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.

892b9e7a8201 · 46GB
    Metadata
  • split.count
    5
  • split.no
    3
  • split.tensors.count
    1025
  • Tensor
    blk.38
  • blk.38.ffn_down_exps.weight
    Q2_K
    [2048, 7168, 256]
  • blk.38.ffn_down_shexp.weight
    Q6_K
    [2048, 7168]
  • blk.38.ffn_gate_exps.weight
    Q2_K
    [7168, 2048, 256]
  • blk.38.ffn_gate_inp.weight
    F32
    [7168, 256]
  • blk.38.ffn_gate_shexp.weight
    Q4_K
    [7168, 2048]
  • blk.38.ffn_norm.weight
    F32
    [7168]
  • blk.38.ffn_up_exps.weight
    Q2_K
    [7168, 2048, 256]
  • blk.38.ffn_up_shexp.weight
    Q4_K
    [7168, 2048]
  • blk.39
  • blk.39.attn_kv_a_mqa.weight
    Q6_K
    [7168, 576]
  • blk.39.attn_kv_a_norm.weight
    F32
    [512]
  • blk.39.attn_kv_b.weight
    Q6_K
    [512, 32768]
  • blk.39.attn_norm.weight
    F32
    [7168]
  • blk.39.attn_output.weight
    Q4_K
    [16384, 7168]
  • blk.39.attn_q_a.weight
    Q4_K
    [7168, 1536]
  • blk.39.attn_q_a_norm.weight
    F32
    [1536]
  • blk.39.attn_q_b.weight
    Q4_K
    [1536, 24576]
  • blk.39.exp_probs_b.bias
    F32
    [256]
  • blk.39.ffn_down_exps.weight
    Q2_K
    [2048, 7168, 256]
  • blk.39.ffn_down_shexp.weight
    Q6_K
    [2048, 7168]
  • blk.39.ffn_gate_exps.weight
    Q2_K
    [7168, 2048, 256]
  • blk.39.ffn_gate_inp.weight
    F32
    [7168, 256]
  • blk.39.ffn_gate_shexp.weight
    Q4_K
    [7168, 2048]
  • blk.39.ffn_norm.weight
    F32
    [7168]
  • blk.39.ffn_up_exps.weight
    Q2_K
    [7168, 2048, 256]
  • blk.39.ffn_up_shexp.weight
    Q4_K
    [7168, 2048]
  • blk.40
  • blk.40.attn_kv_a_mqa.weight
    Q6_K
    [7168, 576]
  • blk.40.attn_kv_a_norm.weight
    F32
    [512]
  • blk.40.attn_kv_b.weight
    Q6_K
    [512, 32768]
  • blk.40.attn_norm.weight
    F32
    [7168]
  • blk.40.attn_output.weight
    Q4_K
    [16384, 7168]
  • blk.40.attn_q_a.weight
    Q4_K
    [7168, 1536]
  • blk.40.attn_q_a_norm.weight
    F32
    [1536]
  • blk.40.attn_q_b.weight
    Q4_K
    [1536, 24576]
  • blk.40.exp_probs_b.bias
    F32
    [256]
  • blk.40.ffn_down_exps.weight
    Q2_K
    [2048, 7168, 256]
  • blk.40.ffn_down_shexp.weight
    Q6_K
    [2048, 7168]
  • blk.40.ffn_gate_exps.weight
    Q2_K
    [7168, 2048, 256]
  • blk.40.ffn_gate_inp.weight
    F32
    [7168, 256]
  • blk.40.ffn_gate_shexp.weight
    Q4_K
    [7168, 2048]
  • blk.40.ffn_norm.weight
    F32
    [7168]
  • blk.40.ffn_up_exps.weight
    Q2_K
    [7168, 2048, 256]
  • blk.40.ffn_up_shexp.weight
    Q4_K
    [7168, 2048]
  • blk.41
  • blk.41.attn_kv_a_mqa.weight
    Q6_K
    [7168, 576]
  • blk.41.attn_kv_a_norm.weight
    F32
    [512]
  • blk.41.attn_kv_b.weight
    Q6_K
    [512, 32768]
  • blk.41.attn_norm.weight
    F32
    [7168]
  • blk.41.attn_output.weight
    Q4_K
    [16384, 7168]
  • blk.41.attn_q_a.weight
    Q4_K
    [7168, 1536]
