lfm2:24b-q8_0

13 hours ago

LFM2 is a family of hybrid models designed for on-device deployment. LFM2-24B-A2B is the largest model in the family, scaling the architecture to 24 billion parameters while keeping inference efficient.

tools 24b
... /
model
d11ac8db9890 · 25GB
    Metadata
  • general.architecture
    lfm2moe
  • general.file_type
    Q8_0
  • lfm2moe.attention.head_count
    32
  • lfm2moe.attention.head_count_kv
    [0, 0, 8, 0, 0, ...]
  • lfm2moe.attention.key_length
    64
  • lfm2moe.attention.layer_norm_rms_epsilon
    1e-05
  • lfm2moe.attention.value_length
    64
  • lfm2moe.block_count
    40
  • lfm2moe.context_length
    32768
  • lfm2moe.embedding_length
    2048
  • lfm2moe.expert_count
    64
  • lfm2moe.expert_feed_forward_length
    1536
  • lfm2moe.expert_gating_func
    2
  • lfm2moe.expert_used_count
    4
  • lfm2moe.feed_forward_length
    11776
  • lfm2moe.leading_dense_block_count
    2
  • lfm2moe.rope.freq_base
    1e+06
  • lfm2moe.shortconv.l_cache
    3
  • lfm2moe.vocab_size
    65536
  • tokenizer.ggml.add_bos_token
    false
  • tokenizer.ggml.add_eos_token
    false
  • tokenizer.ggml.add_padding_token
    false
  • tokenizer.ggml.bos_token_id
    1
  • tokenizer.ggml.eos_token_id
    7
  • tokenizer.ggml.merges
    [Ċ Ċ, Ċ ĊĊ, ĊĊ Ċ, Ċ ĊĊĊ, ĊĊ ĊĊ, ...]
  • tokenizer.ggml.model
    gpt2
  • tokenizer.ggml.padding_token_id
    0
  • tokenizer.ggml.pre
    lfm2
  • tokenizer.ggml.scores
    [0, 1, 2, 3, 4, ...]
  • tokenizer.ggml.token_type
    [3, 3, 3, 3, 3, ...]
  • tokenizer.ggml.tokens
    [<|pad|>, <|startoftext|>, <|endoftext|>, <|fim_pre|>, <|fim_mid|>, ...]
    • Tensor
  • token_embd.weight
    Q8_0
    [2048, 65536]
    • blk.0
  • blk.0.attn_norm.weight
    F32
    [2048]
  • blk.0.ffn_down.weight
    Q8_0
    [11776, 2048]
  • blk.0.ffn_gate.weight
    Q8_0
    [2048, 11776]
  • blk.0.ffn_norm.weight
    F32
    [2048]
  • blk.0.ffn_up.weight
    Q8_0
    [2048, 11776]
  • blk.0.shortconv.conv.weight
    F32
    [3, 2048]
  • blk.0.shortconv.in_proj.weight
    Q8_0
    [2048, 6144]
  • blk.0.shortconv.out_proj.weight
    Q8_0
    [2048, 2048]
    • blk.1
  • blk.1.attn_norm.weight
    F32
    [2048]
  • blk.1.ffn_down.weight
    Q8_0
    [11776, 2048]
  • blk.1.ffn_gate.weight
    Q8_0
    [2048, 11776]
  • blk.1.ffn_norm.weight
    F32
    [2048]
  • blk.1.ffn_up.weight
    Q8_0
    [2048, 11776]
  • blk.1.shortconv.conv.weight
    F32
    [3, 2048]
  • blk.1.shortconv.in_proj.weight
    Q8_0
    [2048, 6144]
  • blk.1.shortconv.out_proj.weight
    Q8_0
    [2048, 2048]
    • blk.2
  • blk.2.attn_k.weight
    Q8_0
    [2048, 512]
  • blk.2.attn_k_norm.weight
    F32
    [64]
  • blk.2.attn_norm.weight
    F32
    [2048]
  • blk.2.attn_output.weight
    Q8_0
    [2048, 2048]
  • blk.2.attn_q.weight
    Q8_0
    [2048, 2048]
  • blk.2.attn_q_norm.weight
    F32
    [64]
  • blk.2.attn_v.weight
    Q8_0
    [2048, 512]
  • blk.2.exp_probs_b.bias
    F32
    [64]
  • blk.2.ffn_down_exps.weight
    Q8_0
    [1536, 2048, 64]
  • blk.2.ffn_gate_exps.weight
    Q8_0
    [2048, 1536, 64]
  • blk.2.ffn_gate_inp.weight
    F32
    [2048, 64]
  • blk.2.ffn_norm.weight
