jyan1 / openelm

22 3 months ago

1 Tag
... /
model
0a07dee90a7b · 6.1GB
    Metadata
  • general.architecture
    openelm
  • general.file_type
    1
  • general.license
    other
  • general.license.link
    LICENSE
  • general.license.name
    apple-sample-code-license
  • general.name
    Openelm
  • general.quantization_version
    2
  • general.size_label
    3.0B
  • general.type
    model
  • openelm.attention.head_count
    [12, 12, 12, 12, 16, ...]
  • openelm.attention.head_count_kv
    [3, 3, 3, 3, 4, ...]
  • openelm.attention.key_length
    128
  • openelm.attention.layer_norm_rms_epsilon
    1e-06
  • openelm.attention.value_length
    128
  • openelm.block_count
    36
  • openelm.context_length
    2048
  • openelm.embedding_length
    3072
  • openelm.feed_forward_length
    [1536, 1792, 2048, 2560, 2816, ...]
  • openelm.rope.dimension_count
    128
  • openelm.rope.freq_base
    10000
  • tokenizer.ggml.add_bos_token
    true
  • tokenizer.ggml.add_eos_token
    false
  • tokenizer.ggml.bos_token_id
    1
  • tokenizer.ggml.eos_token_id
    2
  • tokenizer.ggml.model
    llama
  • tokenizer.ggml.pre
    default
  • tokenizer.ggml.scores
    [0, 0, 0, 0, 0, ...]
  • tokenizer.ggml.token_type
    [2, 3, 3, 6, 6, ...]
  • tokenizer.ggml.tokens
    [<unk>, <s>, </s>, <0x00>, <0x01>, ...]
  • tokenizer.ggml.unknown_token_id
    0
    • Tensor
  • Name
    Type
    Shape
  • token_embd.weight
    F16
    [3072, 32000]
    • blk.0
  • blk.0.attn_k_norm.weight
    F32
    [128]
  • blk.0.attn_norm.weight
    F32
    [3072]
  • blk.0.attn_output.weight
    F16
    [1536, 3072]
  • blk.0.attn_q_norm.weight
    F32
    [128]
  • blk.0.attn_qkv.weight
    F16
    [3072, 2304]
  • blk.0.ffn_down.weight
    F16
    [1536, 3072]
  • blk.0.ffn_gate.weight
    F16
    [3072, 1536]
  • blk.0.ffn_norm.weight
    F32
    [3072]
  • blk.0.ffn_up.weight
    F16
    [3072, 1536]
    • blk.1
  • blk.1.attn_k_norm.weight
    F32
    [128]
  • blk.1.attn_norm.weight
    F32
    [3072]
  • blk.1.attn_output.weight
    F16
    [1536, 3072]
  • blk.1.attn_q_norm.weight
    F32
    [128]
  • blk.1.attn_qkv.weight
    F16
    [3072, 2304]
  • blk.1.ffn_down.weight
    F16
    [1792, 3072]
  • blk.1.ffn_gate.weight
    F16
    [3072, 1792]
  • blk.1.ffn_norm.weight
    F32
    [3072]
  • blk.1.ffn_up.weight
    F16
    [3072, 1792]
    • blk.2
  • blk.2.attn_k_norm.weight
    F32
    [128]
  • blk.2.attn_norm.weight
    F32
    [3072]
  • blk.2.attn_output.weight
    F16
    [1536, 3072]
  • blk.2.attn_q_norm.weight
    F32
    [128]
  • blk.2.attn_qkv.weight
    F16
    [3072, 2304]
  • blk.2.ffn_down.weight
    F16
    [2048, 3072]
  • blk.2.ffn_gate.weight
    F16
    [3072, 2048]
  • blk.2.ffn_norm.weight
    F32
    [3072]
  • blk.2.ffn_up.weight
    F16
    [3072, 2048]
    • blk.3
  • blk.3.attn_k_norm.weight
    F32
    [128]
  • blk.3.attn_norm.weight
    F32
    [3072]
  • blk.3.attn_output.weight
    F16
    [1536, 3072]
  • blk.3.attn_q_norm.weight
    F32
