mannix / llama3-sppo-iter3

Meta Llama-3-8b with Self-Play Preference Optimization for Language Model Alignment at iteration 3

186 4 months ago

21 Tags
... /
model
5862a8175edc · 3.3GB
    Metadata
  • general.architecture
    llama
  • general.file_type
    23
  • general.name
    ..
  • general.quantization_version
    2
  • llama.attention.head_count
    32
  • llama.attention.head_count_kv
    8
  • llama.attention.layer_norm_rms_epsilon
    1e-05
  • llama.block_count
    32
  • llama.context_length
    8192
  • llama.embedding_length
    4096
  • llama.feed_forward_length
    14336
  • llama.rope.dimension_count
    128
  • llama.rope.freq_base
    500000
  • llama.vocab_size
    128256
  • quantize.imatrix.chunks_count
    124
  • quantize.imatrix.dataset
    /shared/opt/work_models/_imatrix/calibration_datav3.txt
  • quantize.imatrix.entries_count
    224
  • quantize.imatrix.file
    imatrix.dat
  • tokenizer.ggml.bos_token_id
    128000
  • tokenizer.ggml.eos_token_id
    128009
  • tokenizer.ggml.merges
    [Ġ Ġ, Ġ ĠĠĠ, ĠĠ ĠĠ, ĠĠĠ Ġ, i n, ...]
  • tokenizer.ggml.model
    gpt2
  • tokenizer.ggml.padding_token_id
    128009
  • tokenizer.ggml.pre
    llama-bpe
  • tokenizer.ggml.token_type
    [1, 1, 1, 1, 1, ...]
  • tokenizer.ggml.tokens
    [!, ", #, $, %, ...]
    • Tensor
  • Name
    Type
    Shape
  • token_embd.weight
    (!unknown_type 21!)
    [4096, 128256]
    • blk.0
  • blk.0.attn_k.weight
    (!unknown_type 22!)
    [4096, 1024]
  • blk.0.attn_norm.weight
    F32
    [4096]
  • blk.0.attn_output.weight
    (!unknown_type 21!)
    [4096, 4096]
  • blk.0.attn_q.weight
    (!unknown_type 22!)
    [4096, 4096]
  • blk.0.attn_v.weight
    Q4_K
    [4096, 1024]
  • blk.0.ffn_down.weight
    I32
    [14336, 4096]
  • blk.0.ffn_gate.weight
    I32
    [4096, 14336]
  • blk.0.ffn_norm.weight
    F32
    [4096]
  • blk.0.ffn_up.weight
    I32
    [4096, 14336]
    • blk.1
  • blk.1.attn_k.weight
    (!unknown_type 22!)
    [4096, 1024]
  • blk.1.attn_norm.weight
    F32
    [4096]
  • blk.1.attn_output.weight
    (!unknown_type 21!)
    [4096, 4096]
  • blk.1.attn_q.weight
    (!unknown_type 22!)
    [4096, 4096]
  • blk.1.attn_v.weight
    Q4_K
    [4096, 1024]
  • blk.1.ffn_down.weight
    I32
    [14336, 4096]
  • blk.1.ffn_gate.weight
    I32
    [4096, 14336]
  • blk.1.ffn_norm.weight
    F32
    [4096]
  • blk.1.ffn_up.weight
    I32
    [4096, 14336]
    • blk.2
  • blk.2.attn_k.weight
    (!unknown_type 22!)
    [4096, 1024]
  • blk.2.attn_norm.weight
    F32
    [4096]
  • blk.2.attn_output.weight
    (!unknown_type 21!)
    [4096, 4096]
  • blk.2.attn_q.weight
    (!unknown_type 22!)
    [4096, 4096]
  • blk.2.attn_v.weight
    Q4_K
    [4096, 1024]
  • blk.2.ffn_down.weight
    I32
    [14336, 4096]
  • blk.2.ffn_gate.weight
    I32
    [4096, 14336]
  • blk.2.ffn_norm.weight
    F32
    [4096]
  • blk.2.ffn_up.weight
    I32
    [4096, 14336]
    • blk.3
  • blk.3.attn_k.weight
    (!unknown_type 22!)