    F32
    [2048]
  • blk.2.ffn_up_exps.weight
    Q8_0
    [2048, 1536, 64]
    • blk.3
  • blk.3.attn_norm.weight
    F32
    [2048]
  • blk.3.exp_probs_b.bias
    F32
    [64]
  • blk.3.ffn_down_exps.weight
    Q8_0
    [1536, 2048, 64]
  • blk.3.ffn_gate_exps.weight
    Q8_0
    [2048, 1536, 64]
  • blk.3.ffn_gate_inp.weight
    F32
    [2048, 64]
  • blk.3.ffn_norm.weight
    F32
    [2048]
  • blk.3.ffn_up_exps.weight
    Q8_0
    [2048, 1536, 64]
  • blk.3.shortconv.conv.weight
    F32
    [3, 2048]
  • blk.3.shortconv.in_proj.weight
    Q8_0
    [2048, 6144]
  • blk.3.shortconv.out_proj.weight
    Q8_0
    [2048, 2048]
    • blk.4
  • blk.4.attn_norm.weight
    F32
    [2048]
  • blk.4.exp_probs_b.bias
    F32
    [64]
  • blk.4.ffn_down_exps.weight
    Q8_0
    [1536, 2048, 64]
  • blk.4.ffn_gate_exps.weight
    Q8_0
    [2048, 1536, 64]
  • blk.4.ffn_gate_inp.weight
    F32
    [2048, 64]
  • blk.4.ffn_norm.weight
    F32
    [2048]
  • blk.4.ffn_up_exps.weight
    Q8_0
    [2048, 1536, 64]
  • blk.4.shortconv.conv.weight
    F32
    [3, 2048]
  • blk.4.shortconv.in_proj.weight
    Q8_0
    [2048, 6144]
  • blk.4.shortconv.out_proj.weight
    Q8_0
    [2048, 2048]
    • blk.5
  • blk.5.attn_norm.weight
    F32
    [2048]
  • blk.5.exp_probs_b.bias
    F32
    [64]
  • blk.5.ffn_down_exps.weight
    Q8_0
    [1536, 2048, 64]
  • blk.5.ffn_gate_exps.weight
    Q8_0
    [2048, 1536, 64]
  • blk.5.ffn_gate_inp.weight
    F32
    [2048, 64]
  • blk.5.ffn_norm.weight
    F32
    [2048]
  • blk.5.ffn_up_exps.weight
    Q8_0
    [2048, 1536, 64]
  • blk.5.shortconv.conv.weight
    F32
    [3, 2048]
  • blk.5.shortconv.in_proj.weight
    Q8_0
    [2048, 6144]
  • blk.5.shortconv.out_proj.weight
    Q8_0
    [2048, 2048]
    • blk.6
  • blk.6.attn_k.weight
    Q8_0
    [2048, 512]
  • blk.6.attn_k_norm.weight
    F32
    [64]
  • blk.6.attn_norm.weight
    F32
    [2048]
  • blk.6.attn_output.weight
    Q8_0
    [2048, 2048]
  • blk.6.attn_q.weight
    Q8_0
    [2048, 2048]
  • blk.6.attn_q_norm.weight
    F32
    [64]
  • blk.6.attn_v.weight
    Q8_0
    [2048, 512]
  • blk.6.exp_probs_b.bias
    F32
    [64]
  • blk.6.ffn_down_exps.weight
    Q8_0
    [1536, 2048, 64]
  • blk.6.ffn_gate_exps.weight
    Q8_0
    [2048, 1536, 64]
  • blk.6.ffn_gate_inp.weight
    F32
    [2048, 64]
  • blk.6.ffn_norm.weight
    F32
    [2048]
  • blk.6.ffn_up_exps.weight
    Q8_0
    [2048, 1536, 64]
    • blk.7
  • blk.7.attn_norm.weight
    F32
    [2048]
  • blk.7.exp_probs_b.bias
    F32
    [64]
  • blk.7.ffn_down_exps.weight
    Q8_0
    [1536, 2048, 64]
  • blk.7.ffn_gate_exps.weight
    Q8_0
    [2048, 1536, 64]
  • blk.7.ffn_gate_inp.weight
    F32
    [2048, 64]
  • blk.7.ffn_norm.weight
    F32
    [2048]
  • blk.7.ffn_up_exps.weight
    Q8_0
    [2048, 1536, 64]
  • blk.7.shortconv.conv.weight
    F32
    [3, 2048]
  • blk.7.shortconv.in_proj.weight
    Q8_0
    [2048, 6144]
  • blk.7.shortconv.out_proj.weight
    Q8_0
    [2048, 2048]
    • blk.8
  • blk.8.attn_norm.weight
    F32
    [2048]
  • blk.8.exp_probs_b.bias
    F32
    [64]
  • blk.8.ffn_down_exps.weight
    Q8_0
    [1536, 2048, 64]
  • blk.8.ffn_gate_exps.weight
