moophlo/ gemma-4-26B-A4B-it-GGUF:UD-IQ4_NL

10 2 hours ago

vision
... /
model
c2d131983ea4 · 13GB
    Metadata
  • general.architecture
    gemma4
  • general.file_type
    IQ4_NL
  • gemma4.attention.head_count
    16
  • gemma4.attention.head_count_kv
    [8, 8, 8, 8, 8, ...]
  • gemma4.attention.key_length
    512
  • gemma4.attention.key_length_swa
    256
  • gemma4.attention.layer_norm_rms_epsilon
    1e-06
  • gemma4.attention.shared_kv_layers
    0
  • gemma4.attention.sliding_window
    1024
  • gemma4.attention.sliding_window_pattern
    [true, true, true, true, true, ...]
  • gemma4.attention.value_length
    512
  • gemma4.attention.value_length_swa
    256
  • gemma4.block_count
    30
  • gemma4.context_length
    262144
  • gemma4.embedding_length
    2816
  • gemma4.embedding_length_per_layer_input
    0
  • gemma4.expert_count
    128
  • gemma4.expert_feed_forward_length
    704
  • gemma4.expert_used_count
    8
  • gemma4.feed_forward_length
    2112
  • gemma4.final_logit_softcapping
    30
  • gemma4.rope.dimension_count
    512
  • gemma4.rope.dimension_count_swa
    256
  • gemma4.rope.freq_base
    1e+06
  • gemma4.rope.freq_base_swa
    10000
  • tokenizer.ggml.add_bos_token
    false
  • tokenizer.ggml.add_space_prefix
    false
  • tokenizer.ggml.bos_token_id
    2
  • tokenizer.ggml.eos_token_id
    106
  • tokenizer.ggml.mask_token_id
    4
  • tokenizer.ggml.merges
    [ , ▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁ ▁, , , ▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁ ▁▁, ...]
  • tokenizer.ggml.model
    gemma4
  • tokenizer.ggml.padding_token_id
    0
  • tokenizer.ggml.scores
    [-1000, -1000, -1000, -1000, -1000, ...]
  • tokenizer.ggml.token_type
    [3, 1, 3, 3, 3, ...]
  • tokenizer.ggml.tokens
    [<pad>, <eos>, <bos>, <unk>, <mask>, ...]
  • tokenizer.ggml.unknown_token_id
    3
  • quantize.imatrix.chunks_count
    141
  • quantize.imatrix.dataset
    unsloth_calibration_gemma-4-26B-A4B-it.txt
  • quantize.imatrix.entries_count
    295
  • quantize.imatrix.file
    gemma-4-26B-A4B-it-GGUF/imatrix_unsloth.gguf
    • Tensor
  • token_embd.weight
    Q8_0
    [2816, 262144]
    • blk.0
  • blk.0.attn_k.weight
    Q8_0
    [2816, 2048]
  • blk.0.attn_k_norm.weight
    F32
    [256]
  • blk.0.attn_norm.weight
    F32
    [2816]
  • blk.0.attn_output.weight
    Q8_0
    [4096, 2816]
  • blk.0.attn_q.weight
    Q8_0
    [2816, 4096]
  • blk.0.attn_q_norm.weight
    F32
    [256]
  • blk.0.attn_v.weight
    Q8_0
    [2816, 2048]
  • blk.0.ffn_down.weight
    Q8_0
    [2112, 2816]
  • blk.0.ffn_down_exps.scale
    F32
    [128]
  • blk.0.ffn_down_exps.weight
    IQ4_NL
    [704, 2816, 128]
  • blk.0.ffn_gate.weight
    Q8_0
    [2816, 2112]
  • blk.0.ffn_gate_inp.scale
    F32
    [2816]
  • blk.0.ffn_gate_inp.weight
    F32
    [2816, 128]
  • blk.0.ffn_gate_up_exps.weight
    IQ3_S
    [2816, 1408, 128]
  • blk.0.ffn_norm.weight
    F32
    [2816]
  • blk.0.ffn_up.weight
    Q8_0
    [2816, 2112]
  • blk.0.layer_output_scale.weight
    F32
    [1]
  • blk.0.post_attention_norm.weight
    F32
    [2816]
  • blk.0.post_ffw_norm.weight
    F32
    [2816]
  • blk.0.post_ffw_norm_1.weight
    F32
    [2816]
  • blk.0.post_ffw_norm_2.weight
    F32
    [2816]
  • blk.0.pre_ffw_norm_2.weight
    F32
    [2816]
    • blk.1
  • blk.1.attn_k.weight
    Q8_0
    [2816, 2048]
  • blk.1.attn_k_norm.weight
    F32
    [256]
  • blk.1.attn_norm.weight
    F32
    [2816]
  • blk.1.attn_output.weight
    Q8_0
    [4096, 2816]
  • blk.1.attn_q.weight
    Q8_0
    [2816, 4096]
  • blk.1.attn_q_norm.weight
    F32
    [256]
  • blk.1.attn_v.weight
    Q8_0
    [2816, 2048]
  • blk.1.ffn_down.weight
    Q8_0
    [2112, 2816]
  • blk.1.ffn_down_exps.scale
    F32
    [128]
  • blk.1.ffn_down_exps.weight
    IQ4_NL
    [704, 2816, 128]
  • blk.1.ffn_gate.weight
    Q8_0
    [2816, 2112]
  • blk.1.ffn_gate_inp.scale
    F32
    [2816]
  • blk.1.ffn_gate_inp.weight
    F32
    [2816, 128]
  • blk.1.ffn_gate_up_exps.weight
    IQ3_S
    [2816, 1408, 128]
  • blk.1.ffn_norm.weight
    F32
    [2816]
  • blk.1.ffn_up.weight
    Q8_0
    [2816, 2112]
  • blk.1.layer_output_scale.weight
    F32
    [1]
  • blk.1.post_attention_norm.weight
    F32
    [2816]
  • blk.1.post_ffw_norm.weight
    F32
    [2816]
  • blk.1.post_ffw_norm_1.weight
    F32
    [2816]
  • blk.1.post_ffw_norm_2.weight
    F32
    [2816]
  • blk.1.pre_ffw_norm_2.weight
    F32
    [2816]
    • blk.2
  • blk.2.attn_k.weight
    Q8_0
    [2816, 2048]
  • blk.2.attn_k_norm.weight
    F32
    [256]
  • blk.2.attn_norm.weight
    F32