    [4096, 1024]
  • blk.3.attn_norm.weight
    F32
    [4096]
  • blk.3.attn_output.weight
    (!unknown_type 21!)
    [4096, 4096]
  • blk.3.attn_q.weight
    (!unknown_type 22!)
    [4096, 4096]
  • blk.3.attn_v.weight
    Q4_K
    [4096, 1024]
  • blk.3.ffn_down.weight
    I32
    [14336, 4096]
  • blk.3.ffn_gate.weight
    I32
    [4096, 14336]
  • blk.3.ffn_norm.weight
    F32
    [4096]
  • blk.3.ffn_up.weight
    I32
    [4096, 14336]
    • blk.4
  • blk.4.attn_k.weight
    (!unknown_type 22!)
    [4096, 1024]
  • blk.4.attn_norm.weight
    F32
    [4096]
  • blk.4.attn_output.weight
    (!unknown_type 21!)
    [4096, 4096]
  • blk.4.attn_q.weight
    (!unknown_type 22!)
    [4096, 4096]
  • blk.4.attn_v.weight
    Q4_K
    [4096, 1024]
  • blk.4.ffn_down.weight
    I32
    [14336, 4096]
  • blk.4.ffn_gate.weight
    I32
    [4096, 14336]
  • blk.4.ffn_norm.weight
    F32
    [4096]
  • blk.4.ffn_up.weight
    I32
    [4096, 14336]
    • blk.5
  • blk.5.attn_k.weight
    (!unknown_type 22!)
    [4096, 1024]
  • blk.5.attn_norm.weight
    F32
    [4096]
  • blk.5.attn_output.weight
    (!unknown_type 21!)
    [4096, 4096]
  • blk.5.attn_q.weight
    (!unknown_type 22!)
    [4096, 4096]
  • blk.5.attn_v.weight
    Q4_K
    [4096, 1024]
  • blk.5.ffn_down.weight
    I32
    [14336, 4096]
  • blk.5.ffn_gate.weight
    I32
    [4096, 14336]
  • blk.5.ffn_norm.weight
    F32
    [4096]
  • blk.5.ffn_up.weight
    I32
    [4096, 14336]
    • blk.6
  • blk.6.attn_k.weight
    (!unknown_type 22!)
    [4096, 1024]
  • blk.6.attn_norm.weight
    F32
    [4096]
  • blk.6.attn_output.weight
    (!unknown_type 21!)
    [4096, 4096]
  • blk.6.attn_q.weight
    (!unknown_type 22!)
    [4096, 4096]
  • blk.6.attn_v.weight
    Q4_K
    [4096, 1024]
  • blk.6.ffn_down.weight
    I32
    [14336, 4096]
  • blk.6.ffn_gate.weight
    I32
    [4096, 14336]
  • blk.6.ffn_norm.weight
    F32
    [4096]
  • blk.6.ffn_up.weight
    I32
    [4096, 14336]
    • blk.7
  • blk.7.attn_k.weight
    (!unknown_type 22!)
    [4096, 1024]
  • blk.7.attn_norm.weight
    F32
    [4096]
  • blk.7.attn_output.weight
    (!unknown_type 21!)
    [4096, 4096]
  • blk.7.attn_q.weight
    (!unknown_type 22!)
    [4096, 4096]
  • blk.7.attn_v.weight
    Q4_K
    [4096, 1024]
  • blk.7.ffn_down.weight
    I32
    [14336, 4096]
  • blk.7.ffn_gate.weight
    I32
    [4096, 14336]
  • blk.7.ffn_norm.weight
    F32
    [4096]
  • blk.7.ffn_up.weight
    I32
    [4096, 14336]
    • blk.8
  • blk.8.attn_k.weight
    (!unknown_type 22!)
