742 6 months ago

DeepSeek's first-generation of reasoning models with comparable performance to OpenAI-o1, including six dense models distilled from DeepSeek-R1 based on Llama and Qwen.

dff704a88236 · 42GB
    Metadata
  • split.count
    5
  • split.no
    4
  • split.tensors.count
    1025
  • Tensor
    blk.50
  • blk.50.ffn_down_exps.weight
    Q2_K
    [2048, 7168, 256]
  • blk.50.ffn_down_shexp.weight
    Q6_K
    [2048, 7168]
  • blk.50.ffn_gate_exps.weight
    Q2_K
    [7168, 2048, 256]
  • blk.50.ffn_gate_inp.weight
    F32
    [7168, 256]
  • blk.50.ffn_gate_shexp.weight
    Q4_K
    [7168, 2048]
  • blk.50.ffn_norm.weight
    F32
    [7168]
  • blk.50.ffn_up_exps.weight
    Q2_K
    [7168, 2048, 256]
  • blk.50.ffn_up_shexp.weight
    Q4_K
    [7168, 2048]
  • blk.51
  • blk.51.attn_kv_a_mqa.weight
    Q6_K
    [7168, 576]
  • blk.51.attn_kv_a_norm.weight
    F32
    [512]
  • blk.51.attn_kv_b.weight
    Q6_K
    [512, 32768]
  • blk.51.attn_norm.weight
    F32
    [7168]
  • blk.51.attn_output.weight
    Q4_K
    [16384, 7168]
  • blk.51.attn_q_a.weight
    Q4_K
    [7168, 1536]
  • blk.51.attn_q_a_norm.weight
    F32
    [1536]
  • blk.51.attn_q_b.weight
    Q4_K
    [1536, 24576]
  • blk.51.exp_probs_b.bias
    F32
    [256]
  • blk.51.ffn_down_exps.weight
    Q2_K
    [2048, 7168, 256]
  • blk.51.ffn_down_shexp.weight
    Q6_K
    [2048, 7168]
  • blk.51.ffn_gate_exps.weight
    Q2_K
    [7168, 2048, 256]
  • blk.51.ffn_gate_inp.weight
    F32
    [7168, 256]
  • blk.51.ffn_gate_shexp.weight
    Q4_K
    [7168, 2048]
  • blk.51.ffn_norm.weight
    F32
    [7168]
  • blk.51.ffn_up_exps.weight
    Q2_K
    [7168, 2048, 256]
  • blk.51.ffn_up_shexp.weight
    Q4_K
    [7168, 2048]
  • blk.52
  • blk.52.attn_kv_a_mqa.weight
    Q6_K
    [7168, 576]
  • blk.52.attn_kv_a_norm.weight
    F32
    [512]
  • blk.52.attn_kv_b.weight
    Q6_K
    [512, 32768]
  • blk.52.attn_norm.weight
    F32
    [7168]
  • blk.52.attn_output.weight
    Q4_K
    [16384, 7168]
  • blk.52.attn_q_a.weight
    Q4_K
    [7168, 1536]
  • blk.52.attn_q_a_norm.weight
    F32
    [1536]
  • blk.52.attn_q_b.weight
    Q4_K
    [1536, 24576]
  • blk.52.exp_probs_b.bias
    F32
    [256]
  • blk.52.ffn_down_exps.weight
    Q2_K
    [2048, 7168, 256]
  • blk.52.ffn_down_shexp.weight
    Q6_K
    [2048, 7168]
  • blk.52.ffn_gate_exps.weight
    Q2_K
    [7168, 2048, 256]
  • blk.52.ffn_gate_inp.weight
    F32
    [7168, 256]
  • blk.52.ffn_gate_shexp.weight
    Q4_K
    [7168, 2048]
  • blk.52.ffn_norm.weight
    F32
    [7168]
