3 5 days ago

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

8d2a48cace87 ยท 50GB
    Metadata
  • split.count
    4
  • split.no
    1
  • split.tensors.count
    1224
  • Tensor
  • token_embd.weight
    Q8_0
    [2560, 248320]
  • blk.0
  • blk.0.attn_gate.weight
    Q8_0
    [2560, 6144]
  • blk.0.attn_qkv.weight
    Q8_0
    [2560, 10240]
  • blk.0.ffn_down_exps.weight
    Q5_1
    [640, 2560, 512]
  • blk.0.ffn_down_shexp.weight
    Q8_0
    [640, 2560]
  • blk.0.ffn_gate_exps.weight
    Q4_K
    [2560, 640, 512]
  • blk.0.ffn_gate_inp.weight
    F32
    [2560, 512]
  • blk.0.ffn_gate_inp_shexp.weight
    F32
    [2560]
  • blk.0.ffn_gate_shexp.weight
    Q8_0
    [2560, 640]
  • blk.0.ffn_up_exps.weight
    Q4_K
    [2560, 640, 512]
  • blk.0.ffn_up_shexp.weight
    Q8_0
    [2560, 640]
  • blk.0.hc_attn_down.weight
    Q8_0
    [10240, 320]
  • blk.0.hc_attn_inject.weight
    F32
    [10240, 4]
  • blk.0.hc_attn_norm.weight
    F32
    [10240]
  • blk.0.hc_attn_up.weight
    Q8_0
    [320, 10240]
  • blk.0.hc_ffn_down.weight
    Q8_0
    [10240, 320]
  • blk.0.hc_ffn_inject.weight
    F32
    [10240, 4]
  • blk.0.hc_ffn_norm.weight
    F32
    [10240]
  • blk.0.hc_ffn_up.weight
    Q8_0
    [320, 10240]
  • blk.0.ssm_a
    F32
    [48]
  • blk.0.ssm_alpha.weight
    F32
    [2560, 48]
  • blk.0.ssm_beta.weight
    F32
    [2560, 48]
  • blk.0.ssm_conv1d.weight
    F32
    [4, 10240]
  • blk.0.ssm_dt.bias
    F32
    [48]
  • blk.0.ssm_norm.weight
    F32
    [128]
  • blk.0.ssm_out.weight
    Q8_0
    [6144, 2560]
  • blk.1
  • blk.1.attn_gate.weight
    Q8_0
    [2560, 6144]
  • blk.1.attn_qkv.weight
    Q8_0
    [2560, 10240]
  • blk.1.ffn_down_exps.weight
    Q5_1
    [640, 2560, 512]
  • blk.1.ffn_down_shexp.weight
    Q8_0
    [640, 2560]
  • blk.1.ffn_gate_exps.weight
    Q4_K
    [2560, 640, 512]
  • blk.1.ffn_gate_inp.weight
    F32
    [2560, 512]
  • blk.1.ffn_gate_inp_shexp.weight
    F32
    [2560]
  • blk.1.ffn_gate_shexp.weight
    Q8_0
    [2560, 640]
  • blk.1.ffn_up_exps.weight
    Q4_K
    [2560, 640, 512]
  • blk.1.ffn_up_shexp.weight
    Q8_0
    [2560, 640]
  • blk.1.hc_attn_down.weight
    Q8_0
    [10240, 320]
  • blk.1.hc_attn_inject.weight
    F32
    [10240, 4]
  • blk.1.hc_attn_norm.weight
    F32
    [10240]
  • blk.1.hc_attn_up.weight
    Q8_0
    [320, 10240]
  • blk.1.hc_ffn_down.weight
    Q8_0
    [10240, 320]
  • blk.1.hc_ffn_inject.weight
    F32
    [10240, 4]
  • blk.1.hc_ffn_norm.weight
    F32
    [10240]
  • blk.1.hc_ffn_up.weight
    Q8_0
    [320, 10240]
  • blk.1.ple_conv1d.weight
    F32
    [4, 10240]
  • blk.1.ple_key.weight
    Q8_0
    [2560, 10240]