    [4096, 1024]
  • blk.8.attn_norm.weight
    F32
    [4096]
  • blk.8.attn_output.weight
    (!unknown_type 21!)
    [4096, 4096]
  • blk.8.attn_q.weight
    (!unknown_type 22!)
    [4096, 4096]
  • blk.8.attn_v.weight
    Q4_K
    [4096, 1024]
  • blk.8.ffn_down.weight
    I32
    [14336, 4096]
  • blk.8.ffn_gate.weight
    I32
    [4096, 14336]
  • blk.8.ffn_norm.weight
    F32
    [4096]
  • blk.8.ffn_up.weight
    I32
    [4096, 14336]
    • blk.9
  • blk.9.attn_k.weight
    (!unknown_type 22!)
    [4096, 1024]
  • blk.9.attn_norm.weight
    F32
    [4096]
  • blk.9.attn_output.weight
    (!unknown_type 21!)
    [4096, 4096]
  • blk.9.attn_q.weight
    (!unknown_type 22!)
    [4096, 4096]
  • blk.9.attn_v.weight
    Q4_K
    [4096, 1024]
  • blk.9.ffn_down.weight
    I32
    [14336, 4096]
  • blk.9.ffn_gate.weight
    I32
    [4096, 14336]
  • blk.9.ffn_norm.weight
    F32
    [4096]
  • blk.9.ffn_up.weight
    I32
    [4096, 14336]
    • blk.10
  • blk.10.attn_k.weight
    (!unknown_type 22!)
    [4096, 1024]
  • blk.10.attn_norm.weight
    F32
    [4096]
  • blk.10.attn_output.weight
    (!unknown_type 21!)
    [4096, 4096]
  • blk.10.attn_q.weight
    (!unknown_type 22!)
    [4096, 4096]
  • blk.10.attn_v.weight
    Q4_K
    [4096, 1024]
  • blk.10.ffn_down.weight
    I32
    [14336, 4096]
  • blk.10.ffn_gate.weight
    I32
    [4096, 14336]
  • blk.10.ffn_norm.weight
    F32
    [4096]
  • blk.10.ffn_up.weight
    I32
    [4096, 14336]
    • blk.11
  • blk.11.attn_k.weight
    (!unknown_type 22!)
    [4096, 1024]
  • blk.11.attn_norm.weight
    F32
    [4096]
  • blk.11.attn_output.weight
    (!unknown_type 21!)
    [4096, 4096]
  • blk.11.attn_q.weight
    (!unknown_type 22!)
    [4096, 4096]
  • blk.11.attn_v.weight
    Q4_K
    [4096, 1024]
  • blk.11.ffn_down.weight
    I32
    [14336, 4096]
  • blk.11.ffn_gate.weight
    I32
    [4096, 14336]
  • blk.11.ffn_norm.weight
    F32
    [4096]
  • blk.11.ffn_up.weight
    I32
    [4096, 14336]
    • blk.12
  • blk.12.attn_k.weight
    (!unknown_type 22!)
    [4096, 1024]
  • blk.12.attn_norm.weight
    F32
    [4096]
  • blk.12.attn_output.weight
    (!unknown_type 21!)
    [4096, 4096]
  • blk.12.attn_q.weight
    (!unknown_type 22!)
    [4096, 4096]
  • blk.12.attn_v.weight
    Q4_K
    [4096, 1024]
  • blk.12.ffn_down.weight
    I32
    [14336, 4096]
  • blk.12.ffn_gate.weight
    I32
    [4096, 14336]
  • blk.12.ffn_norm.weight
    F32
    [4096]
  • blk.12.ffn_up.weight
    I32
    [4096, 14336]
    • blk.13
  • blk.13.attn_k.weight
    (!unknown_type 22!)
    [4096, 1024]
  • blk.13.attn_norm.weight
    F32
    [4096]
  • blk.13.attn_output.weight
    (!unknown_type 21!)
