free02/ glm5.1:part5

149 1 month ago

... /
model
f7090c43c88a · 49GB
    Metadata
  • split.count
    6
  • split.no
    4
  • split.tensors.count
    1809
    • Tensor
    blk.51
  • blk.51.attn_kv_a_mqa.weight
    Q8_0
    [6144, 576]
  • blk.51.attn_kv_a_norm.weight
    F32
    [512]
  • blk.51.attn_norm.weight
    F32
    [6144]
  • blk.51.attn_output.weight
    Q5_K
    [16384, 6144]
  • blk.51.attn_q_a.weight
    Q5_K
    [6144, 2048]
  • blk.51.attn_q_a_norm.weight
    F32
    [2048]
  • blk.51.attn_q_b.weight
    Q8_0
    [2048, 16384]
  • blk.51.attn_v_b.weight
    Q8_0
    [512, 256, 64]
  • blk.51.exp_probs_b.bias
    F32
    [256]
  • blk.51.ffn_down_exps.weight
    IQ3_XXS
    [2048, 6144, 256]
  • blk.51.ffn_down_shexp.weight
    Q6_K
    [2048, 6144]
  • blk.51.ffn_gate_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.51.ffn_gate_inp.weight
    F32
    [6144, 256]
  • blk.51.ffn_gate_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.51.ffn_norm.weight
    F32
    [6144]
  • blk.51.ffn_up_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.51.ffn_up_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.51.indexer.attn_k.weight
    Q8_0
    [6144, 128]
  • blk.51.indexer.attn_q_b.weight
    Q8_0
    [2048, 4096]
  • blk.51.indexer.k_norm.bias
    F32
    [128]
  • blk.51.indexer.k_norm.weight
    F32
    [128]
  • blk.51.indexer.proj.weight
    F32
    [6144, 32]
    • blk.52
  • blk.52.attn_k_b.weight
    Q8_0
    [192, 512, 64]
  • blk.52.attn_kv_a_mqa.weight
    Q8_0
    [6144, 576]
  • blk.52.attn_kv_a_norm.weight
    F32
    [512]
  • blk.52.attn_norm.weight
    F32
    [6144]
  • blk.52.attn_output.weight
    Q5_K
    [16384, 6144]
  • blk.52.attn_q_a.weight
    Q5_K
    [6144, 2048]
  • blk.52.attn_q_a_norm.weight
    F32
    [2048]
  • blk.52.attn_q_b.weight
    Q8_0
    [2048, 16384]
  • blk.52.attn_v_b.weight
    Q8_0
    [512, 256, 64]
  • blk.52.exp_probs_b.bias
    F32
    [256]
  • blk.52.ffn_down_exps.weight
    IQ3_XXS
    [2048, 6144, 256]
  • blk.52.ffn_down_shexp.weight
    Q6_K
    [2048, 6144]
  • blk.52.ffn_gate_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.52.ffn_gate_inp.weight
    F32
    [6144, 256]
  • blk.52.ffn_gate_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.52.ffn_norm.weight
    F32
    [6144]
  • blk.52.ffn_up_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.52.ffn_up_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.52.indexer.attn_k.weight
    Q8_0
    [6144, 128]
  • blk.52.indexer.attn_q_b.weight
    Q8_0
    [2048, 4096]
  • blk.52.indexer.k_norm.bias
    F32
    [128]
  • blk.52.indexer.k_norm.weight
    F32
    [128]
  • blk.52.indexer.proj.weight
    F32
    [6144, 32]
    • blk.53
  • blk.53.attn_k_b.weight
    Q8_0
    [192, 512, 64]
  • blk.53.attn_kv_a_mqa.weight
    Q8_0
    [6144, 576]
  • blk.53.attn_kv_a_norm.weight
    F32
    [512]
  • blk.53.attn_norm.weight
    F32
    [6144]
  • blk.53.attn_output.weight
    Q5_K
    [16384, 6144]
  • blk.53.attn_q_a.weight
    Q5_K
    [6144, 2048]
  • blk.53.attn_q_a_norm.weight