    Q8_0
    [2048, 1536, 64]
  • blk.8.ffn_gate_inp.weight
    F32
    [2048, 64]
  • blk.8.ffn_norm.weight
    F32
    [2048]
  • blk.8.ffn_up_exps.weight
    Q8_0
    [2048, 1536, 64]
  • blk.8.shortconv.conv.weight
    F32
    [3, 2048]
  • blk.8.shortconv.in_proj.weight
    Q8_0
    [2048, 6144]
  • blk.8.shortconv.out_proj.weight
    Q8_0
    [2048, 2048]
    • blk.9
  • blk.9.attn_norm.weight
    F32
    [2048]
  • blk.9.exp_probs_b.bias
    F32
    [64]
  • blk.9.ffn_down_exps.weight
    Q8_0
    [1536, 2048, 64]
  • blk.9.ffn_gate_exps.weight
    Q8_0
    [2048, 1536, 64]
  • blk.9.ffn_gate_inp.weight
    F32
    [2048, 64]
  • blk.9.ffn_norm.weight
    F32
    [2048]
  • blk.9.ffn_up_exps.weight
    Q8_0
    [2048, 1536, 64]
  • blk.9.shortconv.conv.weight
    F32
    [3, 2048]
  • blk.9.shortconv.in_proj.weight
    Q8_0
    [2048, 6144]
  • blk.9.shortconv.out_proj.weight
    Q8_0
    [2048, 2048]
    • blk.10
  • blk.10.attn_k.weight
    Q8_0
    [2048, 512]
  • blk.10.attn_k_norm.weight
    F32
    [64]
  • blk.10.attn_norm.weight
    F32
    [2048]
  • blk.10.attn_output.weight
    Q8_0
    [2048, 2048]
  • blk.10.attn_q.weight
    Q8_0
    [2048, 2048]
  • blk.10.attn_q_norm.weight
    F32
    [64]
  • blk.10.attn_v.weight
    Q8_0
    [2048, 512]
  • blk.10.exp_probs_b.bias
    F32
    [64]
  • blk.10.ffn_down_exps.weight
    Q8_0
    [1536, 2048, 64]
  • blk.10.ffn_gate_exps.weight
    Q8_0
    [2048, 1536, 64]
  • blk.10.ffn_gate_inp.weight
    F32
    [2048, 64]
  • blk.10.ffn_norm.weight
    F32
    [2048]
  • blk.10.ffn_up_exps.weight
    Q8_0
    [2048, 1536, 64]
    • blk.11
  • blk.11.attn_norm.weight
    F32
    [2048]
  • blk.11.exp_probs_b.bias
    F32
    [64]
  • blk.11.ffn_down_exps.weight
    Q8_0
    [1536, 2048, 64]
  • blk.11.ffn_gate_exps.weight
    Q8_0
    [2048, 1536, 64]
  • blk.11.ffn_gate_inp.weight
    F32
    [2048, 64]
  • blk.11.ffn_norm.weight
    F32
    [2048]
  • blk.11.ffn_up_exps.weight
    Q8_0
    [2048, 1536, 64]
  • blk.11.shortconv.conv.weight
    F32
    [3, 2048]
  • blk.11.shortconv.in_proj.weight
    Q8_0
    [2048, 6144]
  • blk.11.shortconv.out_proj.weight
    Q8_0
    [2048, 2048]
    • blk.12
  • blk.12.attn_norm.weight
    F32
    [2048]
  • blk.12.exp_probs_b.bias
    F32
    [64]
  • blk.12.ffn_down_exps.weight
    Q8_0
    [1536, 2048, 64]
  • blk.12.ffn_gate_exps.weight
    Q8_0
    [2048, 1536, 64]
  • blk.12.ffn_gate_inp.weight
    F32
    [2048, 64]
  • blk.12.ffn_norm.weight
    F32
    [2048]
  • blk.12.ffn_up_exps.weight
    Q8_0
    [2048, 1536, 64]
  • blk.12.shortconv.conv.weight
    F32
    [3, 2048]
  • blk.12.shortconv.in_proj.weight
    Q8_0
    [2048, 6144]
  • blk.12.shortconv.out_proj.weight
    Q8_0
    [2048, 2048]
    • blk.13
  • blk.13.attn_norm.weight
    F32
    [2048]
  • blk.13.exp_probs_b.bias
    F32
    [64]
  • blk.13.ffn_down_exps.weight
    Q8_0
    [1536, 2048, 64]
  • blk.13.ffn_gate_exps.weight
    Q8_0
    [2048, 1536, 64]
  • blk.13.ffn_gate_inp.weight
    F32
    [2048, 64]
  • blk.13.ffn_norm.weight
    F32
    [2048]
  • blk.13.ffn_up_exps.weight
    Q8_0
    [2048, 1536, 64]
  • blk.13.shortconv.conv.weight