    [4096, 4096]
  • blk.13.attn_q.weight
    (!unknown_type 22!)
    [4096, 4096]
  • blk.13.attn_v.weight
    Q4_K
    [4096, 1024]
  • blk.13.ffn_down.weight
    I32
    [14336, 4096]
  • blk.13.ffn_gate.weight
    I32
    [4096, 14336]
  • blk.13.ffn_norm.weight
    F32
    [4096]
  • blk.13.ffn_up.weight
    I32
    [4096, 14336]
    • blk.14
  • blk.14.attn_k.weight
    (!unknown_type 22!)
    [4096, 1024]
  • blk.14.attn_norm.weight
    F32
    [4096]
  • blk.14.attn_output.weight
    (!unknown_type 21!)
    [4096, 4096]
  • blk.14.attn_q.weight
    (!unknown_type 22!)
    [4096, 4096]
  • blk.14.attn_v.weight
    Q4_K
    [4096, 1024]
  • blk.14.ffn_down.weight
    I32
    [14336, 4096]
  • blk.14.ffn_gate.weight
    I32
    [4096, 14336]
  • blk.14.ffn_norm.weight
    F32
    [4096]
  • blk.14.ffn_up.weight
    I32
    [4096, 14336]
    • blk.15
  • blk.15.attn_k.weight
    (!unknown_type 22!)
    [4096, 1024]
  • blk.15.attn_norm.weight
    F32
    [4096]
  • blk.15.attn_output.weight
    (!unknown_type 21!)
    [4096, 4096]
  • blk.15.attn_q.weight
    (!unknown_type 22!)
    [4096, 4096]
  • blk.15.attn_v.weight
    Q4_K
    [4096, 1024]
  • blk.15.ffn_down.weight
    I32
    [14336, 4096]
  • blk.15.ffn_gate.weight
    I32
    [4096, 14336]
  • blk.15.ffn_norm.weight
    F32
    [4096]
  • blk.15.ffn_up.weight
    I32
    [4096, 14336]
    • blk.16
  • blk.16.attn_k.weight
    (!unknown_type 22!)
    [4096, 1024]
  • blk.16.attn_norm.weight
    F32
    [4096]
  • blk.16.attn_output.weight
    (!unknown_type 21!)
    [4096, 4096]
  • blk.16.attn_q.weight
    (!unknown_type 22!)
    [4096, 4096]
  • blk.16.attn_v.weight
    Q4_K
    [4096, 1024]
  • blk.16.ffn_down.weight
    I32
    [14336, 4096]
  • blk.16.ffn_gate.weight
    I32
    [4096, 14336]
  • blk.16.ffn_norm.weight
    F32
    [4096]
  • blk.16.ffn_up.weight
    I32
    [4096, 14336]
    • blk.17
  • blk.17.attn_k.weight
    (!unknown_type 22!)
    [4096, 1024]
  • blk.17.attn_norm.weight
    F32
    [4096]
  • blk.17.attn_output.weight
    (!unknown_type 21!)
    [4096, 4096]
  • blk.17.attn_q.weight
    (!unknown_type 22!)
    [4096, 4096]
  • blk.17.attn_v.weight
    Q4_K
    [4096, 1024]
  • blk.17.ffn_down.weight
    I32
    [14336, 4096]
  • blk.17.ffn_gate.weight
    I32
    [4096, 14336]
  • blk.17.ffn_norm.weight
    F32
    [4096]
  • blk.17.ffn_up.weight
    I32
    [4096, 14336]
    • blk.18
  • blk.18.attn_k.weight
    (!unknown_type 22!)
    [4096, 1024]
  • blk.18.attn_norm.weight
    F32
    [4096]
  • blk.18.attn_output.weight
    (!unknown_type 21!)
    [4096, 4096]
  • blk.18.attn_q.weight
    (!unknown_type 22!)