    F32
    [2048]
  • blk.53.attn_q_b.weight
    Q8_0
    [2048, 16384]
  • blk.53.attn_v_b.weight
    Q8_0
    [512, 256, 64]
  • blk.53.exp_probs_b.bias
    F32
    [256]
  • blk.53.ffn_down_exps.weight
    IQ3_XXS
    [2048, 6144, 256]
  • blk.53.ffn_down_shexp.weight
    Q6_K
    [2048, 6144]
  • blk.53.ffn_gate_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.53.ffn_gate_inp.weight
    F32
    [6144, 256]
  • blk.53.ffn_gate_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.53.ffn_norm.weight
    F32
    [6144]
  • blk.53.ffn_up_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.53.ffn_up_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.53.indexer.attn_k.weight
    Q8_0
    [6144, 128]
  • blk.53.indexer.attn_q_b.weight
    Q8_0
    [2048, 4096]
  • blk.53.indexer.k_norm.bias
    F32
    [128]
  • blk.53.indexer.k_norm.weight
    F32
    [128]
  • blk.53.indexer.proj.weight
    F32
    [6144, 32]
    • blk.54
  • blk.54.attn_k_b.weight
    Q8_0
    [192, 512, 64]
  • blk.54.attn_kv_a_mqa.weight
    Q8_0
    [6144, 576]
  • blk.54.attn_kv_a_norm.weight
    F32
    [512]
  • blk.54.attn_norm.weight
    F32
    [6144]
  • blk.54.attn_output.weight
    Q5_K
    [16384, 6144]
  • blk.54.attn_q_a.weight
    Q5_K
    [6144, 2048]
  • blk.54.attn_q_a_norm.weight
    F32
    [2048]
  • blk.54.attn_q_b.weight
    Q8_0
    [2048, 16384]
  • blk.54.attn_v_b.weight
    Q8_0
    [512, 256, 64]
  • blk.54.exp_probs_b.bias
    F32
    [256]
  • blk.54.ffn_down_exps.weight
    IQ3_XXS
    [2048, 6144, 256]
  • blk.54.ffn_down_shexp.weight
    Q6_K
    [2048, 6144]
  • blk.54.ffn_gate_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.54.ffn_gate_inp.weight
    F32
    [6144, 256]
  • blk.54.ffn_gate_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.54.ffn_norm.weight
    F32
    [6144]
  • blk.54.ffn_up_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.54.ffn_up_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.54.indexer.attn_k.weight
    Q8_0
    [6144, 128]
  • blk.54.indexer.attn_q_b.weight
    Q8_0
    [2048, 4096]
  • blk.54.indexer.k_norm.bias
    F32
    [128]
  • blk.54.indexer.k_norm.weight
    F32
    [128]
  • blk.54.indexer.proj.weight
    F32
    [6144, 32]
    • blk.55
  • blk.55.attn_k_b.weight
    Q8_0
    [192, 512, 64]
  • blk.55.attn_kv_a_mqa.weight
    Q8_0
    [6144, 576]
  • blk.55.attn_kv_a_norm.weight
    F32
    [512]
  • blk.55.attn_norm.weight
    F32
    [6144]
  • blk.55.attn_output.weight
    Q5_K
    [16384, 6144]
  • blk.55.attn_q_a.weight
    Q5_K
    [6144, 2048]
  • blk.55.attn_q_a_norm.weight
    F32
    [2048]
  • blk.55.attn_q_b.weight
    Q8_0
    [2048, 16384]
  • blk.55.attn_v_b.weight
    Q8_0
    [512, 256, 64]
  • blk.55.exp_probs_b.bias
    F32
    [256]
  • blk.55.ffn_down_exps.weight
    IQ3_XXS
    [2048, 6144, 256]
  • blk.55.ffn_down_shexp.weight
    Q6_K
    [2048, 6144]
  • blk.55.ffn_gate_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.55.ffn_gate_inp.weight
    F32
    [6144, 256]