    F32
    [3, 2048]
  • blk.13.shortconv.in_proj.weight
    Q8_0
    [2048, 6144]
  • blk.13.shortconv.out_proj.weight
    Q8_0
    [2048, 2048]
    • blk.14
  • blk.14.attn_k.weight
    Q8_0
    [2048, 512]
  • blk.14.attn_k_norm.weight
    F32
    [64]
  • blk.14.attn_norm.weight
    F32
    [2048]
  • blk.14.attn_output.weight
    Q8_0
    [2048, 2048]
  • blk.14.attn_q.weight
    Q8_0
    [2048, 2048]
  • blk.14.attn_q_norm.weight
    F32
    [64]
  • blk.14.attn_v.weight
    Q8_0
    [2048, 512]
  • blk.14.exp_probs_b.bias
    F32
    [64]
  • blk.14.ffn_down_exps.weight
    Q8_0
    [1536, 2048, 64]
  • blk.14.ffn_gate_exps.weight
    Q8_0
    [2048, 1536, 64]
  • blk.14.ffn_gate_inp.weight
    F32
    [2048, 64]
  • blk.14.ffn_norm.weight
    F32
    [2048]
  • blk.14.ffn_up_exps.weight
    Q8_0
    [2048, 1536, 64]
    • blk.15
  • blk.15.attn_norm.weight
    F32
    [2048]
  • blk.15.exp_probs_b.bias
    F32
    [64]
  • blk.15.ffn_down_exps.weight
    Q8_0
    [1536, 2048, 64]
  • blk.15.ffn_gate_exps.weight
    Q8_0
    [2048, 1536, 64]
  • blk.15.ffn_gate_inp.weight
    F32
    [2048, 64]
  • blk.15.ffn_norm.weight
    F32
    [2048]
  • blk.15.ffn_up_exps.weight
    Q8_0
    [2048, 1536, 64]
  • blk.15.shortconv.conv.weight
    F32
    [3, 2048]
  • blk.15.shortconv.in_proj.weight
    Q8_0
    [2048, 6144]
  • blk.15.shortconv.out_proj.weight
    Q8_0
    [2048, 2048]
    • blk.16
  • blk.16.attn_norm.weight
    F32
    [2048]
  • blk.16.exp_probs_b.bias
    F32
    [64]
  • blk.16.ffn_down_exps.weight
    Q8_0
    [1536, 2048, 64]
  • blk.16.ffn_gate_exps.weight
    Q8_0
    [2048, 1536, 64]
  • blk.16.ffn_gate_inp.weight
    F32
    [2048, 64]
  • blk.16.ffn_norm.weight
    F32
    [2048]
  • blk.16.ffn_up_exps.weight
    Q8_0
    [2048, 1536, 64]
  • blk.16.shortconv.conv.weight
    F32
    [3, 2048]
  • blk.16.shortconv.in_proj.weight
    Q8_0
    [2048, 6144]
  • blk.16.shortconv.out_proj.weight
    Q8_0
    [2048, 2048]
    • blk.17
  • blk.17.attn_norm.weight
    F32
    [2048]
  • blk.17.exp_probs_b.bias
    F32
    [64]
  • blk.17.ffn_down_exps.weight
    Q8_0
    [1536, 2048, 64]
  • blk.17.ffn_gate_exps.weight
    Q8_0
    [2048, 1536, 64]
  • blk.17.ffn_gate_inp.weight
    F32
    [2048, 64]
  • blk.17.ffn_norm.weight
    F32
    [2048]
  • blk.17.ffn_up_exps.weight
    Q8_0
    [2048, 1536, 64]
  • blk.17.shortconv.conv.weight
    F32
    [3, 2048]
  • blk.17.shortconv.in_proj.weight
    Q8_0
    [2048, 6144]
  • blk.17.shortconv.out_proj.weight
    Q8_0
    [2048, 2048]
    • blk.18
  • blk.18.attn_k.weight
    Q8_0
    [2048, 512]
  • blk.18.attn_k_norm.weight
    F32
    [64]
  • blk.18.attn_norm.weight
    F32
    [2048]
  • blk.18.attn_output.weight
    Q8_0
    [2048, 2048]
  • blk.18.attn_q.weight
    Q8_0
    [2048, 2048]
  • blk.18.attn_q_norm.weight
    F32
    [64]
  • blk.18.attn_v.weight
    Q8_0
    [2048, 512]
  • blk.18.exp_probs_b.bias
    F32
    [64]
  • blk.18.ffn_down_exps.weight
    Q8_0
    [1536, 2048, 64]
  • blk.18.ffn_gate_exps.weight
    Q8_0
    [2048, 1536, 64]
  • blk.18.ffn_gate_inp.weight
    F32
    [2048, 64]
  • blk.18.ffn_norm.weight