    [4096, 4096]
  • blk.18.attn_v.weight
    Q4_K
    [4096, 1024]
  • blk.18.ffn_down.weight
    I32
    [14336, 4096]
  • blk.18.ffn_gate.weight
    I32
    [4096, 14336]
  • blk.18.ffn_norm.weight
    F32
    [4096]
  • blk.18.ffn_up.weight
    I32
    [4096, 14336]
    • blk.19
  • blk.19.attn_k.weight
    (!unknown_type 22!)
    [4096, 1024]
  • blk.19.attn_norm.weight
    F32
    [4096]
  • blk.19.attn_output.weight
    (!unknown_type 21!)
    [4096, 4096]
  • blk.19.attn_q.weight
    (!unknown_type 22!)
    [4096, 4096]
  • blk.19.attn_v.weight
    Q4_K
    [4096, 1024]
  • blk.19.ffn_down.weight
    I32
    [14336, 4096]
  • blk.19.ffn_gate.weight
    I32
    [4096, 14336]
  • blk.19.ffn_norm.weight
    F32
    [4096]
  • blk.19.ffn_up.weight
    I32
    [4096, 14336]
    • blk.20
  • blk.20.attn_k.weight
    (!unknown_type 22!)
    [4096, 1024]
  • blk.20.attn_norm.weight
    F32
    [4096]
  • blk.20.attn_output.weight
    (!unknown_type 21!)
    [4096, 4096]
  • blk.20.attn_q.weight
    (!unknown_type 22!)
    [4096, 4096]
  • blk.20.attn_v.weight
    Q4_K
    [4096, 1024]
  • blk.20.ffn_down.weight
    I32
    [14336, 4096]
  • blk.20.ffn_gate.weight
    I32
    [4096, 14336]
  • blk.20.ffn_norm.weight
    F32
    [4096]
  • blk.20.ffn_up.weight
    I32
    [4096, 14336]
    • blk.21
  • blk.21.attn_k.weight
    (!unknown_type 22!)
    [4096, 1024]
  • blk.21.attn_norm.weight
    F32
    [4096]
  • blk.21.attn_output.weight
    (!unknown_type 21!)
    [4096, 4096]
  • blk.21.attn_q.weight
    (!unknown_type 22!)
    [4096, 4096]
  • blk.21.attn_v.weight
    Q4_K
    [4096, 1024]
  • blk.21.ffn_down.weight
    I32
    [14336, 4096]
  • blk.21.ffn_gate.weight
    I32
    [4096, 14336]
  • blk.21.ffn_norm.weight
    F32
    [4096]
  • blk.21.ffn_up.weight
    I32
    [4096, 14336]
    • blk.22
  • blk.22.attn_k.weight
    (!unknown_type 22!)
    [4096, 1024]
  • blk.22.attn_norm.weight
    F32
    [4096]
  • blk.22.attn_output.weight
    (!unknown_type 21!)
    [4096, 4096]
  • blk.22.attn_q.weight
    (!unknown_type 22!)
    [4096, 4096]
  • blk.22.attn_v.weight
    Q4_K
    [4096, 1024]
  • blk.22.ffn_down.weight
    I32
    [14336, 4096]
  • blk.22.ffn_gate.weight
    I32
    [4096, 14336]
  • blk.22.ffn_norm.weight
    F32
    [4096]
  • blk.22.ffn_up.weight
    I32
    [4096, 14336]
    • blk.23
  • blk.23.attn_k.weight
    (!unknown_type 22!)
    [4096, 1024]
  • blk.23.attn_norm.weight
    F32
    [4096]
  • blk.23.attn_output.weight
    (!unknown_type 21!)
    [4096, 4096]
  • blk.23.attn_q.weight
    (!unknown_type 22!)