  • blk.55.ffn_gate_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.55.ffn_norm.weight
    F32
    [6144]
  • blk.55.ffn_up_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.55.ffn_up_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.55.indexer.attn_k.weight
    Q8_0
    [6144, 128]
  • blk.55.indexer.attn_q_b.weight
    Q8_0
    [2048, 4096]
  • blk.55.indexer.k_norm.bias
    F32
    [128]
  • blk.55.indexer.k_norm.weight
    F32
    [128]
  • blk.55.indexer.proj.weight
    F32
    [6144, 32]
    • blk.56
  • blk.56.attn_k_b.weight
    Q8_0
    [192, 512, 64]
  • blk.56.attn_kv_a_mqa.weight
    Q8_0
    [6144, 576]
  • blk.56.attn_kv_a_norm.weight
    F32
    [512]
  • blk.56.attn_norm.weight
    F32
    [6144]
  • blk.56.attn_output.weight
    Q5_K
    [16384, 6144]
  • blk.56.attn_q_a.weight
    Q5_K
    [6144, 2048]
  • blk.56.attn_q_a_norm.weight
    F32
    [2048]
  • blk.56.attn_q_b.weight
    Q8_0
    [2048, 16384]
  • blk.56.attn_v_b.weight
    Q8_0
    [512, 256, 64]
  • blk.56.exp_probs_b.bias
    F32
    [256]
  • blk.56.ffn_down_exps.weight
    IQ3_XXS
    [2048, 6144, 256]
  • blk.56.ffn_down_shexp.weight
    Q6_K
    [2048, 6144]
  • blk.56.ffn_gate_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.56.ffn_gate_inp.weight
    F32
    [6144, 256]
  • blk.56.ffn_gate_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.56.ffn_norm.weight
    F32
    [6144]
  • blk.56.ffn_up_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.56.ffn_up_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.56.indexer.attn_k.weight
    Q8_0
    [6144, 128]
  • blk.56.indexer.attn_q_b.weight
    Q8_0
    [2048, 4096]
  • blk.56.indexer.k_norm.bias
    F32
    [128]
  • blk.56.indexer.k_norm.weight
    F32
    [128]
  • blk.56.indexer.proj.weight
    F32
    [6144, 32]
    • blk.57
  • blk.57.attn_k_b.weight
    Q8_0
    [192, 512, 64]
  • blk.57.attn_kv_a_mqa.weight
    Q8_0
    [6144, 576]
  • blk.57.attn_kv_a_norm.weight
    F32
    [512]
  • blk.57.attn_norm.weight
    F32
    [6144]
  • blk.57.attn_output.weight
    Q5_K
    [16384, 6144]
  • blk.57.attn_q_a.weight
    Q5_K
    [6144, 2048]
  • blk.57.attn_q_a_norm.weight
    F32
    [2048]
  • blk.57.attn_q_b.weight
    Q8_0
    [2048, 16384]
  • blk.57.attn_v_b.weight
    Q8_0
    [512, 256, 64]
  • blk.57.exp_probs_b.bias
    F32
    [256]
  • blk.57.ffn_down_exps.weight
    IQ3_XXS
    [2048, 6144, 256]
  • blk.57.ffn_down_shexp.weight
    Q6_K
    [2048, 6144]
  • blk.57.ffn_gate_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.57.ffn_gate_inp.weight
    F32
    [6144, 256]
  • blk.57.ffn_gate_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.57.ffn_norm.weight
    F32
    [6144]
  • blk.57.ffn_up_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.57.ffn_up_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.57.indexer.attn_k.weight
    Q8_0
    [6144, 128]
  • blk.57.indexer.attn_q_b.weight
    Q8_0
    [2048, 4096]
  • blk.57.indexer.k_norm.bias
    F32
    [128]
  • blk.57.indexer.k_norm.weight
    F32
    [128]
  • blk.57.indexer.proj.weight