    [2816]
  • blk.2.attn_output.weight
    Q8_0
    [4096, 2816]
  • blk.2.attn_q.weight
    Q8_0
    [2816, 4096]
  • blk.2.attn_q_norm.weight
    F32
    [256]
  • blk.2.attn_v.weight
    Q8_0
    [2816, 2048]
  • blk.2.ffn_down.weight
    Q8_0
    [2112, 2816]
  • blk.2.ffn_down_exps.scale
    F32
    [128]
  • blk.2.ffn_down_exps.weight
    IQ4_NL
    [704, 2816, 128]
  • blk.2.ffn_gate.weight
    Q8_0
    [2816, 2112]
  • blk.2.ffn_gate_inp.scale
    F32
    [2816]
  • blk.2.ffn_gate_inp.weight
    F32
    [2816, 128]
  • blk.2.ffn_gate_up_exps.weight
    IQ3_S
    [2816, 1408, 128]
  • blk.2.ffn_norm.weight
    F32
    [2816]
  • blk.2.ffn_up.weight
    Q8_0
    [2816, 2112]
  • blk.2.layer_output_scale.weight
    F32
    [1]
  • blk.2.post_attention_norm.weight
    F32
    [2816]
  • blk.2.post_ffw_norm.weight
    F32
    [2816]
  • blk.2.post_ffw_norm_1.weight
    F32
    [2816]
  • blk.2.post_ffw_norm_2.weight
    F32
    [2816]
  • blk.2.pre_ffw_norm_2.weight
    F32
    [2816]
    • blk.3
  • blk.3.attn_k.weight
    Q8_0
    [2816, 2048]
  • blk.3.attn_k_norm.weight
    F32
    [256]
  • blk.3.attn_norm.weight
    F32
    [2816]
  • blk.3.attn_output.weight
    Q8_0
    [4096, 2816]
  • blk.3.attn_q.weight
    Q8_0
    [2816, 4096]
  • blk.3.attn_q_norm.weight
    F32
    [256]
  • blk.3.attn_v.weight
    Q8_0
    [2816, 2048]
  • blk.3.ffn_down.weight
    Q8_0
    [2112, 2816]
  • blk.3.ffn_down_exps.scale
    F32
    [128]
  • blk.3.ffn_down_exps.weight
    IQ4_NL
    [704, 2816, 128]
  • blk.3.ffn_gate.weight
    Q8_0
    [2816, 2112]
  • blk.3.ffn_gate_inp.scale
    F32
    [2816]
  • blk.3.ffn_gate_inp.weight
    F32
    [2816, 128]
  • blk.3.ffn_gate_up_exps.weight
    IQ3_S
    [2816, 1408, 128]
  • blk.3.ffn_norm.weight
    F32
    [2816]
  • blk.3.ffn_up.weight
    Q8_0
    [2816, 2112]
  • blk.3.layer_output_scale.weight
    F32
    [1]
  • blk.3.post_attention_norm.weight
    F32
    [2816]
  • blk.3.post_ffw_norm.weight
    F32
    [2816]
  • blk.3.post_ffw_norm_1.weight
    F32
    [2816]
  • blk.3.post_ffw_norm_2.weight
    F32
    [2816]
  • blk.3.pre_ffw_norm_2.weight
    F32
    [2816]
    • blk.4
  • blk.4.attn_k.weight
    Q8_0
    [2816, 2048]
  • blk.4.attn_k_norm.weight
    F32
    [256]
  • blk.4.attn_norm.weight
    F32
    [2816]
  • blk.4.attn_output.weight
    Q8_0
    [4096, 2816]
  • blk.4.attn_q.weight
    Q8_0
    [2816, 4096]
  • blk.4.attn_q_norm.weight
    F32
    [256]
  • blk.4.attn_v.weight
    Q8_0
    [2816, 2048]
  • blk.4.ffn_down.weight
    Q8_0
    [2112, 2816]
  • blk.4.ffn_down_exps.scale
    F32
    [128]
  • blk.4.ffn_down_exps.weight
    IQ4_NL
    [704, 2816, 128]
  • blk.4.ffn_gate.weight
    Q8_0
    [2816, 2112]
  • blk.4.ffn_gate_inp.scale
    F32
    [2816]
  • blk.4.ffn_gate_inp.weight
    F32
    [2816, 128]
  • blk.4.ffn_gate_up_exps.weight
    IQ3_S
    [2816, 1408, 128]
  • blk.4.ffn_norm.weight
    F32
    [2816]
  • blk.4.ffn_up.weight
    Q8_0
    [2816, 2112]
  • blk.4.layer_output_scale.weight
    F32
    [1]
  • blk.4.post_attention_norm.weight
    F32
    [2816]
  • blk.4.post_ffw_norm.weight
    F32
    [2816]
  • blk.4.post_ffw_norm_1.weight
    F32
    [2816]
  • blk.4.post_ffw_norm_2.weight
    F32
    [2816]
  • blk.4.pre_ffw_norm_2.weight
    F32
    [2816]
    • blk.5
  • blk.5.attn_k.weight
    Q8_0
    [2816, 1024]
  • blk.5.attn_k_norm.weight
    F32
    [512]
  • blk.5.attn_norm.weight
    F32
    [2816]
  • blk.5.attn_output.weight
    Q8_0
    [8192, 2816]
  • blk.5.attn_q.weight
    Q8_0
    [2816, 8192]
  • blk.5.attn_q_norm.weight
    F32
    [512]
  • blk.5.ffn_down.weight
    Q8_0
    [2112, 2816]
  • blk.5.ffn_down_exps.scale
    F32
    [128]
  • blk.5.ffn_down_exps.weight
    IQ4_NL
    [704, 2816, 128]
  • blk.5.ffn_gate.weight
    Q8_0
    [2816, 2112]
  • blk.5.ffn_gate_inp.scale
    F32
    [2816]
  • blk.5.ffn_gate_inp.weight
    F32
    [2816, 128]
  • blk.5.ffn_gate_up_exps.weight
    IQ3_S
    [2816, 1408, 128]
  • blk.5.ffn_norm.weight
    F32
    [2816]
  • blk.5.ffn_up.weight
    Q8_0
    [2816, 2112]
  • blk.5.layer_output_scale.weight
    F32
    [1]
  • blk.5.post_attention_norm.weight
    F32
    [2816]
  • blk.5.post_ffw_norm.weight
    F32
    [2816]
  • blk.5.post_ffw_norm_1.weight
    F32
    [2816]
  • blk.5.post_ffw_norm_2.weight
    F32
    [2816]
  • blk.5.pre_ffw_norm_2.weight
    F32
    [2816]
    • blk.6
  • blk.6.attn_k.weight
    Q8_0
    [2816, 2048]
  • blk.6.attn_k_norm.weight
    F32
    [256]
  • blk.6.attn_norm.weight
    F32
    [2816]
  • blk.6.attn_output.weight
    Q8_0
    [4096, 2816]
