free02/ glm5.1:part4

149 1 month ago

... /
model
39cc900489af · 50GB
    Metadata
  • split.count
    6
  • split.no
    3
  • split.tensors.count
    1809
    • Tensor
    blk.34
  • blk.34.ffn_up_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.34.ffn_up_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.34.indexer.attn_k.weight
    Q8_0
    [6144, 128]
  • blk.34.indexer.attn_q_b.weight
    Q8_0
    [2048, 4096]
  • blk.34.indexer.k_norm.bias
    F32
    [128]
  • blk.34.indexer.k_norm.weight
    F32
    [128]
  • blk.34.indexer.proj.weight
    F32
    [6144, 32]
    • blk.35
  • blk.35.attn_k_b.weight
    Q8_0
    [192, 512, 64]
  • blk.35.attn_kv_a_mqa.weight
    Q8_0
    [6144, 576]
  • blk.35.attn_kv_a_norm.weight
    F32
    [512]
  • blk.35.attn_norm.weight
    F32
    [6144]
  • blk.35.attn_output.weight
    Q5_K
    [16384, 6144]
  • blk.35.attn_q_a.weight
    Q5_K
    [6144, 2048]
  • blk.35.attn_q_a_norm.weight
    F32
    [2048]
  • blk.35.attn_q_b.weight
    Q8_0
    [2048, 16384]
  • blk.35.attn_v_b.weight
    Q8_0
    [512, 256, 64]
  • blk.35.exp_probs_b.bias
    F32
    [256]
  • blk.35.ffn_down_exps.weight
    IQ3_XXS
    [2048, 6144, 256]
  • blk.35.ffn_down_shexp.weight
    Q6_K
    [2048, 6144]
  • blk.35.ffn_gate_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.35.ffn_gate_inp.weight
    F32
    [6144, 256]
  • blk.35.ffn_gate_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.35.ffn_norm.weight
    F32
    [6144]
  • blk.35.ffn_up_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.35.ffn_up_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.35.indexer.attn_k.weight
    Q8_0
    [6144, 128]
  • blk.35.indexer.attn_q_b.weight
    Q8_0
    [2048, 4096]
  • blk.35.indexer.k_norm.bias
    F32
    [128]
  • blk.35.indexer.k_norm.weight
    F32
    [128]
  • blk.35.indexer.proj.weight
    F32
    [6144, 32]
    • blk.36
  • blk.36.attn_k_b.weight
    Q8_0
    [192, 512, 64]
  • blk.36.attn_kv_a_mqa.weight
    Q8_0
    [6144, 576]
  • blk.36.attn_kv_a_norm.weight
    F32
    [512]
  • blk.36.attn_norm.weight
    F32
    [6144]
  • blk.36.attn_output.weight
    Q5_K
    [16384, 6144]
  • blk.36.attn_q_a.weight
    Q5_K
    [6144, 2048]
  • blk.36.attn_q_a_norm.weight
    F32
    [2048]
  • blk.36.attn_q_b.weight
    Q8_0
    [2048, 16384]
  • blk.36.attn_v_b.weight
    Q8_0
    [512, 256, 64]
  • blk.36.exp_probs_b.bias
    F32
    [256]
  • blk.36.ffn_down_exps.weight
    IQ3_XXS
    [2048, 6144, 256]
  • blk.36.ffn_down_shexp.weight
    Q6_K
    [2048, 6144]
  • blk.36.ffn_gate_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.36.ffn_gate_inp.weight
    F32
    [6144, 256]
  • blk.36.ffn_gate_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.36.ffn_norm.weight
    F32
    [6144]
  • blk.36.ffn_up_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.36.ffn_up_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.36.indexer.attn_k.weight
    Q8_0
    [6144, 128]
  • blk.36.indexer.attn_q_b.weight
    Q8_0
    [2048, 4096]
  • blk.36.indexer.k_norm.bias