    F32
    [6144, 32]
    • blk.58
  • blk.58.attn_k_b.weight
    Q8_0
    [192, 512, 64]
  • blk.58.attn_kv_a_mqa.weight
    Q8_0
    [6144, 576]
  • blk.58.attn_kv_a_norm.weight
    F32
    [512]
  • blk.58.attn_norm.weight
    F32
    [6144]
  • blk.58.attn_output.weight
    Q5_K
    [16384, 6144]
  • blk.58.attn_q_a.weight
    Q5_K
    [6144, 2048]
  • blk.58.attn_q_a_norm.weight
    F32
    [2048]
  • blk.58.attn_q_b.weight
    Q8_0
    [2048, 16384]
  • blk.58.attn_v_b.weight
    Q8_0
    [512, 256, 64]
  • blk.58.exp_probs_b.bias
    F32
    [256]
  • blk.58.ffn_down_exps.weight
    IQ3_XXS
    [2048, 6144, 256]
  • blk.58.ffn_down_shexp.weight
    Q6_K
    [2048, 6144]
  • blk.58.ffn_gate_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.58.ffn_gate_inp.weight
    F32
    [6144, 256]
  • blk.58.ffn_gate_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.58.ffn_norm.weight
    F32
    [6144]
  • blk.58.ffn_up_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.58.ffn_up_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.58.indexer.attn_k.weight
    Q8_0
    [6144, 128]
  • blk.58.indexer.attn_q_b.weight
    Q8_0
    [2048, 4096]
  • blk.58.indexer.k_norm.bias
    F32
    [128]
  • blk.58.indexer.k_norm.weight
    F32
    [128]
  • blk.58.indexer.proj.weight
    F32
    [6144, 32]
    • blk.59
  • blk.59.attn_k_b.weight
    Q8_0
    [192, 512, 64]
  • blk.59.attn_kv_a_mqa.weight
    Q8_0
    [6144, 576]
  • blk.59.attn_kv_a_norm.weight
    F32
    [512]
  • blk.59.attn_norm.weight
    F32
    [6144]
  • blk.59.attn_output.weight
    Q5_K
    [16384, 6144]
  • blk.59.attn_q_a.weight
    Q5_K
    [6144, 2048]
  • blk.59.attn_q_a_norm.weight
    F32
    [2048]
  • blk.59.attn_q_b.weight
    Q8_0
    [2048, 16384]
  • blk.59.attn_v_b.weight
    Q8_0
    [512, 256, 64]
  • blk.59.exp_probs_b.bias
    F32
    [256]
  • blk.59.ffn_down_exps.weight
    IQ3_XXS
    [2048, 6144, 256]
  • blk.59.ffn_down_shexp.weight
    Q6_K
    [2048, 6144]
  • blk.59.ffn_gate_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.59.ffn_gate_inp.weight
    F32
    [6144, 256]
  • blk.59.ffn_gate_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.59.ffn_norm.weight
    F32
    [6144]
  • blk.59.ffn_up_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.59.ffn_up_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.59.indexer.attn_k.weight
    Q8_0
    [6144, 128]
  • blk.59.indexer.attn_q_b.weight
    Q8_0
    [2048, 4096]
  • blk.59.indexer.k_norm.bias
    F32
    [128]
  • blk.59.indexer.k_norm.weight
    F32
    [128]
  • blk.59.indexer.proj.weight
    F32
    [6144, 32]
    • blk.60
  • blk.60.attn_k_b.weight
    Q8_0
    [192, 512, 64]
  • blk.60.attn_kv_a_mqa.weight
    Q8_0
    [6144, 576]
  • blk.60.attn_kv_a_norm.weight
    F32
    [512]
  • blk.60.attn_norm.weight
    F32
    [6144]
  • blk.60.attn_output.weight
    Q5_K
    [16384, 6144]
  • blk.60.attn_q_a.weight
    Q5_K
    [6144, 2048]
  • blk.60.attn_q_a_norm.weight
    F32
    [2048]
  • blk.60.attn_q_b.weight
    Q8_0
    [2048, 16384]
