MichelRosselli/
apertus:8b-instruct-2509-bf16

193 3 days ago

Apertus is a 70B and 8B parameter language model designed to push the boundaries of fully-open multilingual and transparent models.

3c304abf37a8 · 16GB
    Metadata
  • general.architecture
    apertus
  • general.file_type
    BF16
  • apertus.attention.head_count
    32
  • apertus.attention.head_count_kv
    8
  • apertus.attention.layer_norm_rms_epsilon
    1e-05
  • apertus.block_count
    32
  • apertus.context_length
    65536
  • apertus.embedding_length
    4096
  • apertus.feed_forward_length
    21504
  • apertus.rope.dimension_count
    128
  • apertus.rope.freq_base
    1.2e+07
  • apertus.vocab_size
    131072
  • tokenizer.ggml.add_bos_token
    true
  • tokenizer.ggml.add_eos_token
    false
  • tokenizer.ggml.add_sep_token
    false
  • tokenizer.ggml.add_space_prefix
    false
  • tokenizer.ggml.bos_token_id
    1
  • tokenizer.ggml.eos_token_id
    68
  • tokenizer.ggml.merges
    [Ġ Ġ, Ġ t, e r, i n, Ġ ĠĠĠ, ...]
  • tokenizer.ggml.model
    gpt2
  • tokenizer.ggml.padding_token_id
    3
  • tokenizer.ggml.pre
    tekken
  • tokenizer.ggml.token_type
    [3, 3, 3, 3, 3, ...]
  • tokenizer.ggml.tokens
    [<unk>, <s>, </s>, <pad>, [/INST], ...]
  • tokenizer.ggml.unknown_token_id
    0
  • xielu.alpha_n
    [40.75, 31.625, 22.875, 16.875, 10.5, ...]
  • xielu.alpha_p
    [166, 174, 128, 72.5, 80.5, ...]
  • xielu.beta
    [0.5, 0.5, 0.5, 0.5, 0.5, ...]
  • xielu.eps
    [-9.983778e-07, -9.983778e-07, -9.983778e-07, -9.983778e-07, -9.983778e-07, ...]
  • Tensor
  • token_embd.weight
    BF16
    [4096, 131072]
  • blk.0
  • blk.0.attn_k.weight
    BF16
    [4096, 1024]
  • blk.0.attn_k_norm.weight
    F32
    [128]
  • blk.0.attn_norm.weight
    F32
    [4096]
  • blk.0.attn_output.weight
    BF16
    [4096, 4096]
  • blk.0.attn_q.weight
    BF16
    [4096, 4096]
  • blk.0.attn_q_norm.weight
    F32
    [128]
  • blk.0.attn_v.weight
    BF16
    [4096, 1024]
  • blk.0.ffn_down.weight
    BF16
    [21504, 4096]
  • blk.0.ffn_norm.weight
    F32
    [4096]
  • blk.0.ffn_up.weight
    BF16
    [4096, 21504]
  • blk.1
  • blk.1.attn_k.weight
    BF16
    [4096, 1024]
  • blk.1.attn_k_norm.weight
    F32
    [128]
  • blk.1.attn_norm.weight
    F32
    [4096]
  • blk.1.attn_output.weight
    BF16
    [4096, 4096]
  • blk.1.attn_q.weight
    BF16
    [4096, 4096]
  • blk.1.attn_q_norm.weight
    F32
    [128]
  • blk.1.attn_v.weight
    BF16
    [4096, 1024]
  • blk.1.ffn_down.weight
    BF16
    [21504, 4096]
  • blk.1.ffn_norm.weight
    F32
    [4096]
  • blk.1.ffn_up.weight
    BF16
    [4096, 21504]
  • blk.2
  • blk.2.attn_k.weight
    BF16
    [4096, 1024]
  • blk.2.attn_k_norm.weight
    F32
    [128]
  • blk.2.attn_norm.weight
    F32
    [4096]
  • blk.2.attn_output.weight
    BF16
    [4096, 4096]
  • blk.2.attn_q.weight
    BF16
    [4096, 4096]
  • blk.2.attn_q_norm.weight
    F32
    [128]
  • blk.2.attn_v.weight
    BF16
    [4096, 1024]
  • blk.2.ffn_down.weight
    BF16