  • blk.6.attn_q.weight
    Q8_0
    [2816, 4096]
  • blk.6.attn_q_norm.weight
    F32
    [256]
  • blk.6.attn_v.weight
    Q8_0
    [2816, 2048]
  • blk.6.ffn_down.weight
    Q8_0
    [2112, 2816]
  • blk.6.ffn_down_exps.scale
    F32
    [128]
  • blk.6.ffn_down_exps.weight
    IQ4_NL
    [704, 2816, 128]
  • blk.6.ffn_gate.weight
    Q8_0
    [2816, 2112]
  • blk.6.ffn_gate_inp.scale
    F32
    [2816]
  • blk.6.ffn_gate_inp.weight
    F32
    [2816, 128]
  • blk.6.ffn_gate_up_exps.weight
    IQ3_S
    [2816, 1408, 128]
  • blk.6.ffn_norm.weight
    F32
    [2816]
  • blk.6.ffn_up.weight
    Q8_0
    [2816, 2112]
  • blk.6.layer_output_scale.weight
    F32
    [1]
  • blk.6.post_attention_norm.weight
    F32
    [2816]
  • blk.6.post_ffw_norm.weight
    F32
    [2816]
  • blk.6.post_ffw_norm_1.weight
    F32
    [2816]
  • blk.6.post_ffw_norm_2.weight
    F32
    [2816]
  • blk.6.pre_ffw_norm_2.weight
    F32
    [2816]
    • blk.7
  • blk.7.attn_k.weight
    Q8_0
    [2816, 2048]
  • blk.7.attn_k_norm.weight
    F32
    [256]
  • blk.7.attn_norm.weight
    F32
    [2816]
  • blk.7.attn_output.weight
    Q8_0
    [4096, 2816]
  • blk.7.attn_q.weight
    Q8_0
    [2816, 4096]
  • blk.7.attn_q_norm.weight
    F32
    [256]
  • blk.7.attn_v.weight
    Q8_0
    [2816, 2048]
  • blk.7.ffn_down.weight
    Q8_0
    [2112, 2816]
  • blk.7.ffn_down_exps.scale
    F32
    [128]
  • blk.7.ffn_down_exps.weight
    IQ4_NL
    [704, 2816, 128]
  • blk.7.ffn_gate.weight
    Q8_0
    [2816, 2112]
  • blk.7.ffn_gate_inp.scale
    F32
    [2816]
  • blk.7.ffn_gate_inp.weight
    F32
    [2816, 128]
  • blk.7.ffn_gate_up_exps.weight
    IQ3_S
    [2816, 1408, 128]
  • blk.7.ffn_norm.weight
    F32
    [2816]
  • blk.7.ffn_up.weight
    Q8_0
    [2816, 2112]
  • blk.7.layer_output_scale.weight
    F32
    [1]
  • blk.7.post_attention_norm.weight
    F32
    [2816]
  • blk.7.post_ffw_norm.weight
    F32
    [2816]
  • blk.7.post_ffw_norm_1.weight
    F32
    [2816]
  • blk.7.post_ffw_norm_2.weight
    F32
    [2816]
  • blk.7.pre_ffw_norm_2.weight
    F32
    [2816]
    • blk.8
  • blk.8.attn_k.weight
    Q8_0
    [2816, 2048]
  • blk.8.attn_k_norm.weight
    F32
    [256]
  • blk.8.attn_norm.weight
    F32
    [2816]
  • blk.8.attn_output.weight
    Q8_0
    [4096, 2816]
  • blk.8.attn_q.weight
    Q8_0
    [2816, 4096]
  • blk.8.attn_q_norm.weight
    F32
    [256]
  • blk.8.attn_v.weight
    Q8_0
    [2816, 2048]
  • blk.8.ffn_down.weight
    Q8_0
    [2112, 2816]
  • blk.8.ffn_down_exps.scale
    F32
    [128]
  • blk.8.ffn_down_exps.weight
    IQ4_NL
    [704, 2816, 128]
  • blk.8.ffn_gate.weight
    Q8_0
    [2816, 2112]
  • blk.8.ffn_gate_inp.scale
    F32
    [2816]
  • blk.8.ffn_gate_inp.weight
    F32
    [2816, 128]
  • blk.8.ffn_gate_up_exps.weight
    IQ3_S
    [2816, 1408, 128]
  • blk.8.ffn_norm.weight
    F32
    [2816]
  • blk.8.ffn_up.weight
    Q8_0
    [2816, 2112]
  • blk.8.layer_output_scale.weight
    F32
    [1]
  • blk.8.post_attention_norm.weight
    F32
    [2816]
  • blk.8.post_ffw_norm.weight
    F32
    [2816]
  • blk.8.post_ffw_norm_1.weight
    F32
    [2816]
  • blk.8.post_ffw_norm_2.weight
    F32
    [2816]
  • blk.8.pre_ffw_norm_2.weight
    F32
    [2816]
    • blk.9
  • blk.9.attn_k.weight
    Q8_0
    [2816, 2048]
  • blk.9.attn_k_norm.weight
    F32
    [256]
  • blk.9.attn_norm.weight
    F32
    [2816]
  • blk.9.attn_output.weight
    Q8_0
    [4096, 2816]
  • blk.9.attn_q.weight
    Q8_0
    [2816, 4096]
  • blk.9.attn_q_norm.weight
    F32
    [256]
  • blk.9.attn_v.weight
    Q8_0
    [2816, 2048]
  • blk.9.ffn_down.weight
    Q8_0
    [2112, 2816]
  • blk.9.ffn_down_exps.scale
    F32
    [128]
  • blk.9.ffn_down_exps.weight
    IQ4_NL
    [704, 2816, 128]
  • blk.9.ffn_gate.weight
    Q8_0
    [2816, 2112]
  • blk.9.ffn_gate_inp.scale
    F32
    [2816]
  • blk.9.ffn_gate_inp.weight
    F32
    [2816, 128]
  • blk.9.ffn_gate_up_exps.weight
    IQ3_S
    [2816, 1408, 128]
  • blk.9.ffn_norm.weight
    F32
    [2816]
  • blk.9.ffn_up.weight
    Q8_0
    [2816, 2112]
  • blk.9.layer_output_scale.weight
    F32
    [1]
  • blk.9.post_attention_norm.weight
    F32
    [2816]
  • blk.9.post_ffw_norm.weight
    F32
    [2816]
  • blk.9.post_ffw_norm_1.weight
    F32
    [2816]
  • blk.9.post_ffw_norm_2.weight
    F32
    [2816]
  • blk.9.pre_ffw_norm_2.weight
    F32
    [2816]
    • blk.10
  • blk.10.attn_k.weight
    Q8_0
    [2816, 2048]
  • blk.10.attn_k_norm.weight
    F32
    [256]
  • blk.10.attn_norm.weight
    F32
    [2816]
  • blk.10.attn_output.weight
    Q8_0
    [4096, 2816]
  • blk.10.attn_q.weight