    [128]
  • blk.3.attn_qkv.weight
    F16
    [3072, 2304]
  • blk.3.ffn_down.weight
    F16
    [2560, 3072]
  • blk.3.ffn_gate.weight
    F16
    [3072, 2560]
  • blk.3.ffn_norm.weight
    F32
    [3072]
  • blk.3.ffn_up.weight
    F16
    [3072, 2560]
    • blk.4
  • blk.4.attn_k_norm.weight
    F32
    [128]
  • blk.4.attn_norm.weight
    F32
    [3072]
  • blk.4.attn_output.weight
    F16
    [2048, 3072]
  • blk.4.attn_q_norm.weight
    F32
    [128]
  • blk.4.attn_qkv.weight
    F16
    [3072, 3072]
  • blk.4.ffn_down.weight
    F16
    [2816, 3072]
  • blk.4.ffn_gate.weight
    F16
    [3072, 2816]
  • blk.4.ffn_norm.weight
    F32
    [3072]
  • blk.4.ffn_up.weight
    F16
    [3072, 2816]
    • blk.5
  • blk.5.attn_k_norm.weight
    F32
    [128]
  • blk.5.attn_norm.weight
    F32
    [3072]
  • blk.5.attn_output.weight
    F16
    [2048, 3072]
  • blk.5.attn_q_norm.weight
    F32
    [128]
  • blk.5.attn_qkv.weight
    F16
    [3072, 3072]
  • blk.5.ffn_down.weight
    F16
    [3072, 3072]
  • blk.5.ffn_gate.weight
    F16
    [3072, 3072]
  • blk.5.ffn_norm.weight
    F32
    [3072]
  • blk.5.ffn_up.weight
    F16
    [3072, 3072]
    • blk.6
  • blk.6.attn_k_norm.weight
    F32
    [128]
  • blk.6.attn_norm.weight
    F32
    [3072]
  • blk.6.attn_output.weight
    F16
    [2048, 3072]
  • blk.6.attn_q_norm.weight
    F32
    [128]
  • blk.6.attn_qkv.weight
    F16
    [3072, 3072]
  • blk.6.ffn_down.weight
    F16
    [3328, 3072]
  • blk.6.ffn_gate.weight
    F16
    [3072, 3328]
  • blk.6.ffn_norm.weight
    F32
    [3072]
  • blk.6.ffn_up.weight
    F16
    [3072, 3328]
    • blk.7
  • blk.7.attn_k_norm.weight
    F32
    [128]
  • blk.7.attn_norm.weight
    F32
    [3072]
  • blk.7.attn_output.weight
    F16
    [2048, 3072]
  • blk.7.attn_q_norm.weight
    F32
    [128]
  • blk.7.attn_qkv.weight
    F16
    [3072, 3072]
  • blk.7.ffn_down.weight
    F16
    [3584, 3072]
  • blk.7.ffn_gate.weight
    F16
    [3072, 3584]
  • blk.7.ffn_norm.weight
    F32
    [3072]
  • blk.7.ffn_up.weight
    F16
    [3072, 3584]
    • blk.8
  • blk.8.attn_k_norm.weight
    F32
    [128]
  • blk.8.attn_norm.weight
    F32
    [3072]
  • blk.8.attn_output.weight
    F16
    [2048, 3072]
  • blk.8.attn_q_norm.weight
    F32
    [128]
  • blk.8.attn_qkv.weight
    F16
    [3072, 3072]
  • blk.8.ffn_down.weight
    F16
    [4096, 3072]
  • blk.8.ffn_gate.weight
    F16
    [3072, 4096]
  • blk.8.ffn_norm.weight
    F32
    [3072]
  • blk.8.ffn_up.weight
    F16
    [3072, 4096]
    • blk.9
  • blk.9.attn_k_norm.weight
    F32
    [128]
  • blk.9.attn_norm.weight
    F32
    [3072]
  • blk.9.attn_output.weight
    F16
    [2048, 3072]
  • blk.9.attn_q_norm.weight
    F32
    [128]
  • blk.9.attn_qkv.weight
    F16
    [3072, 3072]
  • blk.9.ffn_down.weight
    F16
    [4352, 3072]
  • blk.9.ffn_gate.weight
    F16
    [3072, 4352]
  • blk.9.ffn_norm.weight
    F32
    [3072]
  • blk.9.ffn_up.weight
    F16
    [3072, 4352]
    • blk.10
  • blk.10.attn_k_norm.weight