    [21504, 4096]
  • blk.2.ffn_norm.weight
    F32
    [4096]
  • blk.2.ffn_up.weight
    BF16
    [4096, 21504]
  • blk.3
  • blk.3.attn_k.weight
    BF16
    [4096, 1024]
  • blk.3.attn_k_norm.weight
    F32
    [128]
  • blk.3.attn_norm.weight
    F32
    [4096]
  • blk.3.attn_output.weight
    BF16
    [4096, 4096]
  • blk.3.attn_q.weight
    BF16
    [4096, 4096]
  • blk.3.attn_q_norm.weight
    F32
    [128]
  • blk.3.attn_v.weight
    BF16
    [4096, 1024]
  • blk.3.ffn_down.weight
    BF16
    [21504, 4096]
  • blk.3.ffn_norm.weight
    F32
    [4096]
  • blk.3.ffn_up.weight
    BF16
    [4096, 21504]
  • blk.4
  • blk.4.attn_k.weight
    BF16
    [4096, 1024]
  • blk.4.attn_k_norm.weight
    F32
    [128]
  • blk.4.attn_norm.weight
    F32
    [4096]
  • blk.4.attn_output.weight
    BF16
    [4096, 4096]
  • blk.4.attn_q.weight
    BF16
    [4096, 4096]
  • blk.4.attn_q_norm.weight
    F32
    [128]
  • blk.4.attn_v.weight
    BF16
    [4096, 1024]
  • blk.4.ffn_down.weight
    BF16
    [21504, 4096]
  • blk.4.ffn_norm.weight
    F32
    [4096]
  • blk.4.ffn_up.weight
    BF16
    [4096, 21504]
  • blk.5
  • blk.5.attn_k.weight
    BF16
    [4096, 1024]
  • blk.5.attn_k_norm.weight
    F32
    [128]
  • blk.5.attn_norm.weight
    F32
    [4096]
  • blk.5.attn_output.weight
    BF16
    [4096, 4096]
  • blk.5.attn_q.weight
    BF16
    [4096, 4096]
  • blk.5.attn_q_norm.weight
    F32
    [128]
  • blk.5.attn_v.weight
    BF16
    [4096, 1024]
  • blk.5.ffn_down.weight
    BF16
    [21504, 4096]
  • blk.5.ffn_norm.weight
    F32
    [4096]
  • blk.5.ffn_up.weight
    BF16
    [4096, 21504]
  • blk.6
  • blk.6.attn_k.weight
    BF16
    [4096, 1024]
  • blk.6.attn_k_norm.weight
    F32
    [128]
  • blk.6.attn_norm.weight
    F32
    [4096]
  • blk.6.attn_output.weight
    BF16
    [4096, 4096]
  • blk.6.attn_q.weight
    BF16
    [4096, 4096]
  • blk.6.attn_q_norm.weight
    F32
    [128]
  • blk.6.attn_v.weight
    BF16
    [4096, 1024]
  • blk.6.ffn_down.weight
    BF16
    [21504, 4096]
  • blk.6.ffn_norm.weight
    F32
    [4096]
  • blk.6.ffn_up.weight
    BF16
    [4096, 21504]
  • blk.7
  • blk.7.attn_k.weight
    BF16
    [4096, 1024]
  • blk.7.attn_k_norm.weight
    F32
    [128]
  • blk.7.attn_norm.weight
    F32
    [4096]
  • blk.7.attn_output.weight
    BF16
    [4096, 4096]
  • blk.7.attn_q.weight
    BF16
    [4096, 4096]
  • blk.7.attn_q_norm.weight
    F32
    [128]
  • blk.7.attn_v.weight
    BF16
    [4096, 1024]
  • blk.7.ffn_down.weight
    BF16
    [21504, 4096]
  • blk.7.ffn_norm.weight
    F32
    [4096]
  • blk.7.ffn_up.weight
    BF16
    [4096, 21504]
  • blk.8
  • blk.8.attn_k.weight
    BF16
    [4096, 1024]
  • blk.8.attn_k_norm.weight
    F32
    [128]
  • blk.8.attn_norm.weight
    F32
    [4096]
  • blk.8.attn_output.weight
    BF16
    [4096, 4096]
  • blk.8.attn_q.weight
    BF16
    [4096, 4096]
  • blk.8.attn_q_norm.weight
    F32
    [128]
  • blk.8.attn_v.weight
    BF16
    [4096, 1024]
  • blk.8.ffn_down.weight
    BF16