    F32
    [128]
  • blk.36.indexer.k_norm.weight
    F32
    [128]
  • blk.36.indexer.proj.weight
    F32
    [6144, 32]
    • blk.37
  • blk.37.attn_k_b.weight
    Q8_0
    [192, 512, 64]
  • blk.37.attn_kv_a_mqa.weight
    Q8_0
    [6144, 576]
  • blk.37.attn_kv_a_norm.weight
    F32
    [512]
  • blk.37.attn_norm.weight
    F32
    [6144]
  • blk.37.attn_output.weight
    Q5_K
    [16384, 6144]
  • blk.37.attn_q_a.weight
    Q5_K
    [6144, 2048]
  • blk.37.attn_q_a_norm.weight
    F32
    [2048]
  • blk.37.attn_q_b.weight
    Q8_0
    [2048, 16384]
  • blk.37.attn_v_b.weight
    Q8_0
    [512, 256, 64]
  • blk.37.exp_probs_b.bias
    F32
    [256]
  • blk.37.ffn_down_exps.weight
    IQ3_XXS
    [2048, 6144, 256]
  • blk.37.ffn_down_shexp.weight
    Q6_K
    [2048, 6144]
  • blk.37.ffn_gate_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.37.ffn_gate_inp.weight
    F32
    [6144, 256]
  • blk.37.ffn_gate_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.37.ffn_norm.weight
    F32
    [6144]
  • blk.37.ffn_up_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.37.ffn_up_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.37.indexer.attn_k.weight
    Q8_0
    [6144, 128]
  • blk.37.indexer.attn_q_b.weight
    Q8_0
    [2048, 4096]
  • blk.37.indexer.k_norm.bias
    F32
    [128]
  • blk.37.indexer.k_norm.weight
    F32
    [128]
  • blk.37.indexer.proj.weight
    F32
    [6144, 32]
    • blk.38
  • blk.38.attn_k_b.weight
    Q8_0
    [192, 512, 64]
  • blk.38.attn_kv_a_mqa.weight
    Q8_0
    [6144, 576]
  • blk.38.attn_kv_a_norm.weight
    F32
    [512]
  • blk.38.attn_norm.weight
    F32
    [6144]
  • blk.38.attn_output.weight
    Q5_K
    [16384, 6144]
  • blk.38.attn_q_a.weight
    Q5_K
    [6144, 2048]
  • blk.38.attn_q_a_norm.weight
    F32
    [2048]
  • blk.38.attn_q_b.weight
    Q8_0
    [2048, 16384]
  • blk.38.attn_v_b.weight
    Q8_0
    [512, 256, 64]
  • blk.38.exp_probs_b.bias
    F32
    [256]
  • blk.38.ffn_down_exps.weight
    IQ3_XXS
    [2048, 6144, 256]
  • blk.38.ffn_down_shexp.weight
    Q6_K
    [2048, 6144]
  • blk.38.ffn_gate_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.38.ffn_gate_inp.weight
    F32
    [6144, 256]
  • blk.38.ffn_gate_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.38.ffn_norm.weight
    F32
    [6144]
  • blk.38.ffn_up_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.38.ffn_up_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.38.indexer.attn_k.weight
    Q8_0
    [6144, 128]
  • blk.38.indexer.attn_q_b.weight
    Q8_0
    [2048, 4096]
  • blk.38.indexer.k_norm.bias
    F32
    [128]
  • blk.38.indexer.k_norm.weight
    F32
    [128]
  • blk.38.indexer.proj.weight
    F32
    [6144, 32]
    • blk.39
  • blk.39.attn_k_b.weight
    Q8_0
    [192, 512, 64]
  • blk.39.attn_kv_a_mqa.weight
    Q8_0
    [6144, 576]
  • blk.39.attn_kv_a_norm.weight
    F32
    [512]
  • blk.39.attn_norm.weight
    F32
    [6144]
  • blk.39.attn_output.weight
    Q5_K
    [16384, 6144]
  • blk.39.attn_q_a.weight
    Q5_K
    [6144, 2048]