    F32
    [128]
  • blk.10.attn_norm.weight
    F32
    [3072]
  • blk.10.attn_output.weight
    F16
    [2048, 3072]
  • blk.10.attn_q_norm.weight
    F32
    [128]
  • blk.10.attn_qkv.weight
    F16
    [3072, 3072]
  • blk.10.ffn_down.weight
    F16
    [4608, 3072]
  • blk.10.ffn_gate.weight
    F16
    [3072, 4608]
  • blk.10.ffn_norm.weight
    F32
    [3072]
  • blk.10.ffn_up.weight
    F16
    [3072, 4608]
    • blk.11
  • blk.11.attn_k_norm.weight
    F32
    [128]
  • blk.11.attn_norm.weight
    F32
    [3072]
  • blk.11.attn_output.weight
    F16
    [2048, 3072]
  • blk.11.attn_q_norm.weight
    F32
    [128]
  • blk.11.attn_qkv.weight
    F16
    [3072, 3072]
  • blk.11.ffn_down.weight
    F16
    [4864, 3072]
  • blk.11.ffn_gate.weight
    F16
    [3072, 4864]
  • blk.11.ffn_norm.weight
    F32
    [3072]
  • blk.11.ffn_up.weight
    F16
    [3072, 4864]
    • blk.12
  • blk.12.attn_k_norm.weight
    F32
    [128]
  • blk.12.attn_norm.weight
    F32
    [3072]
  • blk.12.attn_output.weight
    F16
    [2048, 3072]
  • blk.12.attn_q_norm.weight
    F32
    [128]
  • blk.12.attn_qkv.weight
    F16
    [3072, 3072]
  • blk.12.ffn_down.weight
    F16
    [5120, 3072]
  • blk.12.ffn_gate.weight
    F16
    [3072, 5120]
  • blk.12.ffn_norm.weight
    F32
    [3072]
  • blk.12.ffn_up.weight
    F16
    [3072, 5120]
    • blk.13
  • blk.13.attn_k_norm.weight
    F32
    [128]
  • blk.13.attn_norm.weight
    F32
    [3072]
  • blk.13.attn_output.weight
    F16
    [2048, 3072]
  • blk.13.attn_q_norm.weight
    F32
    [128]
  • blk.13.attn_qkv.weight
    F16
    [3072, 3072]
  • blk.13.ffn_down.weight
    F16
    [5632, 3072]
  • blk.13.ffn_gate.weight
    F16
    [3072, 5632]
  • blk.13.ffn_norm.weight
    F32
    [3072]
  • blk.13.ffn_up.weight
    F16
    [3072, 5632]
    • blk.14
  • blk.14.attn_k_norm.weight
    F32
    [128]
  • blk.14.attn_norm.weight
    F32
    [3072]
  • blk.14.attn_output.weight
    F16
    [2048, 3072]
  • blk.14.attn_q_norm.weight
    F32
    [128]
  • blk.14.attn_qkv.weight
    F16
    [3072, 3072]
  • blk.14.ffn_down.weight
    F16
    [5888, 3072]
  • blk.14.ffn_gate.weight
    F16
    [3072, 5888]
  • blk.14.ffn_norm.weight
    F32
    [3072]
  • blk.14.ffn_up.weight
    F16
    [3072, 5888]
    • blk.15
  • blk.15.attn_k_norm.weight
    F32
    [128]
  • blk.15.attn_norm.weight
    F32
    [3072]
  • blk.15.attn_output.weight
    F16
    [2048, 3072]
  • blk.15.attn_q_norm.weight
    F32
    [128]
  • blk.15.attn_qkv.weight
    F16
    [3072, 3072]
  • blk.15.ffn_down.weight
    F16
    [6144, 3072]
  • blk.15.ffn_gate.weight
    F16
    [3072, 6144]
  • blk.15.ffn_norm.weight
    F32
    [3072]
  • blk.15.ffn_up.weight
    F16
    [3072, 6144]
    • blk.16
  • blk.16.attn_k_norm.weight
    F32
    [128]
  • blk.16.attn_norm.weight
    F32
    [3072]
  • blk.16.attn_output.weight
    F16
    [2048, 3072]
  • blk.16.attn_q_norm.weight
    F32
    [128]
  • blk.16.attn_qkv.weight
    F16
    [3072, 3072]
  • blk.16.ffn_down.weight
