3 5 days ago

Requant to IQ4_NL, with combined_en_huge as prompt (English + Coding focus)

0f45b1445b2a ยท 12GB
    Metadata
  • split.count
    4
  • split.no
    3
  • split.tensors.count
    1224
  • Tensor
    blk.41
  • blk.41.ffn_down_exps.weight
    Q5_1
    [640, 2560, 512]
  • blk.41.ffn_down_shexp.weight
    Q8_0
    [640, 2560]
  • blk.41.ffn_gate_exps.weight
    Q4_K
    [2560, 640, 512]
  • blk.41.ffn_gate_inp.weight
    F32
    [2560, 512]
  • blk.41.ffn_gate_inp_shexp.weight
    F32
    [2560]
  • blk.41.ffn_gate_shexp.weight
    Q8_0
    [2560, 640]
  • blk.41.ffn_up_exps.weight
    Q4_K
    [2560, 640, 512]
  • blk.41.ffn_up_shexp.weight
    Q8_0
    [2560, 640]
  • blk.41.hc_attn_down.weight
    Q8_0
    [10240, 320]
  • blk.41.hc_attn_inject.weight
    F32
    [10240, 4]
  • blk.41.hc_attn_norm.weight
    F32
    [10240]
  • blk.41.hc_attn_up.weight
    Q8_0
    [320, 10240]
  • blk.41.hc_ffn_down.weight
    Q8_0
    [10240, 320]
  • blk.41.hc_ffn_inject.weight
    F32
    [10240, 4]
  • blk.41.hc_ffn_norm.weight
    F32
    [10240]
  • blk.41.hc_ffn_up.weight
    Q8_0
    [320, 10240]
  • blk.41.ssm_a
    F32
    [48]
  • blk.41.ssm_alpha.weight
    F32
    [2560, 48]
  • blk.41.ssm_beta.weight
    F32
    [2560, 48]
  • blk.41.ssm_conv1d.weight
    F32
    [4, 10240]
  • blk.41.ssm_dt.bias
    F32
    [48]
  • blk.41.ssm_norm.weight
    F32
    [128]
  • blk.41.ssm_out.weight
    Q8_0
    [6144, 2560]
  • blk.42
  • blk.42.attn_gate.weight
    Q8_0
    [2560, 6144]
  • blk.42.attn_qkv.weight
    Q8_0
    [2560, 10240]
  • blk.42.ffn_down_exps.weight
    Q5_1
    [640, 2560, 512]
  • blk.42.ffn_down_shexp.weight
    Q8_0
    [640, 2560]
  • blk.42.ffn_gate_exps.weight
    Q4_K
    [2560, 640, 512]
  • blk.42.ffn_gate_inp.weight
    F32
    [2560, 512]
  • blk.42.ffn_gate_inp_shexp.weight
    F32
    [2560]
  • blk.42.ffn_gate_shexp.weight
    Q8_0
    [2560, 640]
  • blk.42.ffn_up_exps.weight
    Q4_K
    [2560, 640, 512]
  • blk.42.ffn_up_shexp.weight
    Q8_0
    [2560, 640]
  • blk.42.hc_attn_down.weight
    Q8_0
    [10240, 320]
  • blk.42.hc_attn_inject.weight
    F32
    [10240, 4]
  • blk.42.hc_attn_norm.weight
    F32
    [10240]
  • blk.42.hc_attn_up.weight
    Q8_0
    [320, 10240]
  • blk.42.hc_ffn_down.weight
    Q8_0
    [10240, 320]
  • blk.42.hc_ffn_inject.weight
    F32
    [10240, 4]
  • blk.42.hc_ffn_norm.weight
    F32
    [10240]
  • blk.42.hc_ffn_up.weight
    Q8_0
    [320, 10240]
  • blk.42.ssm_a
    F32
    [48]
  • blk.42.ssm_alpha.weight
    F32
    [2560, 48]
  • blk.42.ssm_beta.weight
    F32
    [2560, 48]
  • blk.42.ssm_conv1d.weight
    F32
    [4, 10240]
  • blk.42.ssm_dt.bias
    F32
    [48]
  • blk.42.ssm_norm.weight
    F32
    [128]
  • blk.42.ssm_out.weight
    Q8_0
    [6144, 2560]
  • blk.43
  • blk.43.attn_k.weight
    Q8_0
    [2560, 512]
  • blk.43.attn_k_norm.weight
    F32
    [256]
  • blk.43.attn_output.weight
    Q8_0
    [6144, 2560]
  • blk.43.attn_q.weight
    Q8_0
    [2560, 12288]
  • blk.43.attn_q_norm.weight
    F32
    [256]
  • blk.43.attn_v.weight
    Q8_0
    [2560, 512]
  • blk.43.ffn_down_exps.weight
    Q5_1
    [640, 2560, 512]
  • blk.43.ffn_down_shexp.weight
    Q8_0
    [640, 2560]