    [21504, 4096]
  • blk.8.ffn_norm.weight
    F32
    [4096]
  • blk.8.ffn_up.weight
    BF16
    [4096, 21504]
  • blk.9
  • blk.9.attn_k.weight
    BF16
    [4096, 1024]
  • blk.9.attn_k_norm.weight
    F32
    [128]
  • blk.9.attn_norm.weight
    F32
    [4096]
  • blk.9.attn_output.weight
    BF16
    [4096, 4096]
  • blk.9.attn_q.weight
    BF16
    [4096, 4096]
  • blk.9.attn_q_norm.weight
    F32
    [128]
  • blk.9.attn_v.weight
    BF16
    [4096, 1024]
  • blk.9.ffn_down.weight
    BF16
    [21504, 4096]
  • blk.9.ffn_norm.weight
    F32
    [4096]
  • blk.9.ffn_up.weight
    BF16
    [4096, 21504]
  • blk.10
  • blk.10.attn_k.weight
    BF16
    [4096, 1024]
  • blk.10.attn_k_norm.weight
    F32
    [128]
  • blk.10.attn_norm.weight
    F32
    [4096]
  • blk.10.attn_output.weight
    BF16
    [4096, 4096]
  • blk.10.attn_q.weight
    BF16
    [4096, 4096]
  • blk.10.attn_q_norm.weight
    F32
    [128]
  • blk.10.attn_v.weight
    BF16
    [4096, 1024]
  • blk.10.ffn_down.weight
    BF16
    [21504, 4096]
  • blk.10.ffn_norm.weight
    F32
    [4096]
  • blk.10.ffn_up.weight
    BF16
    [4096, 21504]
  • blk.11
  • blk.11.attn_k.weight
    BF16
    [4096, 1024]
  • blk.11.attn_k_norm.weight
    F32
    [128]
  • blk.11.attn_norm.weight
    F32
    [4096]
  • blk.11.attn_output.weight
    BF16
    [4096, 4096]
  • blk.11.attn_q.weight
    BF16
    [4096, 4096]
  • blk.11.attn_q_norm.weight
    F32
    [128]
  • blk.11.attn_v.weight
    BF16
    [4096, 1024]
  • blk.11.ffn_down.weight
    BF16
    [21504, 4096]
  • blk.11.ffn_norm.weight
    F32
    [4096]
  • blk.11.ffn_up.weight
    BF16
    [4096, 21504]
  • blk.12
  • blk.12.attn_k.weight
    BF16
    [4096, 1024]
  • blk.12.attn_k_norm.weight
    F32
    [128]
  • blk.12.attn_norm.weight
    F32
    [4096]
  • blk.12.attn_output.weight
    BF16
    [4096, 4096]
  • blk.12.attn_q.weight
    BF16
    [4096, 4096]
  • blk.12.attn_q_norm.weight
    F32
    [128]
  • blk.12.attn_v.weight
    BF16
    [4096, 1024]
  • blk.12.ffn_down.weight
    BF16
    [21504, 4096]
  • blk.12.ffn_norm.weight
    F32
    [4096]
  • blk.12.ffn_up.weight
    BF16
    [4096, 21504]
  • blk.13
  • blk.13.attn_k.weight
    BF16
    [4096, 1024]
  • blk.13.attn_k_norm.weight
    F32
    [128]
  • blk.13.attn_norm.weight
    F32
    [4096]
  • blk.13.attn_output.weight
    BF16
    [4096, 4096]
  • blk.13.attn_q.weight
    BF16
    [4096, 4096]
  • blk.13.attn_q_norm.weight
    F32
    [128]
  • blk.13.attn_v.weight
    BF16
    [4096, 1024]
  • blk.13.ffn_down.weight
    BF16
    [21504, 4096]
  • blk.13.ffn_norm.weight
    F32
    [4096]
  • blk.13.ffn_up.weight
    BF16
    [4096, 21504]
  • blk.14
  • blk.14.attn_k.weight
    BF16
    [4096, 1024]
  • blk.14.attn_k_norm.weight
    F32
    [128]
  • blk.14.attn_norm.weight
    F32
    [4096]
  • blk.14.attn_output.weight
    BF16
    [4096, 4096]
  • blk.14.attn_q.weight
    BF16
    [4096, 4096]
  • blk.14.attn_q_norm.weight
    F32
    [128]
  • blk.14.attn_v.weight