  • blk.39.attn_q_a_norm.weight
    F32
    [2048]
  • blk.39.attn_q_b.weight
    Q8_0
    [2048, 16384]
  • blk.39.attn_v_b.weight
    Q8_0
    [512, 256, 64]
  • blk.39.exp_probs_b.bias
    F32
    [256]
  • blk.39.ffn_down_exps.weight
    IQ3_XXS
    [2048, 6144, 256]
  • blk.39.ffn_down_shexp.weight
    Q6_K
    [2048, 6144]
  • blk.39.ffn_gate_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.39.ffn_gate_inp.weight
    F32
    [6144, 256]
  • blk.39.ffn_gate_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.39.ffn_norm.weight
    F32
    [6144]
  • blk.39.ffn_up_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.39.ffn_up_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.39.indexer.attn_k.weight
    Q8_0
    [6144, 128]
  • blk.39.indexer.attn_q_b.weight
    Q8_0
    [2048, 4096]
  • blk.39.indexer.k_norm.bias
    F32
    [128]
  • blk.39.indexer.k_norm.weight
    F32
    [128]
  • blk.39.indexer.proj.weight
    F32
    [6144, 32]
    • blk.40
  • blk.40.attn_k_b.weight
    Q8_0
    [192, 512, 64]
  • blk.40.attn_kv_a_mqa.weight
    Q8_0
    [6144, 576]
  • blk.40.attn_kv_a_norm.weight
    F32
    [512]
  • blk.40.attn_norm.weight
    F32
    [6144]
  • blk.40.attn_output.weight
    Q5_K
    [16384, 6144]
  • blk.40.attn_q_a.weight
    Q5_K
    [6144, 2048]
  • blk.40.attn_q_a_norm.weight
    F32
    [2048]
  • blk.40.attn_q_b.weight
    Q8_0
    [2048, 16384]
  • blk.40.attn_v_b.weight
    Q8_0
    [512, 256, 64]
  • blk.40.exp_probs_b.bias
    F32
    [256]
  • blk.40.ffn_down_exps.weight
    IQ3_XXS
    [2048, 6144, 256]
  • blk.40.ffn_down_shexp.weight
    Q6_K
    [2048, 6144]
  • blk.40.ffn_gate_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.40.ffn_gate_inp.weight
    F32
    [6144, 256]
  • blk.40.ffn_gate_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.40.ffn_norm.weight
    F32
    [6144]
  • blk.40.ffn_up_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.40.ffn_up_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.40.indexer.attn_k.weight
    Q8_0
    [6144, 128]
  • blk.40.indexer.attn_q_b.weight
    Q8_0
    [2048, 4096]
  • blk.40.indexer.k_norm.bias
    F32
    [128]
  • blk.40.indexer.k_norm.weight
    F32
    [128]
  • blk.40.indexer.proj.weight
    F32
    [6144, 32]
    • blk.41
  • blk.41.attn_k_b.weight
    Q8_0
    [192, 512, 64]
  • blk.41.attn_kv_a_mqa.weight
    Q8_0
    [6144, 576]
  • blk.41.attn_kv_a_norm.weight
    F32
    [512]
  • blk.41.attn_norm.weight
    F32
    [6144]
  • blk.41.attn_output.weight
    Q5_K
    [16384, 6144]
  • blk.41.attn_q_a.weight
    Q5_K
    [6144, 2048]
  • blk.41.attn_q_a_norm.weight
    F32
    [2048]
  • blk.41.attn_q_b.weight
    Q8_0
    [2048, 16384]
  • blk.41.attn_v_b.weight
    Q8_0
    [512, 256, 64]
  • blk.41.exp_probs_b.bias
    F32
    [256]
  • blk.41.ffn_down_exps.weight
    IQ3_XXS
    [2048, 6144, 256]
  • blk.41.ffn_down_shexp.weight
    Q6_K
    [2048, 6144]
  • blk.41.ffn_gate_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.41.ffn_gate_inp.weight
    F32
    [6144, 256]