  • blk.43.ffn_gate_exps.weight
    Q4_K
    [2560, 640, 512]
  • blk.43.ffn_gate_inp.weight
    F32
    [2560, 512]
  • blk.43.ffn_gate_inp_shexp.weight
    F32
    [2560]
  • blk.43.ffn_gate_shexp.weight
    Q8_0
    [2560, 640]
  • blk.43.ffn_up_exps.weight
    Q4_K
    [2560, 640, 512]
  • blk.43.ffn_up_shexp.weight
    Q8_0
    [2560, 640]
  • blk.43.hc_attn_down.weight
    Q8_0
    [10240, 320]
  • blk.43.hc_attn_inject.weight
    F32
    [10240, 4]
  • blk.43.hc_attn_norm.weight
    F32
    [10240]
  • blk.43.hc_attn_up.weight
    Q8_0
    [320, 10240]
  • blk.43.hc_ffn_down.weight
    Q8_0
    [10240, 320]
  • blk.43.hc_ffn_inject.weight
    F32
    [10240, 4]
  • blk.43.hc_ffn_norm.weight
    F32
    [10240]
  • blk.43.hc_ffn_up.weight
    Q8_0
    [320, 10240]
  • blk.43.indexer.k_norm.weight
    F32
    [128]
  • blk.43.indexer.k_proj.weight
    BF16
    [2560, 128]
  • blk.43.indexer.q_norm.weight
    F32
    [128]
  • blk.43.indexer.q_proj.weight
    BF16
    [2560, 512]
  • blk.44
  • blk.44.attn_gate.weight
    Q8_0
    [2560, 6144]
  • blk.44.attn_qkv.weight
    Q8_0
    [2560, 10240]
  • blk.44.ffn_down_exps.weight
    Q5_1
    [640, 2560, 512]
  • blk.44.ffn_down_shexp.weight
    Q8_0
    [640, 2560]
  • blk.44.ffn_gate_exps.weight
    Q4_K
    [2560, 640, 512]
  • blk.44.ffn_gate_inp.weight
    F32
    [2560, 512]
  • blk.44.ffn_gate_inp_shexp.weight
    F32
    [2560]
  • blk.44.ffn_gate_shexp.weight
    Q8_0
    [2560, 640]
  • blk.44.ffn_up_exps.weight
    Q4_K
    [2560, 640, 512]
  • blk.44.ffn_up_shexp.weight
    Q8_0
    [2560, 640]
  • blk.44.hc_attn_down.weight
    Q8_0
    [10240, 320]
  • blk.44.hc_attn_inject.weight
    F32
    [10240, 4]
  • blk.44.hc_attn_norm.weight
    F32
    [10240]
  • blk.44.hc_attn_up.weight
    Q8_0
    [320, 10240]
  • blk.44.hc_ffn_down.weight
    Q8_0
    [10240, 320]
  • blk.44.hc_ffn_inject.weight
    F32
    [10240, 4]
  • blk.44.hc_ffn_norm.weight
    F32
    [10240]
  • blk.44.hc_ffn_up.weight
    Q8_0
    [320, 10240]
  • blk.44.ssm_a
    F32
    [48]
  • blk.44.ssm_alpha.weight
    F32
    [2560, 48]
  • blk.44.ssm_beta.weight
    F32
    [2560, 48]
  • blk.44.ssm_conv1d.weight
    F32
    [4, 10240]
  • blk.44.ssm_dt.bias
    F32
    [48]
  • blk.44.ssm_norm.weight
    F32
    [128]
  • blk.44.ssm_out.weight
    Q8_0
    [6144, 2560]
  • blk.45
  • blk.45.attn_gate.weight
    Q8_0
    [2560, 6144]
  • blk.45.attn_qkv.weight
    Q8_0
    [2560, 10240]
  • blk.45.ffn_down_exps.weight
    Q5_1
    [640, 2560, 512]
  • blk.45.ffn_down_shexp.weight
    Q8_0
    [640, 2560]
  • blk.45.ffn_gate_exps.weight
    Q4_K
    [2560, 640, 512]
  • blk.45.ffn_gate_inp.weight
    F32
    [2560, 512]
  • blk.45.ffn_gate_inp_shexp.weight
    F32
    [2560]
  • blk.45.ffn_gate_shexp.weight
    Q8_0
    [2560, 640]
  • blk.45.ffn_up_exps.weight
    Q4_K
    [2560, 640, 512]
  • blk.45.ffn_up_shexp.weight
    Q8_0
    [2560, 640]
  • blk.45.hc_attn_down.weight
    Q8_0
    [10240, 320]
  • blk.45.hc_attn_inject.weight
    F32
    [10240, 4]
  • blk.45.hc_attn_norm.weight
    F32
    [10240]
  • blk.45.hc_attn_up.weight
    Q8_0
    [320, 10240]
  • blk.45.hc_ffn_down.weight
    Q8_0
    [10240, 320]
  • blk.45.hc_ffn_inject.weight
