0
    [2048, 4096]
  • blk.32.indexer.k_norm.bias
    F32
    [128]
  • blk.32.indexer.k_norm.weight
    F32
    [128]
  • blk.32.indexer.proj.weight
    F32
    [6144, 32]
    • blk.33
  • blk.33.attn_k_b.weight
    Q8_0
    [192, 512, 64]
  • blk.33.attn_kv_a_mqa.weight
    Q8_0
    [6144, 576]
  • blk.33.attn_kv_a_norm.weight
    F32
    [512]
  • blk.33.attn_norm.weight
    F32
    [6144]
  • blk.33.attn_output.weight
    Q5_K
    [16384, 6144]
  • blk.33.attn_q_a.weight
    Q5_K
    [6144, 2048]
  • blk.33.attn_q_a_norm.weight
    F32
    [2048]
  • blk.33.attn_q_b.weight
    Q8_0
    [2048, 16384]
  • blk.33.attn_v_b.weight
    Q8_0
    [512, 256, 64]
  • blk.33.exp_probs_b.bias
    F32
    [256]
  • blk.33.ffn_down_exps.weight
    IQ3_XXS
    [2048, 6144, 256]
  • blk.33.ffn_down_shexp.weight
    Q6_K
    [2048, 6144]
  • blk.33.ffn_gate_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.33.ffn_gate_inp.weight
    F32
    [6144, 256]
  • blk.33.ffn_gate_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.33.ffn_norm.weight
    F32
    [6144]
  • blk.33.ffn_up_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.33.ffn_up_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.33.indexer.attn_k.weight
    Q8_0
    [6144, 128]
  • blk.33.indexer.attn_q_b.weight
    Q8_0
    [2048, 4096]
  • blk.33.indexer.k_norm.bias
    F32
    [128]
  • blk.33.indexer.k_norm.weight
    F32
    [128]
  • blk.33.indexer.proj.weight
    F32
    [6144, 32]
    • blk.34
  • blk.34.attn_k_b.weight
    Q8_0
    [192, 512, 64]
  • blk.34.attn_kv_a_mqa.weight
    Q8_0
    [6144, 576]
  • blk.34.attn_kv_a_norm.weight
    F32
    [512]
  • blk.34.attn_norm.weight
    F32
    [6144]
  • blk.34.attn_output.weight
    Q5_K
    [16384, 6144]
  • blk.34.attn_q_a.weight
    Q5_K
    [6144, 2048]
  • blk.34.attn_q_a_norm.weight
    F32
    [2048]
  • blk.34.attn_q_b.weight
    Q8_0
    [2048, 16384]
  • blk.34.attn_v_b.weight
    Q8_0
    [512, 256, 64]
  • blk.34.exp_probs_b.bias
    F32
    [256]
  • blk.34.ffn_down_exps.weight
    IQ3_XXS
    [2048, 6144, 256]
  • blk.34.ffn_down_shexp.weight
    Q6_K
    [2048, 6144]
  • blk.34.ffn_gate_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.34.ffn_gate_inp.weight
    F32
    [6144, 256]
  • blk.34.ffn_gate_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.34.ffn_norm.weight
    F32
    [6144]