    F16
    [6400, 3072]
  • blk.16.ffn_gate.weight
    F16
    [3072, 6400]
  • blk.16.ffn_norm.weight
    F32
    [3072]
  • blk.16.ffn_up.weight
    F16
    [3072, 6400]
    • blk.17
  • blk.17.attn_k_norm.weight
    F32
    [128]
  • blk.17.attn_norm.weight
    F32
    [3072]
  • blk.17.attn_output.weight
    F16
    [2048, 3072]
  • blk.17.attn_q_norm.weight
    F32
    [128]
  • blk.17.attn_qkv.weight
    F16
    [3072, 3072]
  • blk.17.ffn_down.weight
    F16
    [6656, 3072]
  • blk.17.ffn_gate.weight
    F16
    [3072, 6656]
  • blk.17.ffn_norm.weight
    F32
    [3072]
  • blk.17.ffn_up.weight
    F16
    [3072, 6656]
    • blk.18
  • blk.18.attn_k_norm.weight
    F32
    [128]
  • blk.18.attn_norm.weight
    F32
    [3072]
  • blk.18.attn_output.weight
    F16
    [2560, 3072]
  • blk.18.attn_q_norm.weight
    F32
    [128]
  • blk.18.attn_qkv.weight
    F16
    [3072, 3840]
  • blk.18.ffn_down.weight
    F16
    [7168, 3072]
  • blk.18.ffn_gate.weight
    F16
    [3072, 7168]
  • blk.18.ffn_norm.weight
    F32
    [3072]
  • blk.18.ffn_up.weight
    F16
    [3072, 7168]
    • blk.19
  • blk.19.attn_k_norm.weight
    F32
    [128]
  • blk.19.attn_norm.weight
    F32
    [3072]
  • blk.19.attn_output.weight
    F16
    [2560, 3072]
  • blk.19.attn_q_norm.weight
    F32
    [128]
  • blk.19.attn_qkv.weight
    F16
    [3072, 3840]
  • blk.19.ffn_down.weight
    F16
    [7424, 3072]
  • blk.19.ffn_gate.weight
    F16
    [3072, 7424]
  • blk.19.ffn_norm.weight
    F32
    [3072]
  • blk.19.ffn_up.weight
    F16
    [3072, 7424]
    • blk.20
  • blk.20.attn_k_norm.weight
    F32
    [128]
  • blk.20.attn_norm.weight
    F32
    [3072]
  • blk.20.attn_output.weight
    F16
    [2560, 3072]
  • blk.20.attn_q_norm.weight
    F32
    [128]
  • blk.20.attn_qkv.weight
    F16
    [3072, 3840]
  • blk.20.ffn_down.weight
    F16
    [7680, 3072]
  • blk.20.ffn_gate.weight
    F16
    [3072, 7680]
  • blk.20.ffn_norm.weight
    F32
    [3072]
  • blk.20.ffn_up.weight
    F16
    [3072, 7680]
    • blk.21
  • blk.21.attn_k_norm.weight
    F32
    [128]
  • blk.21.attn_norm.weight
    F32
    [3072]
  • blk.21.attn_output.weight
    F16
    [2560, 3072]
  • blk.21.attn_q_norm.weight
    F32
    [128]
  • blk.21.attn_qkv.weight
    F16
    [3072, 3840]
  • blk.21.ffn_down.weight
    F16
    [7936, 3072]
  • blk.21.ffn_gate.weight
    F16
    [3072, 7936]
  • blk.21.ffn_norm.weight
    F32
    [3072]
  • blk.21.ffn_up.weight
    F16
    [3072, 7936]
    • blk.22
  • blk.22.attn_k_norm.weight
    F32
    [128]
  • blk.22.attn_norm.weight
    F32
    [3072]
  • blk.22.attn_output.weight
    F16
    [2560, 3072]
  • blk.22.attn_q_norm.weight
    F32
    [128]
  • blk.22.attn_qkv.weight
    F16
    [3072, 3840]
  • blk.22.ffn_down.weight
    F16
    [8192, 3072]
  • blk.22.ffn_gate.weight
    F16
    [3072, 8192]
  • blk.22.ffn_norm.weight
    F32
    [3072]
  • blk.22.ffn_up.weight
    F16
    [3072, 8192]
    • blk.23
  • blk.23.attn_k_norm.weight
    F32