  • blk.52.ffn_up_exps.weight
    Q2_K
    [7168, 2048, 256]
  • blk.52.ffn_up_shexp.weight
    Q4_K
    [7168, 2048]
  • blk.53
  • blk.53.attn_kv_a_mqa.weight
    Q6_K
    [7168, 576]
  • blk.53.attn_kv_a_norm.weight
    F32
    [512]
  • blk.53.attn_kv_b.weight
    Q6_K
    [512, 32768]
  • blk.53.attn_norm.weight
    F32
    [7168]
  • blk.53.attn_output.weight
    Q4_K
    [16384, 7168]
  • blk.53.attn_q_a.weight
    Q4_K
    [7168, 1536]
  • blk.53.attn_q_a_norm.weight
    F32
    [1536]
  • blk.53.attn_q_b.weight
    Q4_K
    [1536, 24576]
  • blk.53.exp_probs_b.bias
    F32
    [256]
  • blk.53.ffn_down_exps.weight
    Q2_K
    [2048, 7168, 256]
  • blk.53.ffn_down_shexp.weight
    Q6_K
    [2048, 7168]
  • blk.53.ffn_gate_exps.weight
    Q2_K
    [7168, 2048, 256]
  • blk.53.ffn_gate_inp.weight
    F32
    [7168, 256]
  • blk.53.ffn_gate_shexp.weight
    Q4_K
    [7168, 2048]
  • blk.53.ffn_norm.weight
    F32
    [7168]
  • blk.53.ffn_up_exps.weight
    Q2_K
    [7168, 2048, 256]
  • blk.53.ffn_up_shexp.weight
    Q4_K
    [7168, 2048]
  • blk.54
  • blk.54.attn_kv_a_mqa.weight
    Q6_K
    [7168, 576]
  • blk.54.attn_kv_a_norm.weight
    F32
    [512]
  • blk.54.attn_kv_b.weight
    Q6_K
    [512, 32768]
  • blk.54.attn_norm.weight
    F32
    [7168]
  • blk.54.attn_output.weight
    Q4_K
    [16384, 7168]
  • blk.54.attn_q_a.weight
    Q4_K
    [7168, 1536]
  • blk.54.attn_q_a_norm.weight
    F32
    [1536]
  • blk.54.attn_q_b.weight
    Q4_K
    [1536, 24576]
  • blk.54.exp_probs_b.bias
    F32
    [256]
  • blk.54.ffn_down_exps.weight
    Q2_K
    [2048, 7168, 256]
  • blk.54.ffn_down_shexp.weight
    Q6_K
    [2048, 7168]
  • blk.54.ffn_gate_exps.weight
    Q2_K
    [7168, 2048, 256]
  • blk.54.ffn_gate_inp.weight
    F32
    [7168, 256]
  • blk.54.ffn_gate_shexp.weight
    Q4_K
    [7168, 2048]
  • blk.54.ffn_norm.weight
    F32
    [7168]
  • blk.54.ffn_up_exps.weight
    Q2_K
    [7168, 2048, 256]
  • blk.54.ffn_up_shexp.weight
    Q4_K
    [7168, 2048]
  • blk.55
  • blk.55.attn_kv_a_mqa.weight
    Q6_K
    [7168, 576]
  • blk.55.attn_kv_a_norm.weight
    F32
    [512]
  • blk.55.attn_kv_b.weight
    Q6_K
    [512, 32768]
  • blk.55.attn_norm.weight
    F32
    [7168]
  • blk.55.attn_output.weight
    Q4_K
    [16384, 7168]
  • blk.55.attn_q_a.weight
    Q4_K
    [7168, 1536]
  • blk.55.attn_q_a_norm.weight
    F32
    [1536]
  • blk.55.attn_q_b.weight
    Q4_K
    [1536, 24576]
  • blk.55.exp_probs_b.bias
    F32
    [256]
  • blk.55.ffn_down_exps.weight
    Q2_K
    [2048, 7168, 256]
