clore/
deepseek-v3.1:latest

11 1 week ago

... /
model
c26f6745f265 · 380GB
    Metadata
  • general.architecture
    deepseek2
  • general.file_type
    Q4_0
  • deepseek2.attention.head_count
    128
  • deepseek2.attention.head_count_kv
    1
  • deepseek2.attention.key_length
    576
  • deepseek2.attention.key_length_mla
    192
  • deepseek2.attention.kv_lora_rank
    512
  • deepseek2.attention.layer_norm_rms_epsilon
    1e-06
  • deepseek2.attention.q_lora_rank
    1536
  • deepseek2.attention.value_length
    512
  • deepseek2.attention.value_length_mla
    128
  • deepseek2.block_count
    61
  • deepseek2.context_length
    163840
  • deepseek2.embedding_length
    7168
  • deepseek2.expert_count
    256
  • deepseek2.expert_feed_forward_length
    2048
  • deepseek2.expert_gating_func
    2
  • deepseek2.expert_shared_count
    1
  • deepseek2.expert_used_count
    8
  • deepseek2.expert_weights_norm
    true
  • deepseek2.expert_weights_scale
    2.5
  • deepseek2.feed_forward_length
    18432
  • deepseek2.leading_dense_block_count
    3
  • deepseek2.rope.dimension_count
    64
  • deepseek2.rope.freq_base
    10000
  • deepseek2.rope.scaling.factor
    40
  • deepseek2.rope.scaling.original_context_length
    4096
  • deepseek2.rope.scaling.type
    yarn
  • deepseek2.rope.scaling.yarn_log_multiplier
    0.1
  • deepseek2.vocab_size
    129280
  • tokenizer.ggml.add_bos_token
    true
  • tokenizer.ggml.add_eos_token
    false
  • tokenizer.ggml.add_sep_token
    false
  • tokenizer.ggml.bos_token_id
    0
  • tokenizer.ggml.eos_token_id
    1
  • tokenizer.ggml.merges
    [Ġ t, Ġ a, i n, Ġ Ġ, h e, ...]
  • tokenizer.ggml.model
    gpt2
  • tokenizer.ggml.padding_token_id
    2
  • tokenizer.ggml.pre
    deepseek-v3
  • tokenizer.ggml.token_type
    [3, 3, 3, 1, 1, ...]
  • tokenizer.ggml.tokens
    [<｜begin▁of▁sentence｜>, <｜end▁of▁sentence｜>, <｜▁pad▁｜>, !, ", ...]
  • quantize.imatrix.chunks_count
    84
  • quantize.imatrix.dataset
    unsloth_calibration_DeepSeek-V3.1-BF16.txt
  • quantize.imatrix.entries_count
    781
  • quantize.imatrix.file
    DeepSeek-V3.1-GGUF/imatrix_unsloth.gguf
  • split.count
    8
  • split.no
    0
  • split.tensors.count
    1086
    • Tensor
  • token_embd.weight
    Q4_0
    [7168, 129280]
    • blk.0
  • blk.0.attn_k_b.weight
    Q8_0
    [128, 512, 128]
  • blk.0.attn_kv_a_mqa.weight
    Q4_0
    [7168, 576]
  • blk.0.attn_kv_a_norm.weight
    F32
    [512]
  • blk.0.attn_norm.weight
    F32
    [7168]
  • blk.0.attn_output.weight
    Q4_0
    [16384, 7168]
  • blk.0.attn_q_a.weight
    Q4_0
    [7168, 1536]
  • blk.0.attn_q_a_norm.weight
    F32
    [1536]
  • blk.0.attn_q_b.weight
    Q4_0
    [1536, 24576]
  • blk.0.attn_v_b.weight
    Q8_0
    [512, 128, 128]
  • blk.0.ffn_down.weight
    Q4_1
    [18432, 7168]
  • blk.0.ffn_gate.weight
    Q4_0
    [7168, 18432]
  • blk.0.ffn_norm.weight
    F32
    [7168]
  • blk.0.ffn_up.weight
    Q4_0
    [7168, 18432]