  • blk.41.ffn_gate_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.41.ffn_norm.weight
    F32
    [6144]
  • blk.41.ffn_up_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.41.ffn_up_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.41.indexer.attn_k.weight
    Q8_0
    [6144, 128]
  • blk.41.indexer.attn_q_b.weight
    Q8_0
    [2048, 4096]
  • blk.41.indexer.k_norm.bias
    F32
    [128]
  • blk.41.indexer.k_norm.weight
    F32
    [128]
  • blk.41.indexer.proj.weight
    F32
    [6144, 32]
    • blk.42
  • blk.42.attn_k_b.weight
    Q8_0
    [192, 512, 64]
  • blk.42.attn_kv_a_mqa.weight
    Q8_0
    [6144, 576]
  • blk.42.attn_kv_a_norm.weight
    F32
    [512]
  • blk.42.attn_norm.weight
    F32
    [6144]
  • blk.42.attn_output.weight
    Q5_K
    [16384, 6144]
  • blk.42.attn_q_a.weight
    Q5_K
    [6144, 2048]
  • blk.42.attn_q_a_norm.weight
    F32
    [2048]
  • blk.42.attn_q_b.weight
    Q8_0
    [2048, 16384]
  • blk.42.attn_v_b.weight
    Q8_0
    [512, 256, 64]
  • blk.42.exp_probs_b.bias
    F32
    [256]
  • blk.42.ffn_down_exps.weight
    IQ3_XXS
    [2048, 6144, 256]
  • blk.42.ffn_down_shexp.weight
    Q6_K
    [2048, 6144]
  • blk.42.ffn_gate_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.42.ffn_gate_inp.weight
    F32
    [6144, 256]
  • blk.42.ffn_gate_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.42.ffn_norm.weight
    F32
    [6144]
  • blk.42.ffn_up_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.42.ffn_up_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.42.indexer.attn_k.weight
    Q8_0
    [6144, 128]
  • blk.42.indexer.attn_q_b.weight
    Q8_0
    [2048, 4096]
  • blk.42.indexer.k_norm.bias
    F32
    [128]
  • blk.42.indexer.k_norm.weight
    F32
    [128]
  • blk.42.indexer.proj.weight
    F32
    [6144, 32]
    • blk.43
  • blk.43.attn_k_b.weight
    Q8_0
    [192, 512, 64]
  • blk.43.attn_kv_a_mqa.weight
    Q8_0
    [6144, 576]
  • blk.43.attn_kv_a_norm.weight
    F32
    [512]
  • blk.43.attn_norm.weight
    F32
    [6144]
  • blk.43.attn_output.weight
    Q5_K
    [16384, 6144]
  • blk.43.attn_q_a.weight
    Q5_K
    [6144, 2048]
  • blk.43.attn_q_a_norm.weight
    F32
    [2048]
  • blk.43.attn_q_b.weight
    Q8_0
    [2048, 16384]
  • blk.43.attn_v_b.weight
    Q8_0
    [512, 256, 64]
  • blk.43.exp_probs_b.bias
    F32
    [256]
  • blk.43.ffn_down_exps.weight
    IQ3_XXS
    [2048, 6144, 256]
  • blk.43.ffn_down_shexp.weight
    Q6_K
    [2048, 6144]
  • blk.43.ffn_gate_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.43.ffn_gate_inp.weight
    F32
    [6144, 256]
  • blk.43.ffn_gate_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.43.ffn_norm.weight
    F32
    [6144]
  • blk.43.ffn_up_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.43.ffn_up_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.43.indexer.attn_k.weight
    Q8_0
    [6144, 128]
  • blk.43.indexer.attn_q_b.weight
    Q8_0
    [2048, 4096]
  • blk.43.indexer.k_norm.bias
    F32
    [128]
  • blk.43.indexer.k_norm.weight
    F32
    [128]
  • blk.43.indexer.proj.weight