  • blk.1.ple_norm_conv.weight
    F32
    [10240]
  • blk.1.ple_norm_key.weight
    F32
    [10240]
  • blk.1.ple_norm_query.weight
    F32
    [10240]
  • blk.1.ple_value.weight
    Q8_0
    [2560, 2560]
  • blk.1.ssm_a
    F32
    [48]
  • blk.1.ssm_alpha.weight
    F32
    [2560, 48]
  • blk.1.ssm_beta.weight
    F32
    [2560, 48]
  • blk.1.ssm_conv1d.weight
    F32
    [4, 10240]
  • blk.1.ssm_dt.bias
    F32
    [48]
  • blk.1.ssm_norm.weight
    F32
    [128]
  • blk.1.ssm_out.weight
    Q8_0
    [6144, 2560]
  • blk.2
  • blk.2.attn_gate.weight
    Q8_0
    [2560, 6144]
  • blk.2.attn_qkv.weight
    Q8_0
    [2560, 10240]
  • blk.2.ffn_down_exps.weight
    Q8_0
    [640, 2560, 512]
  • blk.2.ffn_down_shexp.weight
    Q8_0
    [640, 2560]
  • blk.2.ffn_gate_exps.weight
    Q5_K
    [2560, 640, 512]
  • blk.2.ffn_gate_inp.weight
    F32
    [2560, 512]
  • blk.2.ffn_gate_inp_shexp.weight
    F32
    [2560]
  • blk.2.ffn_gate_shexp.weight
    Q8_0
    [2560, 640]
  • blk.2.ffn_up_exps.weight
    Q5_K
    [2560, 640, 512]
  • blk.2.ffn_up_shexp.weight
    Q8_0
    [2560, 640]
  • blk.2.hc_attn_down.weight
    Q8_0
    [10240, 320]
  • blk.2.hc_attn_inject.weight
    F32
    [10240, 4]
  • blk.2.hc_attn_norm.weight
    F32
    [10240]
  • blk.2.hc_attn_up.weight
    Q8_0
    [320, 10240]
  • blk.2.hc_ffn_down.weight
    Q8_0
    [10240, 320]
  • blk.2.hc_ffn_inject.weight
    F32
    [10240, 4]
  • blk.2.hc_ffn_norm.weight
    F32
    [10240]
  • blk.2.hc_ffn_up.weight
    Q8_0
    [320, 10240]
  • blk.2.ssm_a
    F32
    [48]
  • blk.2.ssm_alpha.weight
    F32
    [2560, 48]
  • blk.2.ssm_beta.weight
    F32
    [2560, 48]
  • blk.2.ssm_conv1d.weight
    F32
    [4, 10240]
  • blk.2.ssm_dt.bias
    F32
    [48]
  • blk.2.ssm_norm.weight
    F32
    [128]
  • blk.2.ssm_out.weight
    Q8_0
    [6144, 2560]
  • blk.3
  • blk.3.attn_k.weight
    Q8_0
    [2560, 512]
  • blk.3.attn_k_norm.weight
    F32
    [256]
  • blk.3.attn_output.weight
    Q8_0
    [6144, 2560]
  • blk.3.attn_q.weight
    Q8_0
    [2560, 12288]
  • blk.3.attn_q_norm.weight
    F32
    [256]
  • blk.3.attn_v.weight
    Q8_0
    [2560, 512]
  • blk.3.ffn_down_exps.weight
    Q5_1
    [640, 2560, 512]
  • blk.3.ffn_down_shexp.weight
    Q8_0
    [640, 2560]
  • blk.3.ffn_gate_exps.weight
    Q4_K
    [2560, 640, 512]
  • blk.3.ffn_gate_inp.weight
    F32
    [2560, 512]
  • blk.3.ffn_gate_inp_shexp.weight
    F32
    [2560]
  • blk.3.ffn_gate_shexp.weight
    Q8_0
    [2560, 640]
  • blk.3.ffn_up_exps.weight
    Q4_K
    [2560, 640, 512]
  • blk.3.ffn_up_shexp.weight
    Q8_0
    [2560, 640]
  • blk.3.hc_attn_down.weight
    Q8_0