    • blk.1
  • blk.1.attn_k_b.weight
    Q8_0
    [128, 512, 128]
  • blk.1.attn_kv_a_mqa.weight
    Q4_0
    [7168, 576]
  • blk.1.attn_kv_a_norm.weight
    F32
    [512]
  • blk.1.attn_norm.weight
    F32
    [7168]
  • blk.1.attn_output.weight
    Q4_0
    [16384, 7168]
  • blk.1.attn_q_a.weight
    Q4_0
    [7168, 1536]
  • blk.1.attn_q_a_norm.weight
    F32
    [1536]
  • blk.1.attn_q_b.weight
    Q4_0
    [1536, 24576]
  • blk.1.attn_v_b.weight
    Q8_0
    [512, 128, 128]
  • blk.1.ffn_down.weight
    Q4_1
    [18432, 7168]
  • blk.1.ffn_gate.weight
    Q4_0
    [7168, 18432]
  • blk.1.ffn_norm.weight
    F32
    [7168]
  • blk.1.ffn_up.weight
    Q4_0
    [7168, 18432]
    • blk.2
  • blk.2.attn_k_b.weight
    Q8_0
    [128, 512, 128]
  • blk.2.attn_kv_a_mqa.weight
    Q4_0
    [7168, 576]
  • blk.2.attn_kv_a_norm.weight
    F32
    [512]
  • blk.2.attn_norm.weight
    F32
    [7168]
  • blk.2.attn_output.weight
    Q4_0
    [16384, 7168]
  • blk.2.attn_q_a.weight
    Q4_0
    [7168, 1536]
  • blk.2.attn_q_a_norm.weight
    F32
    [1536]
  • blk.2.attn_q_b.weight
    Q4_0
    [1536, 24576]
  • blk.2.attn_v_b.weight
    Q8_0
    [512, 128, 128]
  • blk.2.ffn_down.weight
    Q4_1
    [18432, 7168]
  • blk.2.ffn_gate.weight
    Q4_0
    [7168, 18432]
  • blk.2.ffn_norm.weight
    F32
    [7168]
  • blk.2.ffn_up.weight
    Q4_0
    [7168, 18432]
    • blk.3
  • blk.3.attn_k_b.weight
    Q8_0
    [128, 512, 128]
  • blk.3.attn_kv_a_mqa.weight
    Q4_0
    [7168, 576]
  • blk.3.attn_kv_a_norm.weight
    F32
    [512]
  • blk.3.attn_norm.weight
    F32
    [7168]
  • blk.3.attn_output.weight
    Q4_0
    [16384, 7168]
  • blk.3.attn_q_a.weight
    Q4_0
    [7168, 1536]
  • blk.3.attn_q_a_norm.weight
    F32
    [1536]
  • blk.3.attn_q_b.weight
    Q4_0
    [1536, 24576]
  • blk.3.attn_v_b.weight
    Q8_0
    [512, 128, 128]
  • blk.3.exp_probs_b.bias
    F32
    [256]
  • blk.3.ffn_down_exps.weight
    Q4_1
    [2048, 7168, 256]
  • blk.3.ffn_down_shexp.weight
    Q4_1
    [2048, 7168]
  • blk.3.ffn_gate_exps.weight
    Q4_0
    [7168, 2048, 256]
  • blk.3.ffn_gate_inp.weight
    F32
    [7168, 256]
  • blk.3.ffn_gate_shexp.weight
    Q4_0
    [7168, 2048]
  • blk.3.ffn_norm.weight
    F32
    [7168]
  • blk.3.ffn_up_exps.weight
    Q4_0
    [7168, 2048, 256]
  • blk.3.ffn_up_shexp.weight
    Q4_0
    [7168, 2048]
    • blk.4
  • blk.4.attn_k_b.weight
    Q8_0
    [128, 512, 128]
  • blk.4.attn_kv_a_mqa.weight
    Q4_0
    [7168, 576]
  • blk.4.attn_kv_a_norm.weight
    F32
    [512]
  • blk.4.attn_norm.weight
    F32
    [7168]
  • blk.4.attn_output.weight
    Q4_0
    [16384, 7168]
  • blk.4.attn_q_a.weight
    Q4_0
    [7168, 1536]
  • blk.4.attn_q_a_norm.weight
    F32
    [1536]
  • blk.4.attn_q_b.weight
    Q4_0
    [1536, 24576]
  • blk.4.attn_v_b.weight
    Q8_0
    [512, 128, 128]
  • blk.4.exp_probs_b.bias
    F32
    [256]
  • blk.4.ffn_down_exps.weight
    Q4_1
    [2048, 7168, 256]