    [128]
  • blk.23.attn_norm.weight
    F32
    [3072]
  • blk.23.attn_output.weight
    F16
    [2560, 3072]
  • blk.23.attn_q_norm.weight
    F32
    [128]
  • blk.23.attn_qkv.weight
    F16
    [3072, 3840]
  • blk.23.ffn_down.weight
    F16
    [8704, 3072]
  • blk.23.ffn_gate.weight
    F16
    [3072, 8704]
  • blk.23.ffn_norm.weight
    F32
    [3072]
  • blk.23.ffn_up.weight
    F16
    [3072, 8704]
    • blk.24
  • blk.24.attn_k_norm.weight
    F32
    [128]
  • blk.24.attn_norm.weight
    F32
    [3072]
  • blk.24.attn_output.weight
    F16
    [2560, 3072]
  • blk.24.attn_q_norm.weight
    F32
    [128]
  • blk.24.attn_qkv.weight
    F16
    [3072, 3840]
  • blk.24.ffn_down.weight
    F16
    [8960, 3072]
  • blk.24.ffn_gate.weight
    F16
    [3072, 8960]
  • blk.24.ffn_norm.weight
    F32
    [3072]
  • blk.24.ffn_up.weight
    F16
    [3072, 8960]
    • blk.25
  • blk.25.attn_k_norm.weight
    F32
    [128]
  • blk.25.attn_norm.weight
    F32
    [3072]
  • blk.25.attn_output.weight
    F16
    [2560, 3072]
  • blk.25.attn_q_norm.weight
    F32
    [128]
  • blk.25.attn_qkv.weight
    F16
    [3072, 3840]
  • blk.25.ffn_down.weight
    F16
    [9216, 3072]
  • blk.25.ffn_gate.weight
    F16
    [3072, 9216]
  • blk.25.ffn_norm.weight
    F32
    [3072]
  • blk.25.ffn_up.weight
    F16
    [3072, 9216]
    • blk.26
  • blk.26.attn_k_norm.weight
    F32
    [128]
  • blk.26.attn_norm.weight
    F32
    [3072]
  • blk.26.attn_output.weight
    F16
    [2560, 3072]
  • blk.26.attn_q_norm.weight
    F32
    [128]
  • blk.26.attn_qkv.weight
    F16
    [3072, 3840]
  • blk.26.ffn_down.weight
    F16
    [9472, 3072]
  • blk.26.ffn_gate.weight
    F16
    [3072, 9472]
  • blk.26.ffn_norm.weight
    F32
    [3072]
  • blk.26.ffn_up.weight
    F16
    [3072, 9472]
    • blk.27
  • blk.27.attn_k_norm.weight
    F32
    [128]
  • blk.27.attn_norm.weight
    F32
    [3072]
  • blk.27.attn_output.weight
    F16
    [2560, 3072]
  • blk.27.attn_q_norm.weight
    F32
    [128]
  • blk.27.attn_qkv.weight
    F16
    [3072, 3840]
  • blk.27.ffn_down.weight
    F16
    [9728, 3072]
  • blk.27.ffn_gate.weight
    F16
    [3072, 9728]
  • blk.27.ffn_norm.weight
    F32
    [3072]
  • blk.27.ffn_up.weight
    F16
    [3072, 9728]
    • blk.28
  • blk.28.attn_k_norm.weight
    F32
    [128]
  • blk.28.attn_norm.weight
    F32
    [3072]
  • blk.28.attn_output.weight
    F16
    [2560, 3072]
  • blk.28.attn_q_norm.weight
    F32
    [128]
  • blk.28.attn_qkv.weight
    F16
    [3072, 3840]
  • blk.28.ffn_down.weight
    F16
    [10240, 3072]
  • blk.28.ffn_gate.weight
    F16
    [3072, 10240]
  • blk.28.ffn_norm.weight
    F32
    [3072]
  • blk.28.ffn_up.weight
    F16
    [3072, 10240]
    • blk.29
  • blk.29.attn_k_norm.weight
    F32
    [128]
  • blk.29.attn_norm.weight
    F32
    [3072]
  • blk.29.attn_output.weight
    F16
    [2560, 3072]
  • blk.29.attn_q_norm.weight
    F32
    [128]
  • blk.29.attn_qkv.weight
    F16
    [3072, 3840]
  • blk.29.ffn_down.weight