    Q8_0
    [2816, 4096]
  • blk.10.attn_q_norm.weight
    F32
    [256]
  • blk.10.attn_v.weight
    Q8_0
    [2816, 2048]
  • blk.10.ffn_down.weight
    Q8_0
    [2112, 2816]
  • blk.10.ffn_down_exps.scale
    F32
    [128]
  • blk.10.ffn_down_exps.weight
    IQ4_NL
    [704, 2816, 128]
  • blk.10.ffn_gate.weight
    Q8_0
    [2816, 2112]
  • blk.10.ffn_gate_inp.scale
    F32
    [2816]
  • blk.10.ffn_gate_inp.weight
    F32
    [2816, 128]
  • blk.10.ffn_gate_up_exps.weight
    IQ3_S
    [2816, 1408, 128]
  • blk.10.ffn_norm.weight
    F32
    [2816]
  • blk.10.ffn_up.weight
    Q8_0
    [2816, 2112]
  • blk.10.layer_output_scale.weight
    F32
    [1]
  • blk.10.post_attention_norm.weight
    F32
    [2816]
  • blk.10.post_ffw_norm.weight
    F32
    [2816]
  • blk.10.post_ffw_norm_1.weight
    F32
    [2816]
  • blk.10.post_ffw_norm_2.weight
    F32
    [2816]
  • blk.10.pre_ffw_norm_2.weight
    F32
    [2816]
    • blk.11
  • blk.11.attn_k.weight
    Q8_0
    [2816, 1024]
  • blk.11.attn_k_norm.weight
    F32
    [512]
  • blk.11.attn_norm.weight
    F32
    [2816]
  • blk.11.attn_output.weight
    Q8_0
    [8192, 2816]
  • blk.11.attn_q.weight
    Q8_0
    [2816, 8192]
  • blk.11.attn_q_norm.weight
    F32
    [512]
  • blk.11.ffn_down.weight
    Q8_0
    [2112, 2816]
  • blk.11.ffn_down_exps.scale
    F32
    [128]
  • blk.11.ffn_down_exps.weight
    IQ4_NL
    [704, 2816, 128]
  • blk.11.ffn_gate.weight
    Q8_0
    [2816, 2112]
  • blk.11.ffn_gate_inp.scale
    F32
    [2816]
  • blk.11.ffn_gate_inp.weight
    F32
    [2816, 128]
  • blk.11.ffn_gate_up_exps.weight
    IQ3_S
    [2816, 1408, 128]
  • blk.11.ffn_norm.weight
    F32
    [2816]
  • blk.11.ffn_up.weight
    Q8_0
    [2816, 2112]
  • blk.11.layer_output_scale.weight
    F32
    [1]
  • blk.11.post_attention_norm.weight
    F32
    [2816]
  • blk.11.post_ffw_norm.weight
    F32
    [2816]
  • blk.11.post_ffw_norm_1.weight
    F32
    [2816]
  • blk.11.post_ffw_norm_2.weight
    F32
    [2816]
  • blk.11.pre_ffw_norm_2.weight
    F32
    [2816]
    • blk.12
  • blk.12.attn_k.weight
    Q8_0
    [2816, 2048]
  • blk.12.attn_k_norm.weight
    F32
    [256]
  • blk.12.attn_norm.weight
    F32
    [2816]
  • blk.12.attn_output.weight
    Q8_0
    [4096, 2816]
  • blk.12.attn_q.weight
    Q8_0
    [2816, 4096]
  • blk.12.attn_q_norm.weight
    F32
    [256]
  • blk.12.attn_v.weight
    Q8_0
    [2816, 2048]
  • blk.12.ffn_down.weight
    Q8_0
    [2112, 2816]
  • blk.12.ffn_down_exps.scale
    F32
    [128]
  • blk.12.ffn_down_exps.weight
    IQ4_NL
    [704, 2816, 128]
  • blk.12.ffn_gate.weight
    Q8_0
    [2816, 2112]
  • blk.12.ffn_gate_inp.scale
    F32
    [2816]
  • blk.12.ffn_gate_inp.weight
    F32
    [2816, 128]
  • blk.12.ffn_gate_up_exps.weight
    IQ3_S
    [2816, 1408, 128]
  • blk.12.ffn_norm.weight
    F32
    [2816]
  • blk.12.ffn_up.weight
    Q8_0
    [2816, 2112]
  • blk.12.layer_output_scale.weight
    F32
    [1]
  • blk.12.post_attention_norm.weight
    F32
    [2816]
  • blk.12.post_ffw_norm.weight
    F32
    [2816]
  • blk.12.post_ffw_norm_1.weight
    F32
    [2816]
  • blk.12.post_ffw_norm_2.weight
    F32
    [2816]
  • blk.12.pre_ffw_norm_2.weight
    F32
    [2816]
    • blk.13
  • blk.13.attn_k.weight
    Q8_0
    [2816, 2048]
  • blk.13.attn_k_norm.weight
    F32
    [256]
  • blk.13.attn_norm.weight
    F32
    [2816]
  • blk.13.attn_output.weight
    Q8_0
    [4096, 2816]
  • blk.13.attn_q.weight
    Q8_0
    [2816, 4096]
  • blk.13.attn_q_norm.weight
    F32
    [256]
  • blk.13.attn_v.weight
    Q8_0
    [2816, 2048]
  • blk.13.ffn_down.weight
    Q8_0
    [2112, 2816]
  • blk.13.ffn_down_exps.scale
    F32
    [128]
  • blk.13.ffn_down_exps.weight
    IQ4_NL
    [704, 2816, 128]
  • blk.13.ffn_gate.weight
    Q8_0
    [2816, 2112]
  • blk.13.ffn_gate_inp.scale
    F32
    [2816]
  • blk.13.ffn_gate_inp.weight
    F32
    [2816, 128]
  • blk.13.ffn_gate_up_exps.weight
    IQ3_S
    [2816, 1408, 128]
  • blk.13.ffn_norm.weight
    F32
    [2816]
  • blk.13.ffn_up.weight
    Q8_0
    [2816, 2112]
  • blk.13.layer_output_scale.weight
    F32
    [1]
  • blk.13.post_attention_norm.weight
    F32
    [2816]
  • blk.13.post_ffw_norm.weight
    F32
    [2816]
  • blk.13.post_ffw_norm_1.weight
    F32
    [2816]
  • blk.13.post_ffw_norm_2.weight
    F32
    [2816]
  • blk.13.pre_ffw_norm_2.weight
    F32
    [2816]
    • blk.14
  • blk.14.attn_k.weight
    Q8_0
    [2816, 2048]
  • blk.14.attn_k_norm.weight
    F32
    [256]
  • blk.14.attn_norm.weight
    F32
    [2816]
  • blk.14.attn_output.weight
    Q8_0
    [4096, 2816]