    F32
    [2048]
  • blk.18.ffn_up_exps.weight
    Q8_0
    [2048, 1536, 64]
    • blk.19
  • blk.19.attn_norm.weight
    F32
    [2048]
  • blk.19.exp_probs_b.bias
    F32
    [64]
  • blk.19.ffn_down_exps.weight
    Q8_0
    [1536, 2048, 64]
  • blk.19.ffn_gate_exps.weight
    Q8_0
    [2048, 1536, 64]
  • blk.19.ffn_gate_inp.weight
    F32
    [2048, 64]
  • blk.19.ffn_norm.weight
    F32
    [2048]
  • blk.19.ffn_up_exps.weight
    Q8_0
    [2048, 1536, 64]
  • blk.19.shortconv.conv.weight
    F32
    [3, 2048]
  • blk.19.shortconv.in_proj.weight
    Q8_0
    [2048, 6144]
  • blk.19.shortconv.out_proj.weight
    Q8_0
    [2048, 2048]
    • blk.20
  • blk.20.attn_norm.weight
    F32
    [2048]
  • blk.20.exp_probs_b.bias
    F32
    [64]
  • blk.20.ffn_down_exps.weight
    Q8_0
    [1536, 2048, 64]
  • blk.20.ffn_gate_exps.weight
    Q8_0
    [2048, 1536, 64]
  • blk.20.ffn_gate_inp.weight
    F32
    [2048, 64]
  • blk.20.ffn_norm.weight
    F32
    [2048]
  • blk.20.ffn_up_exps.weight
    Q8_0
    [2048, 1536, 64]
  • blk.20.shortconv.conv.weight
    F32
    [3, 2048]
  • blk.20.shortconv.in_proj.weight
    Q8_0
    [2048, 6144]
  • blk.20.shortconv.out_proj.weight
    Q8_0
    [2048, 2048]
    • blk.21
  • blk.21.attn_norm.weight
    F32
    [2048]
  • blk.21.exp_probs_b.bias
    F32
    [64]
  • blk.21.ffn_down_exps.weight
    Q8_0
    [1536, 2048, 64]
  • blk.21.ffn_gate_exps.weight
    Q8_0
    [2048, 1536, 64]
  • blk.21.ffn_gate_inp.weight
    F32
    [2048, 64]
  • blk.21.ffn_norm.weight
    F32
    [2048]
  • blk.21.ffn_up_exps.weight
    Q8_0
    [2048, 1536, 64]
  • blk.21.shortconv.conv.weight
    F32
    [3, 2048]
  • blk.21.shortconv.in_proj.weight
    Q8_0
    [2048, 6144]
  • blk.21.shortconv.out_proj.weight
    Q8_0
    [2048, 2048]
    • blk.22
  • blk.22.attn_k.weight
    Q8_0
    [2048, 512]
  • blk.22.attn_k_norm.weight
    F32
    [64]
  • blk.22.attn_norm.weight
    F32
    [2048]
  • blk.22.attn_output.weight
    Q8_0
    [2048, 2048]
  • blk.22.attn_q.weight
    Q8_0
    [2048, 2048]
  • blk.22.attn_q_norm.weight
    F32
    [64]
  • blk.22.attn_v.weight
    Q8_0
    [2048, 512]
  • blk.22.exp_probs_b.bias
    F32
    [64]
  • blk.22.ffn_down_exps.weight
    Q8_0
    [1536, 2048, 64]
  • blk.22.ffn_gate_exps.weight
    Q8_0
    [2048, 1536, 64]
  • blk.22.ffn_gate_inp.weight
    F32
    [2048, 64]
  • blk.22.ffn_norm.weight
    F32
    [2048]
  • blk.22.ffn_up_exps.weight
    Q8_0
    [2048, 1536, 64]
    • blk.23
  • blk.23.attn_norm.weight
    F32
    [2048]
  • blk.23.exp_probs_b.bias
    F32
    [64]
  • blk.23.ffn_down_exps.weight
    Q8_0
    [1536, 2048, 64]
  • blk.23.ffn_gate_exps.weight
    Q8_0
    [2048, 1536, 64]
  • blk.23.ffn_gate_inp.weight
    F32
    [2048, 64]
  • blk.23.ffn_norm.weight
    F32
    [2048]
  • blk.23.ffn_up_exps.weight
    Q8_0
    [2048, 1536, 64]
  • blk.23.shortconv.conv.weight
    F32
    [3, 2048]
  • blk.23.shortconv.in_proj.weight
    Q8_0
    [2048, 6144]
  • blk.23.shortconv.out_proj.weight
    Q8_0
    [2048, 2048]
    • blk.24
  • blk.24.attn_norm.weight
    F32
    [2048]
  • blk.24.exp_probs_b.bias
    F32
    [64]
  • blk.24.ffn_down_exps.weight