  • blk.60.attn_v_b.weight
    Q8_0
    [512, 256, 64]
  • blk.60.exp_probs_b.bias
    F32
    [256]
  • blk.60.ffn_down_exps.weight
    IQ3_XXS
    [2048, 6144, 256]
  • blk.60.ffn_down_shexp.weight
    Q6_K
    [2048, 6144]
  • blk.60.ffn_gate_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.60.ffn_gate_inp.weight
    F32
    [6144, 256]
  • blk.60.ffn_gate_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.60.ffn_norm.weight
    F32
    [6144]
  • blk.60.ffn_up_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.60.ffn_up_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.60.indexer.attn_k.weight
    Q8_0
    [6144, 128]
  • blk.60.indexer.attn_q_b.weight
    Q8_0
    [2048, 4096]
  • blk.60.indexer.k_norm.bias
    F32
    [128]
  • blk.60.indexer.k_norm.weight
    F32
    [128]
  • blk.60.indexer.proj.weight
    F32
    [6144, 32]
    • blk.61
  • blk.61.attn_k_b.weight
    Q8_0
    [192, 512, 64]
  • blk.61.attn_kv_a_mqa.weight
    Q8_0
    [6144, 576]
  • blk.61.attn_kv_a_norm.weight
    F32
    [512]
  • blk.61.attn_norm.weight
    F32
    [6144]
  • blk.61.attn_output.weight
    Q5_K
    [16384, 6144]
  • blk.61.attn_q_a.weight
    Q5_K
    [6144, 2048]
  • blk.61.attn_q_a_norm.weight
    F32
    [2048]
  • blk.61.attn_q_b.weight
    Q8_0
    [2048, 16384]
  • blk.61.attn_v_b.weight
    Q8_0
    [512, 256, 64]
  • blk.61.exp_probs_b.bias
    F32
    [256]
  • blk.61.ffn_down_exps.weight
    IQ3_XXS
    [2048, 6144, 256]
  • blk.61.ffn_down_shexp.weight
    Q6_K
    [2048, 6144]
  • blk.61.ffn_gate_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.61.ffn_gate_inp.weight
    F32
    [6144, 256]
  • blk.61.ffn_gate_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.61.ffn_norm.weight
    F32
    [6144]
  • blk.61.ffn_up_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.61.ffn_up_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.61.indexer.attn_k.weight
    Q8_0
    [6144, 128]
  • blk.61.indexer.attn_q_b.weight
    Q8_0
    [2048, 4096]
  • blk.61.indexer.k_norm.bias
    F32
    [128]
  • blk.61.indexer.k_norm.weight
    F32
    [128]
  • blk.61.indexer.proj.weight
    F32
    [6144, 32]
    • blk.62
  • blk.62.attn_k_b.weight
    Q8_0
    [192, 512, 64]
  • blk.62.attn_kv_a_mqa.weight
    Q8_0
    [6144, 576]
  • blk.62.attn_kv_a_norm.weight
    F32
    [512]
  • blk.62.attn_norm.weight
    F32
    [6144]
  • blk.62.attn_output.weight
    Q5_K
    [16384, 6144]
  • blk.62.attn_q_a.weight
    Q5_K
    [6144, 2048]
  • blk.62.attn_q_a_norm.weight
    F32
    [2048]
  • blk.62.attn_q_b.weight
    Q8_0
    [2048, 16384]
  • blk.62.attn_v_b.weight
    Q8_0
    [512, 256, 64]
  • blk.62.exp_probs_b.bias
    F32
    [256]
  • blk.62.ffn_down_exps.weight
    IQ3_XXS
    [2048, 6144, 256]
  • blk.62.ffn_down_shexp.weight
    Q6_K
    [2048, 6144]
  • blk.62.ffn_gate_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.62.ffn_gate_inp.weight
    F32
    [6144, 256]
  • blk.62.ffn_gate_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.62.ffn_norm.weight
    F32
    [6144]