  • blk.14.attn_q.weight
    Q8_0
    [2816, 4096]
  • blk.14.attn_q_norm.weight
    F32
    [256]
  • blk.14.attn_v.weight
    Q8_0
    [2816, 2048]
  • blk.14.ffn_down.weight
    Q8_0
    [2112, 2816]
  • blk.14.ffn_down_exps.scale
    F32
    [128]
  • blk.14.ffn_down_exps.weight
    IQ4_NL
    [704, 2816, 128]
  • blk.14.ffn_gate.weight
    Q8_0
    [2816, 2112]
  • blk.14.ffn_gate_inp.scale
    F32
    [2816]
  • blk.14.ffn_gate_inp.weight
    F32
    [2816, 128]
  • blk.14.ffn_gate_up_exps.weight
    IQ3_S
    [2816, 1408, 128]
  • blk.14.ffn_norm.weight
    F32
    [2816]
  • blk.14.ffn_up.weight
    Q8_0
    [2816, 2112]
  • blk.14.layer_output_scale.weight
    F32
    [1]
  • blk.14.post_attention_norm.weight
    F32
    [2816]
  • blk.14.post_ffw_norm.weight
    F32
    [2816]
  • blk.14.post_ffw_norm_1.weight
    F32
    [2816]
  • blk.14.post_ffw_norm_2.weight
    F32
    [2816]
  • blk.14.pre_ffw_norm_2.weight
    F32
    [2816]
    • blk.15
  • blk.15.attn_k.weight
    Q8_0
    [2816, 2048]
  • blk.15.attn_k_norm.weight
    F32
    [256]
  • blk.15.attn_norm.weight
    F32
    [2816]
  • blk.15.attn_output.weight
    Q8_0
    [4096, 2816]
  • blk.15.attn_q.weight
    Q8_0
    [2816, 4096]
  • blk.15.attn_q_norm.weight
    F32
    [256]
  • blk.15.attn_v.weight
    Q8_0
    [2816, 2048]
  • blk.15.ffn_down.weight
    Q8_0
    [2112, 2816]
  • blk.15.ffn_down_exps.scale
    F32
    [128]
  • blk.15.ffn_down_exps.weight
    IQ4_NL
    [704, 2816, 128]
  • blk.15.ffn_gate.weight
    Q8_0
    [2816, 2112]
  • blk.15.ffn_gate_inp.scale
    F32
    [2816]
  • blk.15.ffn_gate_inp.weight
    F32
    [2816, 128]
  • blk.15.ffn_gate_up_exps.weight
    IQ3_S
    [2816, 1408, 128]
  • blk.15.ffn_norm.weight
    F32
    [2816]
  • blk.15.ffn_up.weight
    Q8_0
    [2816, 2112]
  • blk.15.layer_output_scale.weight
    F32
    [1]
  • blk.15.post_attention_norm.weight
    F32
    [2816]
  • blk.15.post_ffw_norm.weight
    F32
    [2816]
  • blk.15.post_ffw_norm_1.weight
    F32
    [2816]
  • blk.15.post_ffw_norm_2.weight
    F32
    [2816]
  • blk.15.pre_ffw_norm_2.weight
    F32
    [2816]
    • blk.16
  • blk.16.attn_k.weight
    Q8_0
    [2816, 2048]
  • blk.16.attn_k_norm.weight
    F32
    [256]
  • blk.16.attn_norm.weight
    F32
    [2816]
  • blk.16.attn_output.weight
    Q8_0
    [4096, 2816]
  • blk.16.attn_q.weight
    Q8_0
    [2816, 4096]
  • blk.16.attn_q_norm.weight
    F32
    [256]
  • blk.16.attn_v.weight
    Q8_0
    [2816, 2048]
  • blk.16.ffn_down.weight
    Q8_0
    [2112, 2816]
  • blk.16.ffn_down_exps.scale
    F32
    [128]
  • blk.16.ffn_down_exps.weight
    IQ4_NL
    [704, 2816, 128]
  • blk.16.ffn_gate.weight
    Q8_0
    [2816, 2112]
  • blk.16.ffn_gate_inp.scale
    F32
    [2816]
  • blk.16.ffn_gate_inp.weight
    F32
    [2816, 128]
  • blk.16.ffn_gate_up_exps.weight
    IQ3_S
    [2816, 1408, 128]
  • blk.16.ffn_norm.weight
    F32
    [2816]
  • blk.16.ffn_up.weight
    Q8_0
    [2816, 2112]
  • blk.16.layer_output_scale.weight
    F32
    [1]
  • blk.16.post_attention_norm.weight
    F32
    [2816]
  • blk.16.post_ffw_norm.weight
    F32
    [2816]
  • blk.16.post_ffw_norm_1.weight
    F32
    [2816]
  • blk.16.post_ffw_norm_2.weight
    F32
    [2816]
  • blk.16.pre_ffw_norm_2.weight
    F32
    [2816]
    • blk.17
  • blk.17.attn_k.weight
    Q8_0
    [2816, 1024]
  • blk.17.attn_k_norm.weight
    F32
    [512]
  • blk.17.attn_norm.weight
    F32
    [2816]
  • blk.17.attn_output.weight
    Q8_0
    [8192, 2816]
  • blk.17.attn_q.weight
    Q8_0
    [2816, 8192]
  • blk.17.attn_q_norm.weight
    F32
    [512]
  • blk.17.ffn_down.weight
    Q8_0
    [2112, 2816]
  • blk.17.ffn_down_exps.scale
    F32
    [128]
  • blk.17.ffn_down_exps.weight
    IQ4_NL
    [704, 2816, 128]
  • blk.17.ffn_gate.weight
    Q8_0
    [2816, 2112]
  • blk.17.ffn_gate_inp.scale
    F32
    [2816]
  • blk.17.ffn_gate_inp.weight
    F32
    [2816, 128]
  • blk.17.ffn_gate_up_exps.weight
    IQ3_S
    [2816, 1408, 128]
  • blk.17.ffn_norm.weight
    F32
    [2816]
  • blk.17.ffn_up.weight
    Q8_0
    [2816, 2112]
  • blk.17.layer_output_scale.weight
    F32
    [1]
  • blk.17.post_attention_norm.weight
    F32
    [2816]
  • blk.17.post_ffw_norm.weight
    F32
    [2816]
  • blk.17.post_ffw_norm_1.weight
    F32
    [2816]
  • blk.17.post_ffw_norm_2.weight
    F32
    [2816]
  • blk.17.pre_ffw_norm_2.weight
    F32
    [2816]
    • blk.18
  • blk.18.attn_k.weight
    Q8_0
    [2816, 2048]
  • blk.18.attn_k_norm.weight
    F32
    [256]
  • blk.18.attn_norm.weight
    F32
    [2816]
  • blk.18.attn_output.weight