    BF16
    [4096, 1024]
  • blk.14.ffn_down.weight
    BF16
    [21504, 4096]
  • blk.14.ffn_norm.weight
    F32
    [4096]
  • blk.14.ffn_up.weight
    BF16
    [4096, 21504]
  • blk.15
  • blk.15.attn_k.weight
    BF16
    [4096, 1024]
  • blk.15.attn_k_norm.weight
    F32
    [128]
  • blk.15.attn_norm.weight
    F32
    [4096]
  • blk.15.attn_output.weight
    BF16
    [4096, 4096]
  • blk.15.attn_q.weight
    BF16
    [4096, 4096]
  • blk.15.attn_q_norm.weight
    F32
    [128]
  • blk.15.attn_v.weight
    BF16
    [4096, 1024]
  • blk.15.ffn_down.weight
    BF16
    [21504, 4096]
  • blk.15.ffn_norm.weight
    F32
    [4096]
  • blk.15.ffn_up.weight
    BF16
    [4096, 21504]
  • blk.16
  • blk.16.attn_k.weight
    BF16
    [4096, 1024]
  • blk.16.attn_k_norm.weight
    F32
    [128]
  • blk.16.attn_norm.weight
    F32
    [4096]
  • blk.16.attn_output.weight
    BF16
    [4096, 4096]
  • blk.16.attn_q.weight
    BF16
    [4096, 4096]
  • blk.16.attn_q_norm.weight
    F32
    [128]
  • blk.16.attn_v.weight
    BF16
    [4096, 1024]
  • blk.16.ffn_down.weight
    BF16
    [21504, 4096]
  • blk.16.ffn_norm.weight
    F32
    [4096]
  • blk.16.ffn_up.weight
    BF16
    [4096, 21504]
  • blk.17
  • blk.17.attn_k.weight
    BF16
    [4096, 1024]
  • blk.17.attn_k_norm.weight
    F32
    [128]
  • blk.17.attn_norm.weight
    F32
    [4096]
  • blk.17.attn_output.weight
    BF16
    [4096, 4096]
  • blk.17.attn_q.weight
    BF16
    [4096, 4096]
  • blk.17.attn_q_norm.weight
    F32
    [128]
  • blk.17.attn_v.weight
    BF16
    [4096, 1024]
  • blk.17.ffn_down.weight
    BF16
    [21504, 4096]
  • blk.17.ffn_norm.weight
    F32
    [4096]
  • blk.17.ffn_up.weight
    BF16
    [4096, 21504]
  • blk.18
  • blk.18.attn_k.weight
    BF16
    [4096, 1024]
  • blk.18.attn_k_norm.weight
    F32
    [128]
  • blk.18.attn_norm.weight
    F32
    [4096]
  • blk.18.attn_output.weight
    BF16
    [4096, 4096]
  • blk.18.attn_q.weight
    BF16
    [4096, 4096]
  • blk.18.attn_q_norm.weight
    F32
    [128]
  • blk.18.attn_v.weight
    BF16
    [4096, 1024]
  • blk.18.ffn_down.weight
    BF16
    [21504, 4096]
  • blk.18.ffn_norm.weight
    F32
    [4096]
  • blk.18.ffn_up.weight
    BF16
    [4096, 21504]
  • blk.19
  • blk.19.attn_k.weight
    BF16
    [4096, 1024]
  • blk.19.attn_k_norm.weight
    F32
    [128]
  • blk.19.attn_norm.weight
    F32
    [4096]
  • blk.19.attn_output.weight
    BF16
    [4096, 4096]
  • blk.19.attn_q.weight
    BF16
    [4096, 4096]
  • blk.19.attn_q_norm.weight
    F32
    [128]
  • blk.19.attn_v.weight
    BF16
    [4096, 1024]
  • blk.19.ffn_down.weight
    BF16
    [21504, 4096]
  • blk.19.ffn_norm.weight
    F32
    [4096]
  • blk.19.ffn_up.weight
    BF16
    [4096, 21504]
  • blk.20
  • blk.20.attn_k.weight
    BF16
    [4096, 1024]
  • blk.20.attn_k_norm.weight
    F32
    [128]
  • blk.20.attn_norm.weight
    F32
    [4096]
  • blk.20.attn_output.weight
    BF16
    [4096, 4096]
  • blk.20.attn_q.weight
    BF16
    [4096, 4096]