  • blk.62.ffn_up_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.62.ffn_up_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.62.indexer.attn_k.weight
    Q8_0
    [6144, 128]
  • blk.62.indexer.attn_q_b.weight
    Q8_0
    [2048, 4096]
  • blk.62.indexer.k_norm.bias
    F32
    [128]
  • blk.62.indexer.k_norm.weight
    F32
    [128]
  • blk.62.indexer.proj.weight
    F32
    [6144, 32]
    • blk.63
  • blk.63.attn_k_b.weight
    Q8_0
    [192, 512, 64]
  • blk.63.attn_kv_a_mqa.weight
    Q8_0
    [6144, 576]
  • blk.63.attn_kv_a_norm.weight
    F32
    [512]
  • blk.63.attn_norm.weight
    F32
    [6144]
  • blk.63.attn_output.weight
    Q5_K
    [16384, 6144]
  • blk.63.attn_q_a.weight
    Q5_K
    [6144, 2048]
  • blk.63.attn_q_a_norm.weight
    F32
    [2048]
  • blk.63.attn_q_b.weight
    Q8_0
    [2048, 16384]
  • blk.63.attn_v_b.weight
    Q8_0
    [512, 256, 64]
  • blk.63.exp_probs_b.bias
    F32
    [256]
  • blk.63.ffn_down_exps.weight
    IQ3_XXS
    [2048, 6144, 256]
  • blk.63.ffn_down_shexp.weight
    Q6_K
    [2048, 6144]
  • blk.63.ffn_gate_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.63.ffn_gate_inp.weight
    F32
    [6144, 256]
  • blk.63.ffn_gate_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.63.ffn_norm.weight
    F32
    [6144]
  • blk.63.ffn_up_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.63.ffn_up_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.63.indexer.attn_k.weight
    Q8_0
    [6144, 128]
  • blk.63.indexer.attn_q_b.weight
    Q8_0
    [2048, 4096]
  • blk.63.indexer.k_norm.bias
    F32
    [128]
  • blk.63.indexer.k_norm.weight
    F32
    [128]
  • blk.63.indexer.proj.weight
    F32
    [6144, 32]
    • blk.64
  • blk.64.attn_k_b.weight
    Q8_0
    [192, 512, 64]
  • blk.64.attn_kv_a_mqa.weight
    Q8_0
    [6144, 576]
  • blk.64.attn_kv_a_norm.weight
    F32
    [512]
  • blk.64.attn_norm.weight
    F32
    [6144]
  • blk.64.attn_output.weight
    Q5_K
    [16384, 6144]
  • blk.64.attn_q_a.weight
    Q5_K
    [6144, 2048]
  • blk.64.attn_q_a_norm.weight
    F32
    [2048]
  • blk.64.attn_q_b.weight
    Q8_0
    [2048, 16384]
  • blk.64.attn_v_b.weight
    Q8_0
    [512, 256, 64]
  • blk.64.exp_probs_b.bias
    F32
    [256]
  • blk.64.ffn_down_exps.weight
    IQ3_XXS
    [2048, 6144, 256]
  • blk.64.ffn_down_shexp.weight
    Q6_K
    [2048, 6144]
  • blk.64.ffn_gate_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.64.ffn_gate_inp.weight
    F32
    [6144, 256]
  • blk.64.ffn_gate_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.64.ffn_norm.weight
    F32
    [6144]
  • blk.64.ffn_up_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.64.ffn_up_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.64.indexer.attn_k.weight
    Q8_0
    [6144, 128]
  • blk.64.indexer.attn_q_b.weight
    Q8_0
    [2048, 4096]
  • blk.64.indexer.k_norm.bias
    F32
    [128]
  • blk.64.indexer.k_norm.weight
    F32
    [128]
  • blk.64.indexer.proj.weight
    F32
    [6144, 32]
    • blk.65
  • blk.65.attn_k_b.weight
    Q8_0
    [192, 512, 64]