    Q8_0
    [4096, 2816]
  • blk.18.attn_q.weight
    Q8_0
    [2816, 4096]
  • blk.18.attn_q_norm.weight
    F32
    [256]
  • blk.18.attn_v.weight
    Q8_0
    [2816, 2048]
  • blk.18.ffn_down.weight
    Q8_0
    [2112, 2816]
  • blk.18.ffn_down_exps.scale
    F32
    [128]
  • blk.18.ffn_down_exps.weight
    IQ4_NL
    [704, 2816, 128]
  • blk.18.ffn_gate.weight
    Q8_0
    [2816, 2112]
  • blk.18.ffn_gate_inp.scale
    F32
    [2816]
  • blk.18.ffn_gate_inp.weight
    F32
    [2816, 128]
  • blk.18.ffn_gate_up_exps.weight
    IQ3_S
    [2816, 1408, 128]
  • blk.18.ffn_norm.weight
    F32
    [2816]
  • blk.18.ffn_up.weight
    Q8_0
    [2816, 2112]
  • blk.18.layer_output_scale.weight
    F32
    [1]
  • blk.18.post_attention_norm.weight
    F32
    [2816]
  • blk.18.post_ffw_norm.weight
    F32
    [2816]
  • blk.18.post_ffw_norm_1.weight
    F32
    [2816]
  • blk.18.post_ffw_norm_2.weight
    F32
    [2816]
  • blk.18.pre_ffw_norm_2.weight
    F32
    [2816]
    • blk.19
  • blk.19.attn_k.weight
    Q8_0
    [2816, 2048]
  • blk.19.attn_k_norm.weight
    F32
    [256]
  • blk.19.attn_norm.weight
    F32
    [2816]
  • blk.19.attn_output.weight
    Q8_0
    [4096, 2816]
  • blk.19.attn_q.weight
    Q8_0
    [2816, 4096]
  • blk.19.attn_q_norm.weight
    F32
    [256]
  • blk.19.attn_v.weight
    Q8_0
    [2816, 2048]
  • blk.19.ffn_down.weight
    Q8_0
    [2112, 2816]
  • blk.19.ffn_down_exps.scale
    F32
    [128]
  • blk.19.ffn_down_exps.weight
    IQ4_NL
    [704, 2816, 128]
  • blk.19.ffn_gate.weight
    Q8_0
    [2816, 2112]
  • blk.19.ffn_gate_inp.scale
    F32
    [2816]
  • blk.19.ffn_gate_inp.weight
    F32
    [2816, 128]
  • blk.19.ffn_gate_up_exps.weight
    IQ3_S
    [2816, 1408, 128]
  • blk.19.ffn_norm.weight
    F32
    [2816]
  • blk.19.ffn_up.weight
    Q8_0
    [2816, 2112]
  • blk.19.layer_output_scale.weight
    F32
    [1]
  • blk.19.post_attention_norm.weight
    F32
    [2816]
  • blk.19.post_ffw_norm.weight
    F32
    [2816]
  • blk.19.post_ffw_norm_1.weight
    F32
    [2816]
  • blk.19.post_ffw_norm_2.weight
    F32
    [2816]
  • blk.19.pre_ffw_norm_2.weight
    F32
    [2816]
    • blk.20
  • blk.20.attn_k.weight
    Q8_0
    [2816, 2048]
  • blk.20.attn_k_norm.weight
    F32
    [256]
  • blk.20.attn_norm.weight
    F32
    [2816]
  • blk.20.attn_output.weight
    Q8_0
    [4096, 2816]
  • blk.20.attn_q.weight
    Q8_0
    [2816, 4096]
  • blk.20.attn_q_norm.weight
    F32
    [256]
  • blk.20.attn_v.weight
    Q8_0
    [2816, 2048]
  • blk.20.ffn_down.weight
    Q8_0
    [2112, 2816]
  • blk.20.ffn_down_exps.scale
    F32
    [128]
  • blk.20.ffn_down_exps.weight
    IQ4_NL
    [704, 2816, 128]
  • blk.20.ffn_gate.weight
    Q8_0
    [2816, 2112]
  • blk.20.ffn_gate_inp.scale
    F32
    [2816]
  • blk.20.ffn_gate_inp.weight
    F32
    [2816, 128]
  • blk.20.ffn_gate_up_exps.weight
    IQ3_S
    [2816, 1408, 128]
  • blk.20.ffn_norm.weight
    F32
    [2816]
  • blk.20.ffn_up.weight
    Q8_0
    [2816, 2112]
  • blk.20.layer_output_scale.weight
    F32
    [1]
  • blk.20.post_attention_norm.weight
    F32
    [2816]
  • blk.20.post_ffw_norm.weight
    F32
    [2816]
  • blk.20.post_ffw_norm_1.weight
    F32
    [2816]
  • blk.20.post_ffw_norm_2.weight
    F32
    [2816]
  • blk.20.pre_ffw_norm_2.weight
    F32
    [2816]
    • blk.21
  • blk.21.attn_k.weight
    Q8_0
    [2816, 2048]
  • blk.21.attn_k_norm.weight
    F32
    [256]
  • blk.21.attn_norm.weight
    F32
    [2816]
  • blk.21.attn_output.weight
    Q8_0
    [4096, 2816]
  • blk.21.attn_q.weight
    Q8_0
    [2816, 4096]
  • blk.21.attn_q_norm.weight
    F32
    [256]
  • blk.21.attn_v.weight
    Q8_0
    [2816, 2048]
  • blk.21.ffn_down.weight
    Q8_0
    [2112, 2816]
  • blk.21.ffn_down_exps.scale
    F32
    [128]
  • blk.21.ffn_down_exps.weight
    IQ4_NL
    [704, 2816, 128]
  • blk.21.ffn_gate.weight
    Q8_0
    [2816, 2112]
  • blk.21.ffn_gate_inp.scale
    F32
    [2816]
  • blk.21.ffn_gate_inp.weight
    F32
    [2816, 128]
  • blk.21.ffn_gate_up_exps.weight
    IQ3_S
    [2816, 1408, 128]
  • blk.21.ffn_norm.weight
    F32
    [2816]
  • blk.21.ffn_up.weight
    Q8_0
    [2816, 2112]
  • blk.21.layer_output_scale.weight
    F32
    [1]
  • blk.21.post_attention_norm.weight
    F32
    [2816]
  • blk.21.post_ffw_norm.weight
    F32
    [2816]
  • blk.21.post_ffw_norm_1.weight
    F32
    [2816]
  • blk.21.post_ffw_norm_2.weight
    F32
    [2816]
  • blk.21.pre_ffw_norm_2.weight
    F32
    [2816]
    • blk.22
  • blk.22.attn_k.weight
    Q8_0
    [2816, 2048]
  • blk.22.attn_k_norm.weight
    F32
    [256]