  • blk.41.attn_q_a_norm.weight
    F32
    [1536]
  • blk.41.attn_q_b.weight
    Q4_K
    [1536, 24576]
  • blk.41.exp_probs_b.bias
    F32
    [256]
  • blk.41.ffn_down_exps.weight
    Q2_K
    [2048, 7168, 256]
  • blk.41.ffn_down_shexp.weight
    Q6_K
    [2048, 7168]
  • blk.41.ffn_gate_exps.weight
    Q2_K
    [7168, 2048, 256]
  • blk.41.ffn_gate_inp.weight
    F32
    [7168, 256]
  • blk.41.ffn_gate_shexp.weight
    Q4_K
    [7168, 2048]
  • blk.41.ffn_norm.weight
    F32
    [7168]
  • blk.41.ffn_up_exps.weight
    Q2_K
    [7168, 2048, 256]
  • blk.41.ffn_up_shexp.weight
    Q4_K
    [7168, 2048]
  • blk.42
  • blk.42.attn_kv_a_mqa.weight
    Q6_K
    [7168, 576]
  • blk.42.attn_kv_a_norm.weight
    F32
    [512]
  • blk.42.attn_kv_b.weight
    Q6_K
    [512, 32768]
  • blk.42.attn_norm.weight
    F32
    [7168]
  • blk.42.attn_output.weight
    Q4_K
    [16384, 7168]
  • blk.42.attn_q_a.weight
    Q4_K
    [7168, 1536]
  • blk.42.attn_q_a_norm.weight
    F32
    [1536]
  • blk.42.attn_q_b.weight
    Q4_K
    [1536, 24576]
  • blk.42.exp_probs_b.bias
    F32
    [256]
  • blk.42.ffn_down_exps.weight
    Q2_K
    [2048, 7168, 256]
  • blk.42.ffn_down_shexp.weight
    Q6_K
    [2048, 7168]
  • blk.42.ffn_gate_exps.weight
    Q2_K
    [7168, 2048, 256]
  • blk.42.ffn_gate_inp.weight
    F32
    [7168, 256]
  • blk.42.ffn_gate_shexp.weight
    Q4_K
    [7168, 2048]
  • blk.42.ffn_norm.weight
    F32
    [7168]
  • blk.42.ffn_up_exps.weight
    Q2_K
    [7168, 2048, 256]
  • blk.42.ffn_up_shexp.weight
    Q4_K
    [7168, 2048]
  • blk.43
  • blk.43.attn_kv_a_mqa.weight
    Q6_K
    [7168, 576]
  • blk.43.attn_kv_a_norm.weight
    F32
    [512]
  • blk.43.attn_kv_b.weight
    Q6_K
    [512, 32768]
  • blk.43.attn_norm.weight
    F32
    [7168]
  • blk.43.attn_output.weight
    Q4_K
    [16384, 7168]
  • blk.43.attn_q_a.weight
    Q4_K
    [7168, 1536]
  • blk.43.attn_q_a_norm.weight
    F32
    [1536]
  • blk.43.attn_q_b.weight
    Q4_K
    [1536, 24576]
  • blk.43.exp_probs_b.bias
    F32
    [256]
  • blk.43.ffn_down_exps.weight
    Q2_K
    [2048, 7168, 256]
  • blk.43.ffn_down_shexp.weight
    Q6_K
    [2048, 7168]
  • blk.43.ffn_gate_exps.weight
    Q2_K
    [7168, 2048, 256]
  • blk.43.ffn_gate_inp.weight
    F32
    [7168, 256]
  • blk.43.ffn_gate_shexp.weight
    Q4_K
    [7168, 2048]
  • blk.43.ffn_norm.weight
    F32
    [7168]
  • blk.43.ffn_up_exps.weight
    Q2_K
    [7168, 2048, 256]
  • blk.43.ffn_up_shexp.weight
    Q4_K
    [7168, 2048]
  • blk.44
  • blk.44.attn_kv_a_mqa.weight
    Q6_K
    [7168, 576]
  • blk.44.attn_kv_a_norm.weight
    F32
    [512]
  • blk.44.attn_kv_b.weight
    Q6_K
    [512, 32768]
  • blk.44.attn_norm.weight
    F32
    [7168]
  • blk.44.attn_output.weight
    Q4_K
    [16384, 7168]
  • blk.44.attn_q_a.weight
    Q4_K
    [7168, 1536]
  • blk.44.attn_q_a_norm.weight
    F32
    [1536]