    [10240, 320]
  • blk.3.hc_attn_inject.weight
    F32
    [10240, 4]
  • blk.3.hc_attn_norm.weight
    F32
    [10240]
  • blk.3.hc_attn_up.weight
    Q8_0
    [320, 10240]
  • blk.3.hc_ffn_down.weight
    Q8_0
    [10240, 320]
  • blk.3.hc_ffn_inject.weight
    F32
    [10240, 4]
  • blk.3.hc_ffn_norm.weight
    F32
    [10240]
  • blk.3.hc_ffn_up.weight
    Q8_0
    [320, 10240]
  • blk.3.indexer.k_norm.weight
    F32
    [128]
  • blk.3.indexer.k_proj.weight
    BF16
    [2560, 128]
  • blk.3.indexer.q_norm.weight
    F32
    [128]
  • blk.3.indexer.q_proj.weight
    BF16
    [2560, 512]
  • blk.4
  • blk.4.attn_gate.weight
    Q8_0
    [2560, 6144]
  • blk.4.attn_qkv.weight
    Q8_0
    [2560, 10240]
  • blk.4.ffn_down_exps.weight
    Q8_0
    [640, 2560, 512]
  • blk.4.ffn_down_shexp.weight
    Q8_0
    [640, 2560]
  • blk.4.ffn_gate_exps.weight
    Q4_K
    [2560, 640, 512]
  • blk.4.ffn_gate_inp.weight
    F32
    [2560, 512]
  • blk.4.ffn_gate_inp_shexp.weight
    F32
    [2560]
  • blk.4.ffn_gate_shexp.weight
    Q8_0
    [2560, 640]
  • blk.4.ffn_up_exps.weight
    Q4_K
    [2560, 640, 512]
  • blk.4.ffn_up_shexp.weight
    Q8_0
    [2560, 640]
  • blk.4.hc_attn_down.weight
    Q8_0
    [10240, 320]
  • blk.4.hc_attn_inject.weight
    F32
    [10240, 4]
  • blk.4.hc_attn_norm.weight
    F32
    [10240]
  • blk.4.hc_attn_up.weight
    Q8_0
    [320, 10240]
  • blk.4.hc_ffn_down.weight
    Q8_0
    [10240, 320]
  • blk.4.hc_ffn_inject.weight
    F32
    [10240, 4]
  • blk.4.hc_ffn_norm.weight
    F32
    [10240]
  • blk.4.hc_ffn_up.weight
    Q8_0
    [320, 10240]
  • blk.4.ssm_a
    F32
    [48]
  • blk.4.ssm_alpha.weight
    F32
    [2560, 48]
  • blk.4.ssm_beta.weight
    F32
    [2560, 48]
  • blk.4.ssm_conv1d.weight
    F32
    [4, 10240]
  • blk.4.ssm_dt.bias
    F32
    [48]
  • blk.4.ssm_norm.weight
    F32
    [128]
  • blk.4.ssm_out.weight
    Q8_0
    [6144, 2560]
  • blk.5
  • blk.5.attn_gate.weight
    Q8_0
    [2560, 6144]
  • blk.5.attn_qkv.weight
    Q8_0
    [2560, 10240]
  • blk.5.ffn_down_exps.weight
    Q5_1
    [640, 2560, 512]
  • blk.5.ffn_down_shexp.weight
    Q8_0
    [640, 2560]
  • blk.5.ffn_gate_exps.weight
    Q4_K
    [2560, 640, 512]
  • blk.5.ffn_gate_inp.weight
    F32
    [2560, 512]
  • blk.5.ffn_gate_inp_shexp.weight
    F32
    [2560]
  • blk.5.ffn_gate_shexp.weight
    Q8_0
    [2560, 640]
  • blk.5.ffn_up_exps.weight
    Q4_K
    [2560, 640, 512]
  • blk.5.ffn_up_shexp.weight
    Q8_0
    [2560, 640]
  • blk.5.hc_attn_down.weight
    Q8_0
    [10240, 320]
  • blk.5.hc_attn_inject.weight
    F32
    [10240, 4]
  • blk.5.hc_attn_norm.weight
    F32
    [10240]