    F32
    [6144, 32]
    • blk.44
  • blk.44.attn_k_b.weight
    Q8_0
    [192, 512, 64]
  • blk.44.attn_kv_a_mqa.weight
    Q8_0
    [6144, 576]
  • blk.44.attn_kv_a_norm.weight
    F32
    [512]
  • blk.44.attn_norm.weight
    F32
    [6144]
  • blk.44.attn_output.weight
    Q5_K
    [16384, 6144]
  • blk.44.attn_q_a.weight
    Q5_K
    [6144, 2048]
  • blk.44.attn_q_a_norm.weight
    F32
    [2048]
  • blk.44.attn_q_b.weight
    Q8_0
    [2048, 16384]
  • blk.44.attn_v_b.weight
    Q8_0
    [512, 256, 64]
  • blk.44.exp_probs_b.bias
    F32
    [256]
  • blk.44.ffn_down_exps.weight
    IQ3_XXS
    [2048, 6144, 256]
  • blk.44.ffn_down_shexp.weight
    Q6_K
    [2048, 6144]
  • blk.44.ffn_gate_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.44.ffn_gate_inp.weight
    F32
    [6144, 256]
  • blk.44.ffn_gate_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.44.ffn_norm.weight
    F32
    [6144]
  • blk.44.ffn_up_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.44.ffn_up_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.44.indexer.attn_k.weight
    Q8_0
    [6144, 128]
  • blk.44.indexer.attn_q_b.weight
    Q8_0
    [2048, 4096]
  • blk.44.indexer.k_norm.bias
    F32
    [128]
  • blk.44.indexer.k_norm.weight
    F32
    [128]
  • blk.44.indexer.proj.weight
    F32
    [6144, 32]
    • blk.45
  • blk.45.attn_k_b.weight
    Q8_0
    [192, 512, 64]
  • blk.45.attn_kv_a_mqa.weight
    Q8_0
    [6144, 576]
  • blk.45.attn_kv_a_norm.weight
    F32
    [512]
  • blk.45.attn_norm.weight
    F32
    [6144]
  • blk.45.attn_output.weight
    Q5_K
    [16384, 6144]
  • blk.45.attn_q_a.weight
    Q5_K
    [6144, 2048]
  • blk.45.attn_q_a_norm.weight
    F32
    [2048]
  • blk.45.attn_q_b.weight
    Q8_0
    [2048, 16384]
  • blk.45.attn_v_b.weight
    Q8_0
    [512, 256, 64]
  • blk.45.exp_probs_b.bias
    F32
    [256]
  • blk.45.ffn_down_exps.weight
    IQ3_XXS
    [2048, 6144, 256]
  • blk.45.ffn_down_shexp.weight
    Q6_K
    [2048, 6144]
  • blk.45.ffn_gate_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.45.ffn_gate_inp.weight
    F32
    [6144, 256]
  • blk.45.ffn_gate_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.45.ffn_norm.weight
    F32
    [6144]
  • blk.45.ffn_up_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.45.ffn_up_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.45.indexer.attn_k.weight
    Q8_0
    [6144, 128]
  • blk.45.indexer.attn_q_b.weight
    Q8_0
    [2048, 4096]
  • blk.45.indexer.k_norm.bias
    F32
    [128]
  • blk.45.indexer.k_norm.weight
    F32
    [128]
  • blk.45.indexer.proj.weight
    F32
    [6144, 32]
    • blk.46
  • blk.46.attn_k_b.weight
    Q8_0
    [192, 512, 64]
  • blk.46.attn_kv_a_mqa.weight
    Q8_0
    [6144, 576]
  • blk.46.attn_kv_a_norm.weight
    F32
    [512]
  • blk.46.attn_norm.weight
    F32
    [6144]
  • blk.46.attn_output.weight
    Q5_K
    [16384, 6144]
  • blk.46.attn_q_a.weight
    Q5_K
    [6144, 2048]
  • blk.46.attn_q_a_norm.weight
    F32
    [2048]
  • blk.46.attn_q_b.weight
    Q8_0
    [2048, 16384]