  • blk.4.ffn_down_shexp.weight
    Q4_1
    [2048, 7168]
  • blk.4.ffn_gate_exps.weight
    Q4_0
    [7168, 2048, 256]
  • blk.4.ffn_gate_inp.weight
    F32
    [7168, 256]
  • blk.4.ffn_gate_shexp.weight
    Q4_0
    [7168, 2048]
  • blk.4.ffn_norm.weight
    F32
    [7168]
  • blk.4.ffn_up_exps.weight
    Q4_0
    [7168, 2048, 256]
  • blk.4.ffn_up_shexp.weight
    Q4_0
    [7168, 2048]
    • blk.5
  • blk.5.attn_k_b.weight
    Q8_0
    [128, 512, 128]
  • blk.5.attn_kv_a_mqa.weight
    Q4_0
    [7168, 576]
  • blk.5.attn_kv_a_norm.weight
    F32
    [512]
  • blk.5.attn_norm.weight
    F32
    [7168]
  • blk.5.attn_output.weight
    Q4_0
    [16384, 7168]
  • blk.5.attn_q_a.weight
    Q4_0
    [7168, 1536]
  • blk.5.attn_q_a_norm.weight
    F32
    [1536]
  • blk.5.attn_q_b.weight
    Q4_0
    [1536, 24576]
  • blk.5.attn_v_b.weight
    Q8_0
    [512, 128, 128]
  • blk.5.exp_probs_b.bias
    F32
    [256]
  • blk.5.ffn_down_exps.weight
    Q4_1
    [2048, 7168, 256]
  • blk.5.ffn_down_shexp.weight
    Q4_1
    [2048, 7168]
  • blk.5.ffn_gate_exps.weight
    Q4_0
    [7168, 2048, 256]
  • blk.5.ffn_gate_inp.weight
    F32
    [7168, 256]
  • blk.5.ffn_gate_shexp.weight
    Q4_0
    [7168, 2048]
  • blk.5.ffn_norm.weight
    F32
    [7168]
  • blk.5.ffn_up_exps.weight
    Q4_0
    [7168, 2048, 256]
  • blk.5.ffn_up_shexp.weight
    Q4_0
    [7168, 2048]
    • blk.6
  • blk.6.attn_k_b.weight
    Q8_0
    [128, 512, 128]
  • blk.6.attn_kv_a_mqa.weight
    Q4_0
    [7168, 576]
  • blk.6.attn_kv_a_norm.weight
    F32
    [512]
  • blk.6.attn_norm.weight
    F32
    [7168]
  • blk.6.attn_output.weight
    Q4_0
    [16384, 7168]
  • blk.6.attn_q_a.weight
    Q4_0
    [7168, 1536]
  • blk.6.attn_q_a_norm.weight
    F32
    [1536]
  • blk.6.attn_q_b.weight
    Q4_0
    [1536, 24576]
  • blk.6.attn_v_b.weight
    Q8_0
    [512, 128, 128]
  • blk.6.exp_probs_b.bias
    F32
    [256]
  • blk.6.ffn_down_exps.weight
    Q4_1
    [2048, 7168, 256]
  • blk.6.ffn_down_shexp.weight
    Q4_1
    [2048, 7168]
  • blk.6.ffn_gate_exps.weight
    Q4_0
    [7168, 2048, 256]
  • blk.6.ffn_gate_inp.weight
    F32
    [7168, 256]
  • blk.6.ffn_gate_shexp.weight
    Q4_0
    [7168, 2048]
  • blk.6.ffn_norm.weight
    F32
    [7168]
  • blk.6.ffn_up_exps.weight
    Q4_0
    [7168, 2048, 256]
  • blk.6.ffn_up_shexp.weight
    Q4_0
    [7168, 2048]
    • blk.7
  • blk.7.attn_k_b.weight
    Q8_0
    [128, 512, 128]
  • blk.7.attn_kv_a_mqa.weight
    Q4_0
    [7168, 576]
  • blk.7.attn_kv_a_norm.weight
    F32
    [512]
  • blk.7.attn_norm.weight
    F32
    [7168]
  • blk.7.attn_output.weight
    Q4_0
    [16384, 7168]
  • blk.7.attn_q_a.weight
    Q4_0
    [7168, 1536]
  • blk.7.attn_q_a_norm.weight
    F32
    [1536]
  • blk.7.attn_q_b.weight
    Q4_0
    [1536, 24576]
  • blk.7.attn_v_b.weight
    Q8_0
    [512, 128, 128]
  • blk.7.exp_probs_b.bias
    F32
    [256]
  • blk.7.ffn_down_exps.weight
    Q4_0
    [2048, 7168, 256]
  • blk.7.ffn_down_shexp.weight