    F16
    [10496, 3072]
  • blk.29.ffn_gate.weight
    F16
    [3072, 10496]
  • blk.29.ffn_norm.weight
    F32
    [3072]
  • blk.29.ffn_up.weight
    F16
    [3072, 10496]
    • blk.30
  • blk.30.attn_k_norm.weight
    F32
    [128]
  • blk.30.attn_norm.weight
    F32
    [3072]
  • blk.30.attn_output.weight
    F16
    [3072, 3072]
  • blk.30.attn_q_norm.weight
    F32
    [128]
  • blk.30.attn_qkv.weight
    F16
    [3072, 4608]
  • blk.30.ffn_down.weight
    F16
    [10752, 3072]
  • blk.30.ffn_gate.weight
    F16
    [3072, 10752]
  • blk.30.ffn_norm.weight
    F32
    [3072]
  • blk.30.ffn_up.weight
    F16
    [3072, 10752]
    • blk.31
  • blk.31.attn_k_norm.weight
    F32
    [128]
  • blk.31.attn_norm.weight
    F32
    [3072]
  • blk.31.attn_output.weight
    F16
    [3072, 3072]
  • blk.31.attn_q_norm.weight
    F32
    [128]
  • blk.31.attn_qkv.weight
    F16
    [3072, 4608]
  • blk.31.ffn_down.weight
    F16
    [11008, 3072]
  • blk.31.ffn_gate.weight
    F16
    [3072, 11008]
  • blk.31.ffn_norm.weight
    F32
    [3072]
  • blk.31.ffn_up.weight
    F16
    [3072, 11008]
    • blk.32
  • blk.32.attn_k_norm.weight
    F32
    [128]
  • blk.32.attn_norm.weight
    F32
    [3072]
  • blk.32.attn_output.weight
    F16
    [3072, 3072]
  • blk.32.attn_q_norm.weight
    F32
    [128]
  • blk.32.attn_qkv.weight
    F16
    [3072, 4608]
  • blk.32.ffn_down.weight
    F16
    [11264, 3072]
  • blk.32.ffn_gate.weight
    F16
    [3072, 11264]
  • blk.32.ffn_norm.weight
    F32
    [3072]
  • blk.32.ffn_up.weight
    F16
    [3072, 11264]
    • blk.33
  • blk.33.attn_k_norm.weight
    F32
    [128]
  • blk.33.attn_norm.weight
    F32
    [3072]
  • blk.33.attn_output.weight
    F16
    [3072, 3072]
  • blk.33.attn_q_norm.weight
    F32
    [128]
  • blk.33.attn_qkv.weight
    F16
    [3072, 4608]
  • blk.33.ffn_down.weight
    F16
    [11776, 3072]
  • blk.33.ffn_gate.weight
    F16
    [3072, 11776]
  • blk.33.ffn_norm.weight
    F32
    [3072]
  • blk.33.ffn_up.weight
    F16
    [3072, 11776]
    • blk.34
  • blk.34.attn_k_norm.weight
    F32
    [128]
  • blk.34.attn_norm.weight
    F32
    [3072]
  • blk.34.attn_output.weight
    F16
    [3072, 3072]
  • blk.34.attn_q_norm.weight
    F32
    [128]
  • blk.34.attn_qkv.weight
    F16
    [3072, 4608]
  • blk.34.ffn_down.weight
    F16
    [12032, 3072]
  • blk.34.ffn_gate.weight
    F16
    [3072, 12032]
  • blk.34.ffn_norm.weight
    F32
    [3072]
  • blk.34.ffn_up.weight
    F16
    [3072, 12032]
    • blk.35
  • blk.35.attn_k_norm.weight
    F32
    [128]
  • blk.35.attn_norm.weight
    F32
    [3072]
  • blk.35.attn_output.weight
    F16
    [3072, 3072]
  • blk.35.attn_q_norm.weight
    F32
    [128]
  • blk.35.attn_qkv.weight
    F16
    [3072, 4608]
  • blk.35.ffn_down.weight
    F16
    [12288, 3072]
  • blk.35.ffn_gate.weight
    F16
    [3072, 12288]
  • blk.35.ffn_norm.weight
    F32
    [3072]
  • blk.35.ffn_up.weight
    F16
    [3072, 12288]
  • output_norm.weight
    F32
    [3072]