  • blk.46.attn_v_b.weight
    Q8_0
    [512, 256, 64]
  • blk.46.exp_probs_b.bias
    F32
    [256]
  • blk.46.ffn_down_exps.weight
    IQ3_XXS
    [2048, 6144, 256]
  • blk.46.ffn_down_shexp.weight
    Q6_K
    [2048, 6144]
  • blk.46.ffn_gate_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.46.ffn_gate_inp.weight
    F32
    [6144, 256]
  • blk.46.ffn_gate_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.46.ffn_norm.weight
    F32
    [6144]
  • blk.46.ffn_up_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.46.ffn_up_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.46.indexer.attn_k.weight
    Q8_0
    [6144, 128]
  • blk.46.indexer.attn_q_b.weight
    Q8_0
    [2048, 4096]
  • blk.46.indexer.k_norm.bias
    F32
    [128]
  • blk.46.indexer.k_norm.weight
    F32
    [128]
  • blk.46.indexer.proj.weight
    F32
    [6144, 32]
    • blk.47
  • blk.47.attn_k_b.weight
    Q8_0
    [192, 512, 64]
  • blk.47.attn_kv_a_mqa.weight
    Q8_0
    [6144, 576]
  • blk.47.attn_kv_a_norm.weight
    F32
    [512]
  • blk.47.attn_norm.weight
    F32
    [6144]
  • blk.47.attn_output.weight
    Q5_K
    [16384, 6144]
  • blk.47.attn_q_a.weight
    Q5_K
    [6144, 2048]
  • blk.47.attn_q_a_norm.weight
    F32
    [2048]
  • blk.47.attn_q_b.weight
    Q8_0
    [2048, 16384]
  • blk.47.attn_v_b.weight
    Q8_0
    [512, 256, 64]
  • blk.47.exp_probs_b.bias
    F32
    [256]
  • blk.47.ffn_down_exps.weight
    IQ3_XXS
    [2048, 6144, 256]
  • blk.47.ffn_down_shexp.weight
    Q6_K
    [2048, 6144]
  • blk.47.ffn_gate_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.47.ffn_gate_inp.weight
    F32
    [6144, 256]
  • blk.47.ffn_gate_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.47.ffn_norm.weight
    F32
    [6144]
  • blk.47.ffn_up_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.47.ffn_up_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.47.indexer.attn_k.weight
    Q8_0
    [6144, 128]
  • blk.47.indexer.attn_q_b.weight
    Q8_0
    [2048, 4096]
  • blk.47.indexer.k_norm.bias
    F32
    [128]
  • blk.47.indexer.k_norm.weight
    F32
    [128]
  • blk.47.indexer.proj.weight
    F32
    [6144, 32]
    • blk.48
  • blk.48.attn_k_b.weight
    Q8_0
    [192, 512, 64]
  • blk.48.attn_kv_a_mqa.weight
    Q8_0
    [6144, 576]
  • blk.48.attn_kv_a_norm.weight
    F32
    [512]
  • blk.48.attn_norm.weight
    F32
    [6144]
  • blk.48.attn_output.weight
    Q5_K
    [16384, 6144]
  • blk.48.attn_q_a.weight
    Q5_K
    [6144, 2048]
  • blk.48.attn_q_a_norm.weight
    F32
    [2048]
  • blk.48.attn_q_b.weight
    Q8_0
    [2048, 16384]
  • blk.48.attn_v_b.weight
    Q8_0
    [512, 256, 64]
  • blk.48.exp_probs_b.bias
    F32
    [256]
  • blk.48.ffn_down_exps.weight
    IQ3_XXS
    [2048, 6144, 256]
  • blk.48.ffn_down_shexp.weight
    Q6_K
    [2048, 6144]
  • blk.48.ffn_gate_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.48.ffn_gate_inp.weight
    F32
    [6144, 256]
  • blk.48.ffn_gate_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.48.ffn_norm.weight
    F32
    [6144]