    Q8_0
    [1536, 2048, 64]
  • blk.24.ffn_gate_exps.weight
    Q8_0
    [2048, 1536, 64]
  • blk.24.ffn_gate_inp.weight
    F32
    [2048, 64]
  • blk.24.ffn_norm.weight
    F32
    [2048]
  • blk.24.ffn_up_exps.weight
    Q8_0
    [2048, 1536, 64]
  • blk.24.shortconv.conv.weight
    F32
    [3, 2048]
  • blk.24.shortconv.in_proj.weight
    Q8_0
    [2048, 6144]
  • blk.24.shortconv.out_proj.weight
    Q8_0
    [2048, 2048]
    • blk.25
  • blk.25.attn_norm.weight
    F32
    [2048]
  • blk.25.exp_probs_b.bias
    F32
    [64]
  • blk.25.ffn_down_exps.weight
    Q8_0
    [1536, 2048, 64]
  • blk.25.ffn_gate_exps.weight
    Q8_0
    [2048, 1536, 64]
  • blk.25.ffn_gate_inp.weight
    F32
    [2048, 64]
  • blk.25.ffn_norm.weight
    F32
    [2048]
  • blk.25.ffn_up_exps.weight
    Q8_0
    [2048, 1536, 64]
  • blk.25.shortconv.conv.weight
    F32
    [3, 2048]
  • blk.25.shortconv.in_proj.weight
    Q8_0
    [2048, 6144]
  • blk.25.shortconv.out_proj.weight
    Q8_0
    [2048, 2048]
    • blk.26
  • blk.26.attn_k.weight
    Q8_0
    [2048, 512]
  • blk.26.attn_k_norm.weight
    F32
    [64]
  • blk.26.attn_norm.weight
    F32
    [2048]
  • blk.26.attn_output.weight
    Q8_0
    [2048, 2048]
  • blk.26.attn_q.weight
    Q8_0
    [2048, 2048]
  • blk.26.attn_q_norm.weight
    F32
    [64]
  • blk.26.attn_v.weight
    Q8_0
    [2048, 512]
  • blk.26.exp_probs_b.bias
    F32
    [64]
  • blk.26.ffn_down_exps.weight
    Q8_0
    [1536, 2048, 64]
  • blk.26.ffn_gate_exps.weight
    Q8_0
    [2048, 1536, 64]
  • blk.26.ffn_gate_inp.weight
    F32
    [2048, 64]
  • blk.26.ffn_norm.weight
    F32
    [2048]
  • blk.26.ffn_up_exps.weight
    Q8_0
    [2048, 1536, 64]
    • blk.27
  • blk.27.attn_norm.weight
    F32
    [2048]
  • blk.27.exp_probs_b.bias
    F32
    [64]
  • blk.27.ffn_down_exps.weight
    Q8_0
    [1536, 2048, 64]
  • blk.27.ffn_gate_exps.weight
    Q8_0
    [2048, 1536, 64]
  • blk.27.ffn_gate_inp.weight
    F32
    [2048, 64]
  • blk.27.ffn_norm.weight
    F32
    [2048]
  • blk.27.ffn_up_exps.weight
    Q8_0
    [2048, 1536, 64]
  • blk.27.shortconv.conv.weight
    F32
    [3, 2048]
  • blk.27.shortconv.in_proj.weight
    Q8_0
    [2048, 6144]
  • blk.27.shortconv.out_proj.weight
    Q8_0
    [2048, 2048]
    • blk.28
  • blk.28.attn_norm.weight
    F32
    [2048]
  • blk.28.exp_probs_b.bias
    F32
    [64]
  • blk.28.ffn_down_exps.weight
    Q8_0
    [1536, 2048, 64]
  • blk.28.ffn_gate_exps.weight
    Q8_0
    [2048, 1536, 64]
  • blk.28.ffn_gate_inp.weight
    F32
    [2048, 64]
  • blk.28.ffn_norm.weight
    F32
    [2048]
  • blk.28.ffn_up_exps.weight
    Q8_0
    [2048, 1536, 64]
  • blk.28.shortconv.conv.weight
    F32
    [3, 2048]
  • blk.28.shortconv.in_proj.weight
    Q8_0
    [2048, 6144]
  • blk.28.shortconv.out_proj.weight
    Q8_0
    [2048, 2048]
    • blk.29
  • blk.29.attn_norm.weight
    F32
    [2048]
  • blk.29.exp_probs_b.bias
    F32
    [64]
  • blk.29.ffn_down_exps.weight
    Q8_0
    [1536, 2048, 64]
  • blk.29.ffn_gate_exps.weight
    Q8_0
    [2048, 1536, 64]
  • blk.29.ffn_gate_inp.weight
    F32
    [2048, 64]
  • blk.29.ffn_norm.weight
    F32
    [2048]