  • blk.44.attn_q_b.weight
    Q4_K
    [1536, 24576]
  • blk.44.exp_probs_b.bias
    F32
    [256]
  • blk.44.ffn_down_exps.weight
    Q2_K
    [2048, 7168, 256]
  • blk.44.ffn_down_shexp.weight
    Q6_K
    [2048, 7168]
  • blk.44.ffn_gate_exps.weight
    Q2_K
    [7168, 2048, 256]
  • blk.44.ffn_gate_inp.weight
    F32
    [7168, 256]
  • blk.44.ffn_gate_shexp.weight
    Q4_K
    [7168, 2048]
  • blk.44.ffn_norm.weight
    F32
    [7168]
  • blk.44.ffn_up_exps.weight
    Q2_K
    [7168, 2048, 256]
  • blk.44.ffn_up_shexp.weight
    Q4_K
    [7168, 2048]
  • blk.45
  • blk.45.attn_kv_a_mqa.weight
    Q6_K
    [7168, 576]
  • blk.45.attn_kv_a_norm.weight
    F32
    [512]
  • blk.45.attn_kv_b.weight
    Q6_K
    [512, 32768]
  • blk.45.attn_norm.weight
    F32
    [7168]
  • blk.45.attn_output.weight
    Q4_K
    [16384, 7168]
  • blk.45.attn_q_a.weight
    Q4_K
    [7168, 1536]
  • blk.45.attn_q_a_norm.weight
    F32
    [1536]
  • blk.45.attn_q_b.weight
    Q4_K
    [1536, 24576]
  • blk.45.exp_probs_b.bias
    F32
    [256]
  • blk.45.ffn_down_exps.weight
    Q2_K
    [2048, 7168, 256]
  • blk.45.ffn_down_shexp.weight
    Q6_K
    [2048, 7168]
  • blk.45.ffn_gate_exps.weight
    Q2_K
    [7168, 2048, 256]
  • blk.45.ffn_gate_inp.weight
    F32
    [7168, 256]
  • blk.45.ffn_gate_shexp.weight
    Q4_K
    [7168, 2048]
  • blk.45.ffn_norm.weight
    F32
    [7168]
  • blk.45.ffn_up_exps.weight
    Q2_K
    [7168, 2048, 256]
  • blk.45.ffn_up_shexp.weight
    Q4_K
    [7168, 2048]
  • blk.46
  • blk.46.attn_kv_a_mqa.weight
    Q6_K
    [7168, 576]
  • blk.46.attn_kv_a_norm.weight
    F32
    [512]
  • blk.46.attn_kv_b.weight
    Q6_K
    [512, 32768]
  • blk.46.attn_norm.weight
    F32
    [7168]
  • blk.46.attn_output.weight
    Q4_K
    [16384, 7168]
  • blk.46.attn_q_a.weight
    Q4_K
    [7168, 1536]
  • blk.46.attn_q_a_norm.weight
    F32
    [1536]
  • blk.46.attn_q_b.weight
    Q4_K
    [1536, 24576]
  • blk.46.exp_probs_b.bias
    F32
    [256]
  • blk.46.ffn_down_exps.weight
    Q2_K
    [2048, 7168, 256]
  • blk.46.ffn_down_shexp.weight
    Q6_K
    [2048, 7168]
  • blk.46.ffn_gate_exps.weight
    Q2_K
    [7168, 2048, 256]
  • blk.46.ffn_gate_inp.weight
    F32
    [7168, 256]
  • blk.46.ffn_gate_shexp.weight
    Q4_K
    [7168, 2048]
  • blk.46.ffn_norm.weight
    F32
    [7168]
  • blk.46.ffn_up_exps.weight
    Q2_K
    [7168, 2048, 256]
  • blk.46.ffn_up_shexp.weight
    Q4_K
    [7168, 2048]
  • blk.47
  • blk.47.attn_kv_a_mqa.weight
    Q6_K
    [7168, 576]
  • blk.47.attn_kv_a_norm.weight
    F32
    [512]
  • blk.47.attn_kv_b.weight
    Q6_K
    [512, 32768]
  • blk.47.attn_norm.weight
    F32
    [7168]
  • blk.47.attn_output.weight
    Q4_K
    [16384, 7168]
  • blk.47.attn_q_a.weight
    Q4_K
    [7168, 1536]
  • blk.47.attn_q_a_norm.weight
    F32
    [1536]
  • blk.47.attn_q_b.weight
    Q4_K
    [1536, 24576]