  • blk.55.ffn_down_shexp.weight
    Q6_K
    [2048, 7168]
  • blk.55.ffn_gate_exps.weight
    Q2_K
    [7168, 2048, 256]
  • blk.55.ffn_gate_inp.weight
    F32
    [7168, 256]
  • blk.55.ffn_gate_shexp.weight
    Q4_K
    [7168, 2048]
  • blk.55.ffn_norm.weight
    F32
    [7168]
  • blk.55.ffn_up_exps.weight
    Q2_K
    [7168, 2048, 256]
  • blk.55.ffn_up_shexp.weight
    Q4_K
    [7168, 2048]
  • blk.56
  • blk.56.attn_kv_a_mqa.weight
    Q6_K
    [7168, 576]
  • blk.56.attn_kv_a_norm.weight
    F32
    [512]
  • blk.56.attn_kv_b.weight
    Q6_K
    [512, 32768]
  • blk.56.attn_norm.weight
    F32
    [7168]
  • blk.56.attn_output.weight
    Q4_K
    [16384, 7168]
  • blk.56.attn_q_a.weight
    Q4_K
    [7168, 1536]
  • blk.56.attn_q_a_norm.weight
    F32
    [1536]
  • blk.56.attn_q_b.weight
    Q4_K
    [1536, 24576]
  • blk.56.exp_probs_b.bias
    F32
    [256]
  • blk.56.ffn_down_exps.weight
    Q2_K
    [2048, 7168, 256]
  • blk.56.ffn_down_shexp.weight
    Q6_K
    [2048, 7168]
  • blk.56.ffn_gate_exps.weight
    Q2_K
    [7168, 2048, 256]
  • blk.56.ffn_gate_inp.weight
    F32
    [7168, 256]
  • blk.56.ffn_gate_shexp.weight
    Q4_K
    [7168, 2048]
  • blk.56.ffn_norm.weight
    F32
    [7168]
  • blk.56.ffn_up_exps.weight
    Q2_K
    [7168, 2048, 256]
  • blk.56.ffn_up_shexp.weight
    Q4_K
    [7168, 2048]
  • blk.57
  • blk.57.attn_kv_a_mqa.weight
    Q6_K
    [7168, 576]
  • blk.57.attn_kv_a_norm.weight
    F32
    [512]
  • blk.57.attn_kv_b.weight
    Q6_K
    [512, 32768]
  • blk.57.attn_norm.weight
    F32
    [7168]
  • blk.57.attn_output.weight
    Q4_K
    [16384, 7168]
  • blk.57.attn_q_a.weight
    Q4_K
    [7168, 1536]
  • blk.57.attn_q_a_norm.weight
    F32
    [1536]
  • blk.57.attn_q_b.weight
    Q4_K
    [1536, 24576]
  • blk.57.exp_probs_b.bias
    F32
    [256]
  • blk.57.ffn_down_exps.weight
    Q2_K
    [2048, 7168, 256]
  • blk.57.ffn_down_shexp.weight
    Q6_K
    [2048, 7168]
  • blk.57.ffn_gate_exps.weight
    Q2_K
    [7168, 2048, 256]
  • blk.57.ffn_gate_inp.weight
    F32
    [7168, 256]
  • blk.57.ffn_gate_shexp.weight
    Q4_K
    [7168, 2048]
  • blk.57.ffn_norm.weight
    F32
    [7168]
  • blk.57.ffn_up_exps.weight
    Q2_K
    [7168, 2048, 256]
  • blk.57.ffn_up_shexp.weight
    Q4_K
    [7168, 2048]
  • blk.58
  • blk.58.attn_kv_a_mqa.weight
    Q6_K
    [7168, 576]
  • blk.58.attn_kv_a_norm.weight
    F32
    [512]
  • blk.58.attn_kv_b.weight
    Q6_K
    [512, 32768]
  • blk.58.attn_norm.weight
    F32
    [7168]
  • blk.58.attn_output.weight
    Q4_K
    [16384, 7168]