  • blk.5.hc_attn_up.weight
    Q8_0
    [320, 10240]
  • blk.5.hc_ffn_down.weight
    Q8_0
    [10240, 320]
  • blk.5.hc_ffn_inject.weight
    F32
    [10240, 4]
  • blk.5.hc_ffn_norm.weight
    F32
    [10240]
  • blk.5.hc_ffn_up.weight
    Q8_0
    [320, 10240]
  • blk.5.ssm_a
    F32
    [48]
  • blk.5.ssm_alpha.weight
    F32
    [2560, 48]
  • blk.5.ssm_beta.weight
    F32
    [2560, 48]
  • blk.5.ssm_conv1d.weight
    F32
    [4, 10240]
  • blk.5.ssm_dt.bias
    F32
    [48]
  • blk.5.ssm_norm.weight
    F32
    [128]
  • blk.5.ssm_out.weight
    Q8_0
    [6144, 2560]
  • blk.6
  • blk.6.attn_gate.weight
    Q8_0
    [2560, 6144]
  • blk.6.attn_qkv.weight
    Q8_0
    [2560, 10240]
  • blk.6.ffn_down_exps.weight
    Q5_1
    [640, 2560, 512]
  • blk.6.ffn_down_shexp.weight
    Q8_0
    [640, 2560]
  • blk.6.ffn_gate_exps.weight
    Q4_K
    [2560, 640, 512]
  • blk.6.ffn_gate_inp.weight
    F32
    [2560, 512]
  • blk.6.ffn_gate_inp_shexp.weight
    F32
    [2560]
  • blk.6.ffn_gate_shexp.weight
    Q8_0
    [2560, 640]
  • blk.6.ffn_up_exps.weight
    Q4_K
    [2560, 640, 512]
  • blk.6.ffn_up_shexp.weight
    Q8_0
    [2560, 640]
  • blk.6.hc_attn_down.weight
    Q8_0
    [10240, 320]
  • blk.6.hc_attn_inject.weight
    F32
    [10240, 4]
  • blk.6.hc_attn_norm.weight
    F32
    [10240]
  • blk.6.hc_attn_up.weight
    Q8_0
    [320, 10240]
  • blk.6.hc_ffn_down.weight
    Q8_0
    [10240, 320]
  • blk.6.hc_ffn_inject.weight
    F32
    [10240, 4]
  • blk.6.hc_ffn_norm.weight
    F32
    [10240]
  • blk.6.hc_ffn_up.weight
    Q8_0
    [320, 10240]
  • blk.6.ssm_a
    F32
    [48]
  • blk.6.ssm_alpha.weight
    F32
    [2560, 48]
  • blk.6.ssm_beta.weight
    F32
    [2560, 48]
  • blk.6.ssm_conv1d.weight
    F32
    [4, 10240]
  • blk.6.ssm_dt.bias
    F32
    [48]
  • blk.6.ssm_norm.weight
    F32
    [128]
  • blk.6.ssm_out.weight
    Q8_0
    [6144, 2560]
  • blk.7
  • blk.7.attn_k.weight
    Q8_0
    [2560, 512]
  • blk.7.attn_k_norm.weight
    F32
    [256]
  • blk.7.attn_output.weight
    Q8_0
    [6144, 2560]
  • blk.7.attn_q.weight
    Q8_0
    [2560, 12288]
  • blk.7.attn_q_norm.weight
    F32
    [256]
  • blk.7.attn_v.weight
    Q8_0
    [2560, 512]
  • blk.7.ffn_down_exps.weight
    Q5_1
    [640, 2560, 512]
  • blk.7.ffn_down_shexp.weight
    Q8_0
    [640, 2560]
  • blk.7.ffn_gate_exps.weight
    Q4_K
    [2560, 640, 512]
  • blk.7.ffn_gate_inp.weight
    F32
    [2560, 512]
  • blk.7.ffn_gate_inp_shexp.weight
    F32
    [2560]
  • blk.7.ffn_gate_shexp.weight
    Q8_0
    [2560, 640]
  • blk.7.ffn_up_exps.weight
    Q4_K
    [2560, 640, 512]
  • blk.7.ffn_up_shexp.weight
    Q8_0
    [2560, 640]