  • blk.29.ffn_up_exps.weight
    Q8_0
    [2048, 1536, 64]
  • blk.29.shortconv.conv.weight
    F32
    [3, 2048]
  • blk.29.shortconv.in_proj.weight
    Q8_0
    [2048, 6144]
  • blk.29.shortconv.out_proj.weight
    Q8_0
    [2048, 2048]
    • blk.30
  • blk.30.attn_k.weight
    Q8_0
    [2048, 512]
  • blk.30.attn_k_norm.weight
    F32
    [64]
  • blk.30.attn_norm.weight
    F32
    [2048]
  • blk.30.attn_output.weight
    Q8_0
    [2048, 2048]
  • blk.30.attn_q.weight
    Q8_0
    [2048, 2048]
  • blk.30.attn_q_norm.weight
    F32
    [64]
  • blk.30.attn_v.weight
    Q8_0
    [2048, 512]
  • blk.30.exp_probs_b.bias
    F32
    [64]
  • blk.30.ffn_down_exps.weight
    Q8_0
    [1536, 2048, 64]
  • blk.30.ffn_gate_exps.weight
    Q8_0
    [2048, 1536, 64]
  • blk.30.ffn_gate_inp.weight
    F32
    [2048, 64]
  • blk.30.ffn_norm.weight
    F32
    [2048]
  • blk.30.ffn_up_exps.weight
    Q8_0
    [2048, 1536, 64]
    • blk.31
  • blk.31.attn_norm.weight
    F32
    [2048]
  • blk.31.exp_probs_b.bias
    F32
    [64]
  • blk.31.ffn_down_exps.weight
    Q8_0
    [1536, 2048, 64]
  • blk.31.ffn_gate_exps.weight
    Q8_0
    [2048, 1536, 64]
  • blk.31.ffn_gate_inp.weight
    F32
    [2048, 64]
  • blk.31.ffn_norm.weight
    F32
    [2048]
  • blk.31.ffn_up_exps.weight
    Q8_0
    [2048, 1536, 64]
  • blk.31.shortconv.conv.weight
    F32
    [3, 2048]
  • blk.31.shortconv.in_proj.weight
    Q8_0
    [2048, 6144]
  • blk.31.shortconv.out_proj.weight
    Q8_0
    [2048, 2048]
    • blk.32
  • blk.32.attn_norm.weight
    F32
    [2048]
  • blk.32.exp_probs_b.bias
    F32
    [64]
  • blk.32.ffn_down_exps.weight
    Q8_0
    [1536, 2048, 64]
  • blk.32.ffn_gate_exps.weight
    Q8_0
    [2048, 1536, 64]
  • blk.32.ffn_gate_inp.weight
    F32
    [2048, 64]
  • blk.32.ffn_norm.weight
    F32
    [2048]
  • blk.32.ffn_up_exps.weight
    Q8_0
    [2048, 1536, 64]
  • blk.32.shortconv.conv.weight
    F32
    [3, 2048]
  • blk.32.shortconv.in_proj.weight
    Q8_0
    [2048, 6144]
  • blk.32.shortconv.out_proj.weight
    Q8_0
    [2048, 2048]
    • blk.33
  • blk.33.attn_norm.weight
    F32
    [2048]
  • blk.33.exp_probs_b.bias
    F32
    [64]
  • blk.33.ffn_down_exps.weight
    Q8_0
    [1536, 2048, 64]
  • blk.33.ffn_gate_exps.weight
    Q8_0
    [2048, 1536, 64]
  • blk.33.ffn_gate_inp.weight
    F32
    [2048, 64]
  • blk.33.ffn_norm.weight
    F32
    [2048]
  • blk.33.ffn_up_exps.weight
    Q8_0
    [2048, 1536, 64]
  • blk.33.shortconv.conv.weight
    F32
    [3, 2048]
  • blk.33.shortconv.in_proj.weight
    Q8_0
    [2048, 6144]
  • blk.33.shortconv.out_proj.weight
    Q8_0
    [2048, 2048]
    • blk.34
  • blk.34.attn_k.weight
    Q8_0
    [2048, 512]
  • blk.34.attn_k_norm.weight
    F32
    [64]
  • blk.34.attn_norm.weight
    F32
    [2048]
  • blk.34.attn_output.weight
    Q8_0
    [2048, 2048]
  • blk.34.attn_q.weight
    Q8_0
    [2048, 2048]
  • blk.34.attn_q_norm.weight
    F32
    [64]
  • blk.34.attn_v.weight
    Q8_0
    [2048, 512]
  • blk.34.exp_probs_b.bias
    F32
    [64]
  • blk.34.ffn_down_exps.weight
    Q8_0
    [1536, 2048, 64]
  • blk.34.ffn_gate_exps.weight
    Q8_0
    [2048, 1536, 64]