    F32
    [10240, 4]
  • blk.45.hc_ffn_norm.weight
    F32
    [10240]
  • blk.45.hc_ffn_up.weight
    Q8_0
    [320, 10240]
  • blk.45.ssm_a
    F32
    [48]
  • blk.45.ssm_alpha.weight
    F32
    [2560, 48]
  • blk.45.ssm_beta.weight
    F32
    [2560, 48]
  • blk.45.ssm_conv1d.weight
    F32
    [4, 10240]
  • blk.45.ssm_dt.bias
    F32
    [48]
  • blk.45.ssm_norm.weight
    F32
    [128]
  • blk.45.ssm_out.weight
    Q8_0
    [6144, 2560]
  • blk.46
  • blk.46.attn_gate.weight
    Q8_0
    [2560, 6144]
  • blk.46.attn_qkv.weight
    Q8_0
    [2560, 10240]
  • blk.46.ffn_down_exps.weight
    Q8_0
    [640, 2560, 512]
  • blk.46.ffn_down_shexp.weight
    Q8_0
    [640, 2560]
  • blk.46.ffn_gate_exps.weight
    Q4_K
    [2560, 640, 512]
  • blk.46.ffn_gate_inp.weight
    F32
    [2560, 512]
  • blk.46.ffn_gate_inp_shexp.weight
    F32
    [2560]
  • blk.46.ffn_gate_shexp.weight
    Q8_0
    [2560, 640]
  • blk.46.ffn_up_exps.weight
    Q4_K
    [2560, 640, 512]
  • blk.46.ffn_up_shexp.weight
    Q8_0
    [2560, 640]
  • blk.46.hc_attn_down.weight
    Q8_0
    [10240, 320]
  • blk.46.hc_attn_inject.weight
    F32
    [10240, 4]
  • blk.46.hc_attn_norm.weight
    F32
    [10240]
  • blk.46.hc_attn_up.weight
    Q8_0
    [320, 10240]
  • blk.46.hc_ffn_down.weight
    Q8_0
    [10240, 320]
  • blk.46.hc_ffn_inject.weight
    F32
    [10240, 4]
  • blk.46.hc_ffn_norm.weight
    F32
    [10240]
  • blk.46.hc_ffn_up.weight
    Q8_0
    [320, 10240]
  • blk.46.ssm_a
    F32
    [48]
  • blk.46.ssm_alpha.weight
    F32
    [2560, 48]
  • blk.46.ssm_beta.weight
    F32
    [2560, 48]
  • blk.46.ssm_conv1d.weight
    F32
    [4, 10240]
  • blk.46.ssm_dt.bias
    F32
    [48]
  • blk.46.ssm_norm.weight
    F32
    [128]
  • blk.46.ssm_out.weight
    Q8_0
    [6144, 2560]
  • blk.47
  • blk.47.attn_k.weight
    Q8_0
    [2560, 512]
  • blk.47.attn_k_norm.weight
    F32
    [256]
  • blk.47.attn_output.weight
    Q8_0
    [6144, 2560]
  • blk.47.attn_q.weight
    Q8_0
    [2560, 12288]
  • blk.47.attn_q_norm.weight
    F32
    [256]
  • blk.47.attn_v.weight
    Q8_0
    [2560, 512]
  • blk.47.ffn_down_exps.weight
    Q8_0
    [640, 2560, 512]
  • blk.47.ffn_down_shexp.weight
    Q8_0
    [640, 2560]
  • blk.47.ffn_gate_exps.weight
    Q4_K
    [2560, 640, 512]
  • blk.47.ffn_gate_inp.weight
    F32
    [2560, 512]
  • blk.47.ffn_gate_inp_shexp.weight
    F32
    [2560]
  • blk.47.ffn_gate_shexp.weight
    Q8_0
    [2560, 640]
  • blk.47.ffn_up_exps.weight
    Q4_K
    [2560, 640, 512]
  • blk.47.ffn_up_shexp.weight
    Q8_0
    [2560, 640]
  • blk.47.hc_attn_down.weight
    Q8_0
    [10240, 320]
  • blk.47.hc_attn_inject.weight
    F32
    [10240, 4]
  • blk.47.hc_attn_norm.weight
    F32
    [10240]
  • blk.47.hc_attn_up.weight
    Q8_0
    [320, 10240]
  • blk.47.hc_ffn_down.weight
    Q8_0
    [10240, 320]
  • blk.47.hc_ffn_inject.weight
    F32
    [10240, 4]
  • blk.47.hc_ffn_norm.weight
    F32
    [10240]
  • blk.47.hc_ffn_up.weight
    Q8_0
    [320, 10240]
  • blk.47.indexer.k_norm.weight
    F32
    [128]
  • blk.47.indexer.k_proj.weight
    BF16
    [2560, 128]
  • blk.47.indexer.q_norm.weight
    F32
    [128]
  • blk.47.indexer.q_proj.weight
    BF16
    [2560, 512]