orcarouter/ DeepSeek-V4-Flash-Vision-Uncensored:mxfp4

21 yesterday

... /
model
85d7c0db5c4a · 45GB
    Metadata
  • general.architecture
    deepseek4
  • general.file_type
    (!unknown_file_type 38!)
  • deepseek4.attention.compress_ratios
    [0, 0, 4, 128, 4, ...]
  • deepseek4.attention.compress_rope_freq_base
    160000
  • deepseek4.attention.head_count
    64
  • deepseek4.attention.head_count_kv
    1
  • deepseek4.attention.indexer.head_count
    64
  • deepseek4.attention.indexer.key_length
    128
  • deepseek4.attention.indexer.top_k
    512
  • deepseek4.attention.key_length
    512
  • deepseek4.attention.layer_norm_rms_epsilon
    1e-20
  • deepseek4.attention.output_group_count
    8
  • deepseek4.attention.output_lora_rank
    1024
  • deepseek4.attention.q_lora_rank
    1024
  • deepseek4.attention.sliding_window
    128
  • deepseek4.attention.value_length
    512
  • deepseek4.block_count
    43
  • deepseek4.context_length
    1048576
  • deepseek4.embedding_length
    4096
  • deepseek4.embedding_length_out
    16384
  • deepseek4.expert_count
    256
  • deepseek4.expert_feed_forward_length
    2048
  • deepseek4.expert_gating_func
    4
  • deepseek4.expert_shared_count
    1
  • deepseek4.expert_used_count
    6
  • deepseek4.expert_weights_norm
    true
  • deepseek4.expert_weights_scale
    1.5
  • deepseek4.hash_layer_count
    3
  • deepseek4.hyper_connection.count
    4
  • deepseek4.hyper_connection.epsilon
    1e-06
  • deepseek4.hyper_connection.sinkhorn_iterations
    20
  • deepseek4.rope.dimension_count
    64
  • deepseek4.rope.freq_base
    10000
  • deepseek4.rope.scaling.factor
    16
  • deepseek4.rope.scaling.original_context_length
    65536
  • deepseek4.rope.scaling.type
    yarn
  • deepseek4.rope.scaling.yarn_beta_fast
    32
  • deepseek4.rope.scaling.yarn_beta_slow
    1
  • deepseek4.swiglu_clamp_exp
    [10, 10, 10, 10, 10, ...]
  • deepseek4.swiglu_clamp_shexp
    [10, 10, 10, 10, 10, ...]
  • tokenizer.ggml.add_bos_token
    false
  • tokenizer.ggml.add_eos_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
    1
  • tokenizer.ggml.pre
    joyai-llm
  • tokenizer.ggml.token_type
    [3, 3, 3, 1, 1, ...]
  • tokenizer.ggml.tokens
    [<｜begin▁of▁sentence｜>, <｜end▁of▁sentence｜>, <｜▁pad▁｜>, !, ", ...]
  • split.count
    4
  • split.no
    0
  • split.tensors.count
    1328
    • Tensor
    blk.0
  • blk.0.ffn_down_exps.weight
    MXFP4
    [2048, 4096, 256]
  • blk.0.ffn_gate_exps.weight
    MXFP4
    [4096, 2048, 256]
  • blk.0.ffn_gate_tid2eid.weight
    I32
    [6, 129280]
  • blk.0.ffn_up_exps.weight
    MXFP4
    [4096, 2048, 256]
    • blk.1
  • blk.1.ffn_down_exps.weight
    MXFP4
    [2048, 4096, 256]
  • blk.1.ffn_gate_exps.weight
    MXFP4
    [4096, 2048, 256]
  • blk.1.ffn_gate_tid2eid.weight
    I32
    [6, 129280]
  • blk.1.ffn_up_exps.weight
    MXFP4
    [4096, 2048, 256]
    • blk.2
  • blk.2.ffn_down_exps.weight
    MXFP4
    [2048, 4096, 256]
  • blk.2.ffn_gate_exps.weight
    MXFP4
    [4096, 2048, 256]
  • blk.2.ffn_gate_tid2eid.weight
    I32
    [6, 129280]
  • blk.2.ffn_up_exps.weight
    MXFP4
    [4096, 2048, 256]
    • blk.3
  • blk.3.ffn_down_exps.weight
    MXFP4
    [2048, 4096, 256]
  • blk.3.ffn_gate_exps.weight
    MXFP4
    [4096, 2048, 256]
  • blk.3.ffn_up_exps.weight
    MXFP4
    [4096, 2048, 256]
    • blk.4
  • blk.4.ffn_down_exps.weight
    MXFP4
    [2048, 4096, 256]
  • blk.4.ffn_gate_exps.weight
    MXFP4
    [4096, 2048, 256]
  • blk.4.ffn_up_exps.weight
    MXFP4
    [4096, 2048, 256]
    • blk.5
  • blk.5.ffn_down_exps.weight
    MXFP4
    [2048, 4096, 256]
  • blk.5.ffn_gate_exps.weight
    MXFP4
    [4096, 2048, 256]
  • blk.5.ffn_up_exps.weight
    MXFP4
    [4096, 2048, 256]
    • blk.6
  • blk.6.ffn_down_exps.weight
    MXFP4
    [2048, 4096, 256]
  • blk.6.ffn_gate_exps.weight
    MXFP4
    [4096, 2048, 256]
  • blk.6.ffn_up_exps.weight
    MXFP4
    [4096, 2048, 256]
    • blk.7
  • blk.7.ffn_down_exps.weight
    MXFP4
    [2048, 4096, 256]
  • blk.7.ffn_gate_exps.weight
    MXFP4
    [4096, 2048, 256]
  • blk.7.ffn_up_exps.weight
    MXFP4
    [4096, 2048, 256]
    • blk.8
  • blk.8.ffn_down_exps.weight
    MXFP4
    [2048, 4096, 256]
  • blk.8.ffn_gate_exps.weight
    MXFP4
    [4096, 2048, 256]
  • blk.8.ffn_up_exps.weight
    MXFP4
    [4096, 2048, 256]
    • blk.9
  • blk.9.ffn_down_exps.weight
    MXFP4
    [2048, 4096, 256]
  • blk.9.ffn_gate_exps.weight
    MXFP4
    [4096, 2048, 256]
  • blk.9.ffn_up_exps.weight
    MXFP4
    [4096, 2048, 256]
    • blk.10
  • blk.10.ffn_down_exps.weight
    MXFP4
    [2048, 4096, 256]
  • blk.10.ffn_gate_exps.weight
    MXFP4
    [4096, 2048, 256]
  • blk.10.ffn_up_exps.weight
    MXFP4
    [4096, 2048, 256]
    • blk.11
  • blk.11.ffn_down_exps.weight
    MXFP4
    [2048, 4096, 256]
  • blk.11.ffn_gate_exps.weight
    MXFP4
    [4096, 2048, 256]
  • blk.11.ffn_up_exps.weight
    MXFP4
    [4096, 2048, 256]
    • blk.12
  • blk.12.ffn_down_exps.weight
    MXFP4
    [2048, 4096, 256]
  • blk.12.ffn_gate_exps.weight
    MXFP4
    [4096, 2048, 256]
  • blk.12.ffn_up_exps.weight
    MXFP4
    [4096, 2048, 256]