  • blk.34.ffn_gate_inp.weight
    F32
    [2048, 64]
  • blk.34.ffn_norm.weight
    F32
    [2048]
  • blk.34.ffn_up_exps.weight
    Q8_0
    [2048, 1536, 64]
    • blk.35
  • blk.35.attn_norm.weight
    F32
    [2048]
  • blk.35.exp_probs_b.bias
    F32
    [64]
  • blk.35.ffn_down_exps.weight
    Q8_0
    [1536, 2048, 64]
  • blk.35.ffn_gate_exps.weight
    Q8_0
    [2048, 1536, 64]
  • blk.35.ffn_gate_inp.weight
    F32
    [2048, 64]
  • blk.35.ffn_norm.weight
    F32
    [2048]
  • blk.35.ffn_up_exps.weight
    Q8_0
    [2048, 1536, 64]
  • blk.35.shortconv.conv.weight
    F32
    [3, 2048]
  • blk.35.shortconv.in_proj.weight
    Q8_0
    [2048, 6144]
  • blk.35.shortconv.out_proj.weight
    Q8_0
    [2048, 2048]
    • blk.36
  • blk.36.attn_norm.weight
    F32
    [2048]
  • blk.36.exp_probs_b.bias
    F32
    [64]
  • blk.36.ffn_down_exps.weight
    Q8_0
    [1536, 2048, 64]
  • blk.36.ffn_gate_exps.weight
    Q8_0
    [2048, 1536, 64]
  • blk.36.ffn_gate_inp.weight
    F32
    [2048, 64]
  • blk.36.ffn_norm.weight
    F32
    [2048]
  • blk.36.ffn_up_exps.weight
    Q8_0
    [2048, 1536, 64]
  • blk.36.shortconv.conv.weight
    F32
    [3, 2048]
  • blk.36.shortconv.in_proj.weight
    Q8_0
    [2048, 6144]
  • blk.36.shortconv.out_proj.weight
    Q8_0
    [2048, 2048]
    • blk.37
  • blk.37.attn_norm.weight
    F32
    [2048]
  • blk.37.exp_probs_b.bias
    F32
    [64]
  • blk.37.ffn_down_exps.weight
    Q8_0
    [1536, 2048, 64]
  • blk.37.ffn_gate_exps.weight
    Q8_0
    [2048, 1536, 64]
  • blk.37.ffn_gate_inp.weight
    F32
    [2048, 64]
  • blk.37.ffn_norm.weight
    F32
    [2048]
  • blk.37.ffn_up_exps.weight
    Q8_0
    [2048, 1536, 64]
  • blk.37.shortconv.conv.weight
    F32
    [3, 2048]
  • blk.37.shortconv.in_proj.weight
    Q8_0
    [2048, 6144]
  • blk.37.shortconv.out_proj.weight
    Q8_0
    [2048, 2048]
    • blk.38
  • blk.38.attn_k.weight
    Q8_0
    [2048, 512]
  • blk.38.attn_k_norm.weight
    F32
    [64]
  • blk.38.attn_norm.weight
    F32
    [2048]
  • blk.38.attn_output.weight
    Q8_0
    [2048, 2048]
  • blk.38.attn_q.weight
    Q8_0
    [2048, 2048]
  • blk.38.attn_q_norm.weight
    F32
    [64]
  • blk.38.attn_v.weight
    Q8_0
    [2048, 512]
  • blk.38.exp_probs_b.bias
    F32
    [64]
  • blk.38.ffn_down_exps.weight
    Q8_0
    [1536, 2048, 64]
  • blk.38.ffn_gate_exps.weight
    Q8_0
    [2048, 1536, 64]
  • blk.38.ffn_gate_inp.weight
    F32
    [2048, 64]
  • blk.38.ffn_norm.weight
    F32
    [2048]
  • blk.38.ffn_up_exps.weight
    Q8_0
    [2048, 1536, 64]
    • blk.39
  • blk.39.attn_norm.weight
    F32
    [2048]
  • blk.39.exp_probs_b.bias
    F32
    [64]
  • blk.39.ffn_down_exps.weight
    Q8_0
    [1536, 2048, 64]
  • blk.39.ffn_gate_exps.weight
    Q8_0
    [2048, 1536, 64]
  • blk.39.ffn_gate_inp.weight
    F32
    [2048, 64]
  • blk.39.ffn_norm.weight
    F32
    [2048]
  • blk.39.ffn_up_exps.weight
    Q8_0
    [2048, 1536, 64]
  • blk.39.shortconv.conv.weight
    F32
    [3, 2048]
  • blk.39.shortconv.in_proj.weight
    Q8_0
    [2048, 6144]
  • blk.39.shortconv.out_proj.weight
    Q8_0
    [2048, 2048]
  • token_embd_norm.weight
    F32
    [2048]