  • blk.22.attn_norm.weight
    F32
    [2816]
  • blk.22.attn_output.weight
    Q8_0
    [4096, 2816]
  • blk.22.attn_q.weight
    Q8_0
    [2816, 4096]
  • blk.22.attn_q_norm.weight
    F32
    [256]
  • blk.22.attn_v.weight
    Q8_0
    [2816, 2048]
  • blk.22.ffn_down.weight
    Q8_0
    [2112, 2816]
  • blk.22.ffn_down_exps.scale
    F32
    [128]
  • blk.22.ffn_down_exps.weight
    IQ4_NL
    [704, 2816, 128]
  • blk.22.ffn_gate.weight
    Q8_0
    [2816, 2112]
  • blk.22.ffn_gate_inp.scale
    F32
    [2816]
  • blk.22.ffn_gate_inp.weight
    F32
    [2816, 128]
  • blk.22.ffn_gate_up_exps.weight
    IQ3_S
    [2816, 1408, 128]
  • blk.22.ffn_norm.weight
    F32
    [2816]
  • blk.22.ffn_up.weight
    Q8_0
    [2816, 2112]
  • blk.22.layer_output_scale.weight
    F32
    [1]
  • blk.22.post_attention_norm.weight
    F32
    [2816]
  • blk.22.post_ffw_norm.weight
    F32
    [2816]
  • blk.22.post_ffw_norm_1.weight
    F32
    [2816]
  • blk.22.post_ffw_norm_2.weight
    F32
    [2816]
  • blk.22.pre_ffw_norm_2.weight
    F32
    [2816]
    • blk.23
  • blk.23.attn_k.weight
    Q8_0
    [2816, 1024]
  • blk.23.attn_k_norm.weight
    F32
    [512]
  • blk.23.attn_norm.weight
    F32
    [2816]
  • blk.23.attn_output.weight
    Q8_0
    [8192, 2816]
  • blk.23.attn_q.weight
    Q8_0
    [2816, 8192]
  • blk.23.attn_q_norm.weight
    F32
    [512]
  • blk.23.ffn_down.weight
    Q8_0
    [2112, 2816]
  • blk.23.ffn_down_exps.scale
    F32
    [128]
  • blk.23.ffn_down_exps.weight
    IQ4_NL
    [704, 2816, 128]
  • blk.23.ffn_gate.weight
    Q8_0
    [2816, 2112]
  • blk.23.ffn_gate_inp.scale
    F32
    [2816]
  • blk.23.ffn_gate_inp.weight
    F32
    [2816, 128]
  • blk.23.ffn_gate_up_exps.weight
    IQ3_S
    [2816, 1408, 128]
  • blk.23.ffn_norm.weight
    F32
    [2816]
  • blk.23.ffn_up.weight
    Q8_0
    [2816, 2112]
  • blk.23.layer_output_scale.weight
    F32
    [1]
  • blk.23.post_attention_norm.weight
    F32
    [2816]
  • blk.23.post_ffw_norm.weight
    F32
    [2816]
  • blk.23.post_ffw_norm_1.weight
    F32
    [2816]
  • blk.23.post_ffw_norm_2.weight
    F32
    [2816]
  • blk.23.pre_ffw_norm_2.weight
    F32
    [2816]
    • blk.24
  • blk.24.attn_k.weight
    Q8_0
    [2816, 2048]
  • blk.24.attn_k_norm.weight
    F32
    [256]
  • blk.24.attn_norm.weight
    F32
    [2816]
  • blk.24.attn_output.weight
    Q8_0
    [4096, 2816]
  • blk.24.attn_q.weight
    Q8_0
    [2816, 4096]
  • blk.24.attn_q_norm.weight
    F32
    [256]
  • blk.24.attn_v.weight
    Q8_0
    [2816, 2048]
  • blk.24.ffn_down.weight
    Q8_0
    [2112, 2816]
  • blk.24.ffn_down_exps.scale
    F32
    [128]
  • blk.24.ffn_down_exps.weight
    IQ4_NL
    [704, 2816, 128]
  • blk.24.ffn_gate.weight
    Q8_0
    [2816, 2112]
  • blk.24.ffn_gate_inp.scale
    F32
    [2816]
  • blk.24.ffn_gate_inp.weight
    F32
    [2816, 128]
  • blk.24.ffn_gate_up_exps.weight
    IQ3_S
    [2816, 1408, 128]
  • blk.24.ffn_norm.weight
    F32
    [2816]
  • blk.24.ffn_up.weight
    Q8_0
    [2816, 2112]
  • blk.24.layer_output_scale.weight
    F32
    [1]
  • blk.24.post_attention_norm.weight
    F32
    [2816]
  • blk.24.post_ffw_norm.weight
    F32
    [2816]
  • blk.24.post_ffw_norm_1.weight
    F32
    [2816]
  • blk.24.post_ffw_norm_2.weight
    F32
    [2816]
  • blk.24.pre_ffw_norm_2.weight
    F32
    [2816]
    • blk.25
  • blk.25.attn_k.weight
    Q8_0
    [2816, 2048]
  • blk.25.attn_k_norm.weight
    F32
    [256]
  • blk.25.attn_norm.weight
    F32
    [2816]
  • blk.25.attn_output.weight
    Q8_0
    [4096, 2816]
  • blk.25.attn_q.weight
    Q8_0
    [2816, 4096]
  • blk.25.attn_q_norm.weight
    F32
    [256]
  • blk.25.attn_v.weight
    Q8_0
    [2816, 2048]
  • blk.25.ffn_down.weight
    Q8_0
    [2112, 2816]
  • blk.25.ffn_down_exps.scale
    F32
    [128]
  • blk.25.ffn_down_exps.weight
    IQ4_NL
    [704, 2816, 128]
  • blk.25.ffn_gate.weight
    Q8_0
    [2816, 2112]
  • blk.25.ffn_gate_inp.scale
    F32
    [2816]
  • blk.25.ffn_gate_inp.weight
    F32
    [2816, 128]
  • blk.25.ffn_gate_up_exps.weight
    IQ3_S
    [2816, 1408, 128]
  • blk.25.ffn_norm.weight
    F32
    [2816]
  • blk.25.ffn_up.weight
    Q8_0
    [2816, 2112]
  • blk.25.layer_output_scale.weight
    F32
    [1]
  • blk.25.post_attention_norm.weight
    F32
    [2816]
  • blk.25.post_ffw_norm.weight
    F32
    [2816]
  • blk.25.post_ffw_norm_1.weight
    F32
    [2816]
  • blk.25.post_ffw_norm_2.weight
    F32
    [2816]
  • blk.25.pre_ffw_norm_2.weight
    F32
    [2816]
    • blk.26
  • blk.26.attn_k.weight
    Q8_0
    [2816, 2048]
  • blk.26.attn_k_norm.weight
    F32