  • blk.20.attn_q_norm.weight
    F32
    [128]
  • blk.20.attn_v.weight
    BF16
    [4096, 1024]
  • blk.20.ffn_down.weight
    BF16
    [21504, 4096]
  • blk.20.ffn_norm.weight
    F32
    [4096]
  • blk.20.ffn_up.weight
    BF16
    [4096, 21504]
  • blk.21
  • blk.21.attn_k.weight
    BF16
    [4096, 1024]
  • blk.21.attn_k_norm.weight
    F32
    [128]
  • blk.21.attn_norm.weight
    F32
    [4096]
  • blk.21.attn_output.weight
    BF16
    [4096, 4096]
  • blk.21.attn_q.weight
    BF16
    [4096, 4096]
  • blk.21.attn_q_norm.weight
    F32
    [128]
  • blk.21.attn_v.weight
    BF16
    [4096, 1024]
  • blk.21.ffn_down.weight
    BF16
    [21504, 4096]
  • blk.21.ffn_norm.weight
    F32
    [4096]
  • blk.21.ffn_up.weight
    BF16
    [4096, 21504]
  • blk.22
  • blk.22.attn_k.weight
    BF16
    [4096, 1024]
  • blk.22.attn_k_norm.weight
    F32
    [128]
  • blk.22.attn_norm.weight
    F32
    [4096]
  • blk.22.attn_output.weight
    BF16
    [4096, 4096]
  • blk.22.attn_q.weight
    BF16
    [4096, 4096]
  • blk.22.attn_q_norm.weight
    F32
    [128]
  • blk.22.attn_v.weight
    BF16
    [4096, 1024]
  • blk.22.ffn_down.weight
    BF16
    [21504, 4096]
  • blk.22.ffn_norm.weight
    F32
    [4096]
  • blk.22.ffn_up.weight
    BF16
    [4096, 21504]
  • blk.23
  • blk.23.attn_k.weight
    BF16
    [4096, 1024]
  • blk.23.attn_k_norm.weight
    F32
    [128]
  • blk.23.attn_norm.weight
    F32
    [4096]
  • blk.23.attn_output.weight
    BF16
    [4096, 4096]
  • blk.23.attn_q.weight
    BF16
    [4096, 4096]
  • blk.23.attn_q_norm.weight
    F32
    [128]
  • blk.23.attn_v.weight
    BF16
    [4096, 1024]
  • blk.23.ffn_down.weight
    BF16
    [21504, 4096]
  • blk.23.ffn_norm.weight
    F32
    [4096]
  • blk.23.ffn_up.weight
    BF16
    [4096, 21504]
  • blk.24
  • blk.24.attn_k.weight
    BF16
    [4096, 1024]
  • blk.24.attn_k_norm.weight
    F32
    [128]
  • blk.24.attn_norm.weight
    F32
    [4096]
  • blk.24.attn_output.weight
    BF16
    [4096, 4096]
  • blk.24.attn_q.weight
    BF16
    [4096, 4096]
  • blk.24.attn_q_norm.weight
    F32
    [128]
  • blk.24.attn_v.weight
    BF16
    [4096, 1024]
  • blk.24.ffn_down.weight
    BF16
    [21504, 4096]
  • blk.24.ffn_norm.weight
    F32
    [4096]
  • blk.24.ffn_up.weight
    BF16
    [4096, 21504]
  • blk.25
  • blk.25.attn_k.weight
    BF16
    [4096, 1024]
  • blk.25.attn_k_norm.weight
    F32
    [128]
  • blk.25.attn_norm.weight
    F32
    [4096]
  • blk.25.attn_output.weight
    BF16
    [4096, 4096]
  • blk.25.attn_q.weight
    BF16
    [4096, 4096]
  • blk.25.attn_q_norm.weight
    F32
    [128]
  • blk.25.attn_v.weight
    BF16
    [4096, 1024]
  • blk.25.ffn_down.weight
    BF16
    [21504, 4096]
  • blk.25.ffn_norm.weight
    F32
    [4096]
  • blk.25.ffn_up.weight
    BF16
    [4096, 21504]
  • blk.26
  • blk.26.attn_k.weight
    BF16
    [4096, 1024]
  • blk.26.attn_k_norm.weight
    F32
    [128]
  • blk.26.attn_norm.weight
    F32
    [4096]
  • blk.26.attn_output.weight
    BF16