  • blk.65.attn_kv_a_mqa.weight
    Q8_0
    [6144, 576]
  • blk.65.attn_kv_a_norm.weight
    F32
    [512]
  • blk.65.attn_norm.weight
    F32
    [6144]
  • blk.65.attn_output.weight
    Q5_K
    [16384, 6144]
  • blk.65.attn_q_a.weight
    Q5_K
    [6144, 2048]
  • blk.65.attn_q_a_norm.weight
    F32
    [2048]
  • blk.65.attn_q_b.weight
    Q8_0
    [2048, 16384]
  • blk.65.attn_v_b.weight
    Q8_0
    [512, 256, 64]
  • blk.65.exp_probs_b.bias
    F32
    [256]
  • blk.65.ffn_down_exps.weight
    IQ3_XXS
    [2048, 6144, 256]
  • blk.65.ffn_down_shexp.weight
    Q6_K
    [2048, 6144]
  • blk.65.ffn_gate_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.65.ffn_gate_inp.weight
    F32
    [6144, 256]
  • blk.65.ffn_gate_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.65.ffn_norm.weight
    F32
    [6144]
  • blk.65.ffn_up_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.65.ffn_up_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.65.indexer.attn_k.weight
    Q8_0
    [6144, 128]
  • blk.65.indexer.attn_q_b.weight
    Q8_0
    [2048, 4096]
  • blk.65.indexer.k_norm.bias
    F32
    [128]
  • blk.65.indexer.k_norm.weight
    F32
    [128]
  • blk.65.indexer.proj.weight
    F32
    [6144, 32]
    • blk.66
  • blk.66.attn_k_b.weight
    Q8_0
    [192, 512, 64]
  • blk.66.attn_kv_a_mqa.weight
    Q8_0
    [6144, 576]
  • blk.66.attn_kv_a_norm.weight
    F32
    [512]
  • blk.66.attn_norm.weight
    F32
    [6144]
  • blk.66.attn_output.weight
    Q5_K
    [16384, 6144]
  • blk.66.attn_q_a.weight
    Q5_K
    [6144, 2048]
  • blk.66.attn_q_a_norm.weight
    F32
    [2048]
  • blk.66.attn_q_b.weight
    Q8_0
    [2048, 16384]
  • blk.66.attn_v_b.weight
    Q8_0
    [512, 256, 64]
  • blk.66.exp_probs_b.bias
    F32
    [256]
  • blk.66.ffn_down_exps.weight
    IQ3_XXS
    [2048, 6144, 256]
  • blk.66.ffn_down_shexp.weight
    Q6_K
    [2048, 6144]
  • blk.66.ffn_gate_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.66.ffn_gate_inp.weight
    F32
    [6144, 256]
  • blk.66.ffn_gate_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.66.ffn_norm.weight
    F32
    [6144]
  • blk.66.ffn_up_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.66.ffn_up_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.66.indexer.attn_k.weight
    Q8_0
    [6144, 128]
  • blk.66.indexer.attn_q_b.weight
    Q8_0
    [2048, 4096]
  • blk.66.indexer.k_norm.bias
    F32
    [128]
  • blk.66.indexer.k_norm.weight
    F32
    [128]
  • blk.66.indexer.proj.weight
    F32
    [6144, 32]
    • blk.67
  • blk.67.attn_k_b.weight
    Q8_0
    [192, 512, 64]
  • blk.67.attn_kv_a_mqa.weight
    Q8_0
    [6144, 576]
  • blk.67.attn_kv_a_norm.weight
    F32
    [512]
  • blk.67.attn_norm.weight
    F32
    [6144]
  • blk.67.attn_output.weight
    Q5_K
    [16384, 6144]
  • blk.67.attn_q_a.weight
    Q5_K
    [6144, 2048]
  • blk.67.attn_q_a_norm.weight
    F32
    [2048]
  • blk.67.attn_q_b.weight
    Q8_0
    [2048, 16384]
  • blk.67.attn_v_b.weight
    Q8_0
    [512, 256, 64]
  • blk.67.exp_probs_b.bias
    F32
    [256]