    [256]
  • blk.26.attn_norm.weight
    F32
    [2816]
  • blk.26.attn_output.weight
    Q8_0
    [4096, 2816]
  • blk.26.attn_q.weight
    Q8_0
    [2816, 4096]
  • blk.26.attn_q_norm.weight
    F32
    [256]
  • blk.26.attn_v.weight
    Q8_0
    [2816, 2048]
  • blk.26.ffn_down.weight
    Q8_0
    [2112, 2816]
  • blk.26.ffn_down_exps.scale
    F32
    [128]
  • blk.26.ffn_down_exps.weight
    IQ4_NL
    [704, 2816, 128]
  • blk.26.ffn_gate.weight
    Q8_0
    [2816, 2112]
  • blk.26.ffn_gate_inp.scale
    F32
    [2816]
  • blk.26.ffn_gate_inp.weight
    F32
    [2816, 128]
  • blk.26.ffn_gate_up_exps.weight
    IQ3_S
    [2816, 1408, 128]
  • blk.26.ffn_norm.weight
    F32
    [2816]
  • blk.26.ffn_up.weight
    Q8_0
    [2816, 2112]
  • blk.26.layer_output_scale.weight
    F32
    [1]
  • blk.26.post_attention_norm.weight
    F32
    [2816]
  • blk.26.post_ffw_norm.weight
    F32
    [2816]
  • blk.26.post_ffw_norm_1.weight
    F32
    [2816]
  • blk.26.post_ffw_norm_2.weight
    F32
    [2816]
  • blk.26.pre_ffw_norm_2.weight
    F32
    [2816]
    • blk.27
  • blk.27.attn_k.weight
    Q8_0
    [2816, 2048]
  • blk.27.attn_k_norm.weight
    F32
    [256]
  • blk.27.attn_norm.weight
    F32
    [2816]
  • blk.27.attn_output.weight
    Q8_0
    [4096, 2816]
  • blk.27.attn_q.weight
    Q8_0
    [2816, 4096]
  • blk.27.attn_q_norm.weight
    F32
    [256]
  • blk.27.attn_v.weight
    Q8_0
    [2816, 2048]
  • blk.27.ffn_down.weight
    Q8_0
    [2112, 2816]
  • blk.27.ffn_down_exps.scale
    F32
    [128]
  • blk.27.ffn_down_exps.weight
    IQ4_NL
    [704, 2816, 128]
  • blk.27.ffn_gate.weight
    Q8_0
    [2816, 2112]
  • blk.27.ffn_gate_inp.scale
    F32
    [2816]
  • blk.27.ffn_gate_inp.weight
    F32
    [2816, 128]
  • blk.27.ffn_gate_up_exps.weight
    IQ3_S
    [2816, 1408, 128]
  • blk.27.ffn_norm.weight
    F32
    [2816]
  • blk.27.ffn_up.weight
    Q8_0
    [2816, 2112]
  • blk.27.layer_output_scale.weight
    F32
    [1]
  • blk.27.post_attention_norm.weight
    F32
    [2816]
  • blk.27.post_ffw_norm.weight
    F32
    [2816]
  • blk.27.post_ffw_norm_1.weight
    F32
    [2816]
  • blk.27.post_ffw_norm_2.weight
    F32
    [2816]
  • blk.27.pre_ffw_norm_2.weight
    F32
    [2816]
    • blk.28
  • blk.28.attn_k.weight
    Q8_0
    [2816, 2048]
  • blk.28.attn_k_norm.weight
    F32
    [256]
  • blk.28.attn_norm.weight
    F32
    [2816]
  • blk.28.attn_output.weight
    Q8_0
    [4096, 2816]
  • blk.28.attn_q.weight
    Q8_0
    [2816, 4096]
  • blk.28.attn_q_norm.weight
    F32
    [256]
  • blk.28.attn_v.weight
    Q8_0
    [2816, 2048]
  • blk.28.ffn_down.weight
    Q8_0
    [2112, 2816]
  • blk.28.ffn_down_exps.scale
    F32
    [128]
  • blk.28.ffn_down_exps.weight
    IQ4_NL
    [704, 2816, 128]
  • blk.28.ffn_gate.weight
    Q8_0
    [2816, 2112]
  • blk.28.ffn_gate_inp.scale
    F32
    [2816]
  • blk.28.ffn_gate_inp.weight
    F32
    [2816, 128]
  • blk.28.ffn_gate_up_exps.weight
    IQ3_S
    [2816, 1408, 128]
  • blk.28.ffn_norm.weight
    F32
    [2816]
  • blk.28.ffn_up.weight
    Q8_0
    [2816, 2112]
  • blk.28.layer_output_scale.weight
    F32
    [1]
  • blk.28.post_attention_norm.weight
    F32
    [2816]
  • blk.28.post_ffw_norm.weight
    F32
    [2816]
  • blk.28.post_ffw_norm_1.weight
    F32
    [2816]
  • blk.28.post_ffw_norm_2.weight
    F32
    [2816]
  • blk.28.pre_ffw_norm_2.weight
    F32
    [2816]
    • blk.29
  • blk.29.attn_k.weight
    Q8_0
    [2816, 1024]
  • blk.29.attn_k_norm.weight
    F32
    [512]
  • blk.29.attn_norm.weight
    F32
    [2816]
  • blk.29.attn_output.weight
    Q8_0
    [8192, 2816]
  • blk.29.attn_q.weight
    Q8_0
    [2816, 8192]
  • blk.29.attn_q_norm.weight
    F32
    [512]
  • blk.29.ffn_down.weight
    Q8_0
    [2112, 2816]
  • blk.29.ffn_down_exps.scale
    F32
    [128]
  • blk.29.ffn_down_exps.weight
    IQ4_NL
    [704, 2816, 128]
  • blk.29.ffn_gate.weight
    Q8_0
    [2816, 2112]
  • blk.29.ffn_gate_inp.scale
    F32
    [2816]
  • blk.29.ffn_gate_inp.weight
    F32
    [2816, 128]
  • blk.29.ffn_gate_up_exps.weight
    IQ3_S
    [2816, 1408, 128]
  • blk.29.ffn_norm.weight
    F32
    [2816]
  • blk.29.ffn_up.weight
    Q8_0
    [2816, 2112]
  • blk.29.layer_output_scale.weight
    F32
    [1]
  • blk.29.post_attention_norm.weight
    F32
    [2816]
  • blk.29.post_ffw_norm.weight
    F32
    [2816]
  • blk.29.post_ffw_norm_1.weight
    F32
    [2816]
  • blk.29.post_ffw_norm_2.weight
    F32
    [2816]
  • blk.29.pre_ffw_norm_2.weight
    F32
    [2816]
  • rope_freqs.weight
    F32
    [256]
  • output_norm.weight
    F32
    [2816]