    [4096, 4096]
  • blk.26.attn_q.weight
    BF16
    [4096, 4096]
  • blk.26.attn_q_norm.weight
    F32
    [128]
  • blk.26.attn_v.weight
    BF16
    [4096, 1024]
  • blk.26.ffn_down.weight
    BF16
    [21504, 4096]
  • blk.26.ffn_norm.weight
    F32
    [4096]
  • blk.26.ffn_up.weight
    BF16
    [4096, 21504]
  • blk.27
  • blk.27.attn_k.weight
    BF16
    [4096, 1024]
  • blk.27.attn_k_norm.weight
    F32
    [128]
  • blk.27.attn_norm.weight
    F32
    [4096]
  • blk.27.attn_output.weight
    BF16
    [4096, 4096]
  • blk.27.attn_q.weight
    BF16
    [4096, 4096]
  • blk.27.attn_q_norm.weight
    F32
    [128]
  • blk.27.attn_v.weight
    BF16
    [4096, 1024]
  • blk.27.ffn_down.weight
    BF16
    [21504, 4096]
  • blk.27.ffn_norm.weight
    F32
    [4096]
  • blk.27.ffn_up.weight
    BF16
    [4096, 21504]
  • blk.28
  • blk.28.attn_k.weight
    BF16
    [4096, 1024]
  • blk.28.attn_k_norm.weight
    F32
    [128]
  • blk.28.attn_norm.weight
    F32
    [4096]
  • blk.28.attn_output.weight
    BF16
    [4096, 4096]
  • blk.28.attn_q.weight
    BF16
    [4096, 4096]
  • blk.28.attn_q_norm.weight
    F32
    [128]
  • blk.28.attn_v.weight
    BF16
    [4096, 1024]
  • blk.28.ffn_down.weight
    BF16
    [21504, 4096]
  • blk.28.ffn_norm.weight
    F32
    [4096]
  • blk.28.ffn_up.weight
    BF16
    [4096, 21504]
  • blk.29
  • blk.29.attn_k.weight
    BF16
    [4096, 1024]
  • blk.29.attn_k_norm.weight
    F32
    [128]
  • blk.29.attn_norm.weight
    F32
    [4096]
  • blk.29.attn_output.weight
    BF16
    [4096, 4096]
  • blk.29.attn_q.weight
    BF16
    [4096, 4096]
  • blk.29.attn_q_norm.weight
    F32
    [128]
  • blk.29.attn_v.weight
    BF16
    [4096, 1024]
  • blk.29.ffn_down.weight
    BF16
    [21504, 4096]
  • blk.29.ffn_norm.weight
    F32
    [4096]
  • blk.29.ffn_up.weight
    BF16
    [4096, 21504]
  • blk.30
  • blk.30.attn_k.weight
    BF16
    [4096, 1024]
  • blk.30.attn_k_norm.weight
    F32
    [128]
  • blk.30.attn_norm.weight
    F32
    [4096]
  • blk.30.attn_output.weight
    BF16
    [4096, 4096]
  • blk.30.attn_q.weight
    BF16
    [4096, 4096]
  • blk.30.attn_q_norm.weight
    F32
    [128]
  • blk.30.attn_v.weight
    BF16
    [4096, 1024]
  • blk.30.ffn_down.weight
    BF16
    [21504, 4096]
  • blk.30.ffn_norm.weight
    F32
    [4096]
  • blk.30.ffn_up.weight
    BF16
    [4096, 21504]
  • blk.31
  • blk.31.attn_k.weight
    BF16
    [4096, 1024]
  • blk.31.attn_k_norm.weight
    F32
    [128]
  • blk.31.attn_norm.weight
    F32
    [4096]
  • blk.31.attn_output.weight
    BF16
    [4096, 4096]
  • blk.31.attn_q.weight
    BF16
    [4096, 4096]
  • blk.31.attn_q_norm.weight
    F32
    [128]
  • blk.31.attn_v.weight
    BF16
    [4096, 1024]
  • blk.31.ffn_down.weight
    BF16
    [21504, 4096]
  • blk.31.ffn_norm.weight
    F32
    [4096]
  • blk.31.ffn_up.weight
    BF16
    [4096, 21504]
  • output.weight
    BF16
    [4096, 131072]
  • rope_freqs.weight
    F32
    [64]
  • output_norm.weight
    F32
    [4096]