  • blk.7.hc_attn_down.weight
    Q8_0
    [10240, 320]
  • blk.7.hc_attn_inject.weight
    F32
    [10240, 4]
  • blk.7.hc_attn_norm.weight
    F32
    [10240]
  • blk.7.hc_attn_up.weight
    Q8_0
    [320, 10240]
  • blk.7.hc_ffn_down.weight
    Q8_0
    [10240, 320]
  • blk.7.hc_ffn_inject.weight
    F32
    [10240, 4]
  • blk.7.hc_ffn_norm.weight
    F32
    [10240]
  • blk.7.hc_ffn_up.weight
    Q8_0
    [320, 10240]
  • blk.7.indexer.k_norm.weight
    F32
    [128]
  • blk.7.indexer.k_proj.weight
    BF16
    [2560, 128]
  • blk.7.indexer.q_norm.weight
    F32
    [128]
  • blk.7.indexer.q_proj.weight
    BF16
    [2560, 512]
  • blk.8
  • blk.8.attn_gate.weight
    Q8_0
    [2560, 6144]
  • blk.8.attn_qkv.weight
    Q8_0
    [2560, 10240]
  • blk.8.ffn_down_exps.weight
    Q5_1
    [640, 2560, 512]
  • blk.8.ffn_down_shexp.weight
    Q8_0
    [640, 2560]
  • blk.8.ffn_gate_exps.weight
    Q4_K
    [2560, 640, 512]
  • blk.8.ffn_gate_inp.weight
    F32
    [2560, 512]
  • blk.8.ffn_gate_inp_shexp.weight
    F32
    [2560]
  • blk.8.ffn_gate_shexp.weight
    Q8_0
    [2560, 640]
  • blk.8.ffn_up_exps.weight
    Q4_K
    [2560, 640, 512]
  • blk.8.ffn_up_shexp.weight
    Q8_0
    [2560, 640]
  • blk.8.hc_attn_down.weight
    Q8_0
    [10240, 320]
  • blk.8.hc_attn_inject.weight
    F32
    [10240, 4]
  • blk.8.hc_attn_norm.weight
    F32
    [10240]
  • blk.8.hc_attn_up.weight
    Q8_0
    [320, 10240]
  • blk.8.hc_ffn_down.weight
    Q8_0
    [10240, 320]
  • blk.8.hc_ffn_inject.weight
    F32
    [10240, 4]
  • blk.8.hc_ffn_norm.weight
    F32
    [10240]
  • blk.8.hc_ffn_up.weight
    Q8_0
    [320, 10240]
  • blk.8.ssm_a
    F32
    [48]
  • blk.8.ssm_alpha.weight
    F32
    [2560, 48]
  • blk.8.ssm_beta.weight
    F32
    [2560, 48]
  • blk.8.ssm_conv1d.weight
    F32
    [4, 10240]
  • blk.8.ssm_dt.bias
    F32
    [48]
  • blk.8.ssm_norm.weight
    F32
    [128]
  • blk.8.ssm_out.weight
    Q8_0
    [6144, 2560]
  • blk.9
  • blk.9.attn_gate.weight
    Q8_0
    [2560, 6144]
  • blk.9.attn_qkv.weight
    Q8_0
    [2560, 10240]
  • blk.9.ffn_down_exps.weight
    Q5_1
    [640, 2560, 512]
  • blk.9.ffn_down_shexp.weight
    Q8_0
    [640, 2560]
  • blk.9.ffn_gate_exps.weight
    Q4_K
    [2560, 640, 512]
  • blk.9.ffn_gate_inp.weight
    F32
    [2560, 512]
  • blk.9.ffn_gate_inp_shexp.weight
    F32
    [2560]
  • blk.9.ffn_gate_shexp.weight
    Q8_0
    [2560, 640]
  • blk.9.ffn_up_exps.weight
    Q4_K
    [2560, 640, 512]
  • blk.9.ffn_up_shexp.weight
    Q8_0
    [2560, 640]
  • blk.9.hc_attn_down.weight
    Q8_0
    [10240, 320]
  • blk.9.hc_attn_inject.weight
    F32
    [10240, 4]