  • blk.58.attn_q_a.weight
    Q4_K
    [7168, 1536]
  • blk.58.attn_q_a_norm.weight
    F32
    [1536]
  • blk.58.attn_q_b.weight
    Q4_K
    [1536, 24576]
  • blk.58.exp_probs_b.bias
    F32
    [256]
  • blk.58.ffn_down_exps.weight
    Q2_K
    [2048, 7168, 256]
  • blk.58.ffn_down_shexp.weight
    Q6_K
    [2048, 7168]
  • blk.58.ffn_gate_exps.weight
    Q2_K
    [7168, 2048, 256]
  • blk.58.ffn_gate_inp.weight
    F32
    [7168, 256]
  • blk.58.ffn_gate_shexp.weight
    Q4_K
    [7168, 2048]
  • blk.58.ffn_norm.weight
    F32
    [7168]
  • blk.58.ffn_up_exps.weight
    Q2_K
    [7168, 2048, 256]
  • blk.58.ffn_up_shexp.weight
    Q4_K
    [7168, 2048]
  • blk.59
  • blk.59.attn_kv_a_mqa.weight
    Q6_K
    [7168, 576]
  • blk.59.attn_kv_a_norm.weight
    F32
    [512]
  • blk.59.attn_kv_b.weight
    Q6_K
    [512, 32768]
  • blk.59.attn_norm.weight
    F32
    [7168]
  • blk.59.attn_output.weight
    Q4_K
    [16384, 7168]
  • blk.59.attn_q_a.weight
    Q4_K
    [7168, 1536]
  • blk.59.attn_q_a_norm.weight
    F32
    [1536]
  • blk.59.attn_q_b.weight
    Q4_K
    [1536, 24576]
  • blk.59.exp_probs_b.bias
    F32
    [256]
  • blk.59.ffn_down_exps.weight
    Q2_K
    [2048, 7168, 256]
  • blk.59.ffn_down_shexp.weight
    Q6_K
    [2048, 7168]
  • blk.59.ffn_gate_exps.weight
    Q2_K
    [7168, 2048, 256]
  • blk.59.ffn_gate_inp.weight
    F32
    [7168, 256]
  • blk.59.ffn_gate_shexp.weight
    Q4_K
    [7168, 2048]
  • blk.59.ffn_norm.weight
    F32
    [7168]
  • blk.59.ffn_up_exps.weight
    Q2_K
    [7168, 2048, 256]
  • blk.59.ffn_up_shexp.weight
    Q4_K
    [7168, 2048]
  • blk.60
  • blk.60.attn_kv_a_mqa.weight
    Q6_K
    [7168, 576]
  • blk.60.attn_kv_a_norm.weight
    F32
    [512]
  • blk.60.attn_kv_b.weight
    Q6_K
    [512, 32768]
  • blk.60.attn_norm.weight
    F32
    [7168]
  • blk.60.attn_output.weight
    Q4_K
    [16384, 7168]
  • blk.60.attn_q_a.weight
    Q4_K
    [7168, 1536]
  • blk.60.attn_q_a_norm.weight
    F32
    [1536]
  • blk.60.attn_q_b.weight
    Q4_K
    [1536, 24576]
  • blk.60.exp_probs_b.bias
    F32
    [256]
  • blk.60.ffn_down_exps.weight
    Q2_K
    [2048, 7168, 256]
  • blk.60.ffn_down_shexp.weight
    Q6_K
    [2048, 7168]
  • blk.60.ffn_gate_exps.weight
    Q2_K
    [7168, 2048, 256]
  • blk.60.ffn_gate_inp.weight
    F32
    [7168, 256]
  • blk.60.ffn_gate_shexp.weight
    Q4_K
    [7168, 2048]
  • blk.60.ffn_norm.weight
    F32
    [7168]
  • blk.60.ffn_up_exps.weight
    Q2_K
    [7168, 2048, 256]
  • blk.60.ffn_up_shexp.weight
    Q4_K
    [7168, 2048]