  • blk.48.ffn_up_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.48.ffn_up_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.48.indexer.attn_k.weight
    Q8_0
    [6144, 128]
  • blk.48.indexer.attn_q_b.weight
    Q8_0
    [2048, 4096]
  • blk.48.indexer.k_norm.bias
    F32
    [128]
  • blk.48.indexer.k_norm.weight
    F32
    [128]
  • blk.48.indexer.proj.weight
    F32
    [6144, 32]
    • blk.49
  • blk.49.attn_k_b.weight
    Q8_0
    [192, 512, 64]
  • blk.49.attn_kv_a_mqa.weight
    Q8_0
    [6144, 576]
  • blk.49.attn_kv_a_norm.weight
    F32
    [512]
  • blk.49.attn_norm.weight
    F32
    [6144]
  • blk.49.attn_output.weight
    Q5_K
    [16384, 6144]
  • blk.49.attn_q_a.weight
    Q5_K
    [6144, 2048]
  • blk.49.attn_q_a_norm.weight
    F32
    [2048]
  • blk.49.attn_q_b.weight
    Q8_0
    [2048, 16384]
  • blk.49.attn_v_b.weight
    Q8_0
    [512, 256, 64]
  • blk.49.exp_probs_b.bias
    F32
    [256]
  • blk.49.ffn_down_exps.weight
    IQ3_XXS
    [2048, 6144, 256]
  • blk.49.ffn_down_shexp.weight
    Q6_K
    [2048, 6144]
  • blk.49.ffn_gate_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.49.ffn_gate_inp.weight
    F32
    [6144, 256]
  • blk.49.ffn_gate_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.49.ffn_norm.weight
    F32
    [6144]
  • blk.49.ffn_up_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.49.ffn_up_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.49.indexer.attn_k.weight
    Q8_0
    [6144, 128]
  • blk.49.indexer.attn_q_b.weight
    Q8_0
    [2048, 4096]
  • blk.49.indexer.k_norm.bias
    F32
    [128]
  • blk.49.indexer.k_norm.weight
    F32
    [128]
  • blk.49.indexer.proj.weight
    F32
    [6144, 32]
    • blk.50
  • blk.50.attn_k_b.weight
    Q8_0
    [192, 512, 64]
  • blk.50.attn_kv_a_mqa.weight
    Q8_0
    [6144, 576]
  • blk.50.attn_kv_a_norm.weight
    F32
    [512]
  • blk.50.attn_norm.weight
    F32
    [6144]
  • blk.50.attn_output.weight
    Q5_K
    [16384, 6144]
  • blk.50.attn_q_a.weight
    Q5_K
    [6144, 2048]
  • blk.50.attn_q_a_norm.weight
    F32
    [2048]
  • blk.50.attn_q_b.weight
    Q8_0
    [2048, 16384]
  • blk.50.attn_v_b.weight
    Q8_0
    [512, 256, 64]
  • blk.50.exp_probs_b.bias
    F32
    [256]
  • blk.50.ffn_down_exps.weight
    IQ3_XXS
    [2048, 6144, 256]
  • blk.50.ffn_down_shexp.weight
    Q6_K
    [2048, 6144]
  • blk.50.ffn_gate_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.50.ffn_gate_inp.weight
    F32
    [6144, 256]
  • blk.50.ffn_gate_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.50.ffn_norm.weight
    F32
    [6144]
  • blk.50.ffn_up_exps.weight
    IQ2_XXS
    [6144, 2048, 256]
  • blk.50.ffn_up_shexp.weight
    Q5_K
    [6144, 2048]
  • blk.50.indexer.attn_k.weight
    Q8_0
    [6144, 128]
  • blk.50.indexer.attn_q_b.weight
    Q8_0
    [2048, 4096]
  • blk.50.indexer.k_norm.bias
    F32
    [128]
  • blk.50.indexer.k_norm.weight
    F32
    [128]
  • blk.50.indexer.proj.weight
    F32
    [6144, 32]
    • blk.51
  • blk.51.attn_k_b.weight
    Q8_0
    [192, 512, 64]