  • blk.47.exp_probs_b.bias
    F32
    [256]
  • blk.47.ffn_down_exps.weight
    Q2_K
    [2048, 7168, 256]
  • blk.47.ffn_down_shexp.weight
    Q6_K
    [2048, 7168]
  • blk.47.ffn_gate_exps.weight
    Q2_K
    [7168, 2048, 256]
  • blk.47.ffn_gate_inp.weight
    F32
    [7168, 256]
  • blk.47.ffn_gate_shexp.weight
    Q4_K
    [7168, 2048]
  • blk.47.ffn_norm.weight
    F32
    [7168]
  • blk.47.ffn_up_exps.weight
    Q2_K
    [7168, 2048, 256]
  • blk.47.ffn_up_shexp.weight
    Q4_K
    [7168, 2048]
  • blk.48
  • blk.48.attn_kv_a_mqa.weight
    Q6_K
    [7168, 576]
  • blk.48.attn_kv_a_norm.weight
    F32
    [512]
  • blk.48.attn_kv_b.weight
    Q6_K
    [512, 32768]
  • blk.48.attn_norm.weight
    F32
    [7168]
  • blk.48.attn_output.weight
    Q4_K
    [16384, 7168]
  • blk.48.attn_q_a.weight
    Q4_K
    [7168, 1536]
  • blk.48.attn_q_a_norm.weight
    F32
    [1536]
  • blk.48.attn_q_b.weight
    Q4_K
    [1536, 24576]
  • blk.48.exp_probs_b.bias
    F32
    [256]
  • blk.48.ffn_down_exps.weight
    Q2_K
    [2048, 7168, 256]
  • blk.48.ffn_down_shexp.weight
    Q6_K
    [2048, 7168]
  • blk.48.ffn_gate_exps.weight
    Q2_K
    [7168, 2048, 256]
  • blk.48.ffn_gate_inp.weight
    F32
    [7168, 256]
  • blk.48.ffn_gate_shexp.weight
    Q4_K
    [7168, 2048]
  • blk.48.ffn_norm.weight
    F32
    [7168]
  • blk.48.ffn_up_exps.weight
    Q2_K
    [7168, 2048, 256]
  • blk.48.ffn_up_shexp.weight
    Q4_K
    [7168, 2048]
  • blk.49
  • blk.49.attn_kv_a_mqa.weight
    Q6_K
    [7168, 576]
  • blk.49.attn_kv_a_norm.weight
    F32
    [512]
  • blk.49.attn_kv_b.weight
    Q6_K
    [512, 32768]
  • blk.49.attn_norm.weight
    F32
    [7168]
  • blk.49.attn_output.weight
    Q4_K
    [16384, 7168]
  • blk.49.attn_q_a.weight
    Q4_K
    [7168, 1536]
  • blk.49.attn_q_a_norm.weight
    F32
    [1536]
  • blk.49.attn_q_b.weight
    Q4_K
    [1536, 24576]
  • blk.49.exp_probs_b.bias
    F32
    [256]
  • blk.49.ffn_down_exps.weight
    Q2_K
    [2048, 7168, 256]
  • blk.49.ffn_down_shexp.weight
    Q6_K
    [2048, 7168]
  • blk.49.ffn_gate_exps.weight
    Q2_K
    [7168, 2048, 256]
  • blk.49.ffn_gate_inp.weight
    F32
    [7168, 256]
  • blk.49.ffn_gate_shexp.weight
    Q4_K
    [7168, 2048]
  • blk.49.ffn_norm.weight
    F32
    [7168]
  • blk.49.ffn_up_exps.weight
    Q2_K
    [7168, 2048, 256]
  • blk.49.ffn_up_shexp.weight
    Q4_K
    [7168, 2048]
  • blk.50
  • blk.50.attn_kv_a_mqa.weight
    Q6_K
    [7168, 576]
  • blk.50.attn_kv_a_norm.weight
    F32
    [512]
  • blk.50.attn_kv_b.weight
    Q6_K
    [512, 32768]
  • blk.50.attn_norm.weight
    F32
    [7168]
  • blk.50.attn_output.weight
    Q4_K
    [16384, 7168]
  • blk.50.attn_q_a.weight
    Q4_K
    [7168, 1536]
  • blk.50.attn_q_a_norm.weight
    F32
    [1536]
  • blk.50.attn_q_b.weight
    Q4_K
    [1536, 24576]
  • blk.50.exp_probs_b.bias
    F32
    [256]