    Q4_0
    [2048, 7168]
  • blk.7.ffn_gate_exps.weight
    Q4_0
    [7168, 2048, 256]
  • blk.7.ffn_gate_inp.weight
    F32
    [7168, 256]
  • blk.7.ffn_gate_shexp.weight
    Q4_0
    [7168, 2048]
  • blk.7.ffn_norm.weight
    F32
    [7168]
  • blk.7.ffn_up_exps.weight
    Q4_0
    [7168, 2048, 256]
  • blk.7.ffn_up_shexp.weight
    Q4_0
    [7168, 2048]
    • blk.8
  • blk.8.attn_k_b.weight
    Q8_0
    [128, 512, 128]
  • blk.8.attn_kv_a_mqa.weight
    Q4_0
    [7168, 576]
  • blk.8.attn_kv_a_norm.weight
    F32
    [512]
  • blk.8.attn_norm.weight
    F32
    [7168]
  • blk.8.attn_output.weight
    Q4_0
    [16384, 7168]
  • blk.8.attn_q_a.weight
    Q4_0
    [7168, 1536]
  • blk.8.attn_q_a_norm.weight
    F32
    [1536]
  • blk.8.attn_q_b.weight
    Q4_0
    [1536, 24576]
  • blk.8.attn_v_b.weight
    Q8_0
    [512, 128, 128]
  • blk.8.exp_probs_b.bias
    F32
    [256]
  • blk.8.ffn_down_exps.weight
    Q4_0
    [2048, 7168, 256]
  • blk.8.ffn_down_shexp.weight
    Q4_0
    [2048, 7168]
  • blk.8.ffn_gate_exps.weight
    Q4_0
    [7168, 2048, 256]
  • blk.8.ffn_gate_inp.weight
    F32
    [7168, 256]
  • blk.8.ffn_gate_shexp.weight
    Q4_0
    [7168, 2048]
  • blk.8.ffn_norm.weight
    F32
    [7168]
  • blk.8.ffn_up_exps.weight
    Q4_0
    [7168, 2048, 256]
  • blk.8.ffn_up_shexp.weight
    Q4_0
    [7168, 2048]
    • blk.9
  • blk.9.attn_k_b.weight
    Q8_0
    [128, 512, 128]
  • blk.9.attn_kv_a_mqa.weight
    Q4_0
    [7168, 576]
  • blk.9.attn_kv_a_norm.weight
    F32
    [512]
  • blk.9.attn_norm.weight
    F32
    [7168]
  • blk.9.attn_output.weight
    Q4_0
    [16384, 7168]
  • blk.9.attn_q_a.weight
    Q4_0
    [7168, 1536]
  • blk.9.attn_q_a_norm.weight
    F32
    [1536]
  • blk.9.attn_q_b.weight
    Q4_0
    [1536, 24576]
  • blk.9.attn_v_b.weight
    Q8_0
    [512, 128, 128]
  • blk.9.exp_probs_b.bias
    F32
    [256]
  • blk.9.ffn_down_exps.weight
    Q4_0
    [2048, 7168, 256]
  • blk.9.ffn_down_shexp.weight
    Q4_0
    [2048, 7168]
  • blk.9.ffn_gate_exps.weight
    Q4_0
    [7168, 2048, 256]
  • blk.9.ffn_gate_inp.weight
    F32
    [7168, 256]
  • blk.9.ffn_gate_shexp.weight
    Q4_0
    [7168, 2048]
  • blk.9.ffn_norm.weight
    F32
    [7168]
  • blk.9.ffn_up_exps.weight
    Q4_0
    [7168, 2048, 256]
  • blk.9.ffn_up_shexp.weight
    Q4_0
    [7168, 2048]
    • blk.10
  • blk.10.attn_k_b.weight
    Q8_0
    [128, 512, 128]
  • blk.10.attn_kv_a_mqa.weight
    Q4_0
    [7168, 576]
  • blk.10.attn_kv_a_norm.weight
    F32
    [512]
  • blk.10.attn_norm.weight
    F32
    [7168]
  • blk.10.attn_output.weight
    Q4_0
    [16384, 7168]
  • blk.10.attn_q_a.weight
    Q4_0
    [7168, 1536]
  • blk.10.attn_q_a_norm.weight
    F32
    [1536]
  • blk.10.attn_q_b.weight
    Q4_0
    [1536, 24576]
  • blk.10.attn_v_b.weight
    Q8_0
    [512, 128, 128]
  • blk.10.exp_probs_b.bias
    F32
    [256]
  • output.weight
    Q6_K
    [7168, 129280]
  • output_norm.weight
    F32
    [7168]