  • blk.9.hc_attn_norm.weight
    F32
    [10240]
  • blk.9.hc_attn_up.weight
    Q8_0
    [320, 10240]
  • blk.9.hc_ffn_down.weight
    Q8_0
    [10240, 320]
  • blk.9.hc_ffn_inject.weight
    F32
    [10240, 4]
  • blk.9.hc_ffn_norm.weight
    F32
    [10240]
  • blk.9.hc_ffn_up.weight
    Q8_0
    [320, 10240]
  • blk.9.ssm_a
    F32
    [48]
  • blk.9.ssm_alpha.weight
    F32
    [2560, 48]
  • blk.9.ssm_beta.weight
    F32
    [2560, 48]
  • blk.9.ssm_conv1d.weight
    F32
    [4, 10240]
  • blk.9.ssm_dt.bias
    F32
    [48]
  • blk.9.ssm_norm.weight
    F32
    [128]
  • blk.9.ssm_out.weight
    Q8_0
    [6144, 2560]
  • blk.10
  • blk.10.attn_gate.weight
    Q8_0
    [2560, 6144]
  • blk.10.attn_qkv.weight
    Q8_0
    [2560, 10240]
  • blk.10.ffn_down_exps.weight
    Q5_1
    [640, 2560, 512]
  • blk.10.ffn_down_shexp.weight
    Q8_0
    [640, 2560]
  • blk.10.ffn_gate_exps.weight
    Q4_K
    [2560, 640, 512]
  • blk.10.ffn_gate_inp.weight
    F32
    [2560, 512]
  • blk.10.ffn_gate_inp_shexp.weight
    F32
    [2560]
  • blk.10.ffn_gate_shexp.weight
    Q8_0
    [2560, 640]
  • blk.10.ffn_up_exps.weight
    Q4_K
    [2560, 640, 512]
  • blk.10.ffn_up_shexp.weight
    Q8_0
    [2560, 640]
  • blk.10.hc_attn_down.weight
    Q8_0
    [10240, 320]
  • blk.10.hc_attn_inject.weight
    F32
    [10240, 4]
  • blk.10.hc_attn_norm.weight
    F32
    [10240]
  • blk.10.hc_attn_up.weight
    Q8_0
    [320, 10240]
  • blk.10.hc_ffn_down.weight
    Q8_0
    [10240, 320]
  • blk.10.hc_ffn_inject.weight
    F32
    [10240, 4]
  • blk.10.hc_ffn_norm.weight
    F32
    [10240]
  • blk.10.hc_ffn_up.weight
    Q8_0
    [320, 10240]
  • blk.10.ssm_a
    F32
    [48]
  • blk.10.ssm_alpha.weight
    F32
    [2560, 48]
  • blk.10.ssm_beta.weight
    F32
    [2560, 48]
  • blk.10.ssm_conv1d.weight
    F32
    [4, 10240]
  • blk.10.ssm_dt.bias
    F32
    [48]
  • blk.10.ssm_norm.weight
    F32
    [128]
  • blk.10.ssm_out.weight
    Q8_0
    [6144, 2560]
  • blk.11
  • blk.11.attn_k.weight
    Q8_0
    [2560, 512]
  • blk.11.attn_k_norm.weight
    F32
    [256]
  • blk.11.attn_output.weight
    Q8_0
    [6144, 2560]
  • blk.11.attn_q.weight
    Q8_0
    [2560, 12288]
  • blk.11.attn_q_norm.weight
    F32
    [256]
  • blk.11.attn_v.weight
    Q8_0
    [2560, 512]
  • blk.11.ffn_down_exps.weight
    Q5_1
    [640, 2560, 512]
  • blk.11.ffn_down_shexp.weight
    Q8_0
    [640, 2560]
  • output.weight
    Q8_0
    [2560, 248320]
  • output_hc_down.weight
    Q8_0
    [10240, 320]
  • output_hc_norm.weight
    F32
    [10240]
  • output_hc_up.weight
    Q8_0
    [320, 10240]
  • per_layer_token_embd.weight
    IQ4_NL
    [160, 320001536]