    [4096, 4096]
  • blk.23.attn_v.weight
    Q4_K
    [4096, 1024]
  • blk.23.ffn_down.weight
    I32
    [14336, 4096]
  • blk.23.ffn_gate.weight
    I32
    [4096, 14336]
  • blk.23.ffn_norm.weight
    F32
    [4096]
  • blk.23.ffn_up.weight
    I32
    [4096, 14336]
    • blk.24
  • blk.24.attn_k.weight
    (!unknown_type 22!)
    [4096, 1024]
  • blk.24.attn_norm.weight
    F32
    [4096]
  • blk.24.attn_output.weight
    (!unknown_type 21!)
    [4096, 4096]
  • blk.24.attn_q.weight
    (!unknown_type 22!)
    [4096, 4096]
  • blk.24.attn_v.weight
    Q4_K
    [4096, 1024]
  • blk.24.ffn_down.weight
    I32
    [14336, 4096]
  • blk.24.ffn_gate.weight
    I32
    [4096, 14336]
  • blk.24.ffn_norm.weight
    F32
    [4096]
  • blk.24.ffn_up.weight
    I32
    [4096, 14336]
    • blk.25
  • blk.25.attn_k.weight
    (!unknown_type 22!)
    [4096, 1024]
  • blk.25.attn_norm.weight
    F32
    [4096]
  • blk.25.attn_output.weight
    (!unknown_type 21!)
    [4096, 4096]
  • blk.25.attn_q.weight
    (!unknown_type 22!)
    [4096, 4096]
  • blk.25.attn_v.weight
    Q4_K
    [4096, 1024]
  • blk.25.ffn_down.weight
    I32
    [14336, 4096]
  • blk.25.ffn_gate.weight
    I32
    [4096, 14336]
  • blk.25.ffn_norm.weight
    F32
    [4096]
  • blk.25.ffn_up.weight
    I32
    [4096, 14336]
    • blk.26
  • blk.26.attn_k.weight
    (!unknown_type 22!)
    [4096, 1024]
  • blk.26.attn_norm.weight
    F32
    [4096]
  • blk.26.attn_output.weight
    (!unknown_type 21!)
    [4096, 4096]
  • blk.26.attn_q.weight
    (!unknown_type 22!)
    [4096, 4096]
  • blk.26.attn_v.weight
    Q4_K
    [4096, 1024]
  • blk.26.ffn_down.weight
    I32
    [14336, 4096]
  • blk.26.ffn_gate.weight
    I32
    [4096, 14336]
  • blk.26.ffn_norm.weight
    F32
    [4096]
  • blk.26.ffn_up.weight
    I32
    [4096, 14336]
    • blk.27
  • blk.27.attn_k.weight
    (!unknown_type 22!)
    [4096, 1024]
  • blk.27.attn_norm.weight
    F32
    [4096]
  • blk.27.attn_output.weight
    (!unknown_type 21!)
    [4096, 4096]
  • blk.27.attn_q.weight
    (!unknown_type 22!)
    [4096, 4096]
  • blk.27.attn_v.weight
    Q4_K
    [4096, 1024]
  • blk.27.ffn_down.weight
    I32
    [14336, 4096]
  • blk.27.ffn_gate.weight
    I32
    [4096, 14336]
  • blk.27.ffn_norm.weight
    F32
    [4096]
  • blk.27.ffn_up.weight
    I32
    [4096, 14336]
    • blk.28
  • blk.28.attn_k.weight
    (!unknown_type 22!)
    [4096, 1024]
  • blk.28.attn_norm.weight
    F32
    [4096]
  • blk.28.attn_output.weight
    (!unknown_type 21!)
    [4096, 4096]
  • blk.28.attn_q.weight
    (!unknown_type 22!)
    [4096, 4096]
  • blk.28.attn_v.weight
    Q4_K
    [4096, 1024]
  • blk.28.ffn_down.weight
    I32
    [14336, 4096]
  • blk.28.ffn_gate.weight
    I32
    [4096, 14336]
  • blk.28.ffn_norm.weight
    F32
    [4096]
  • blk.28.ffn_up.weight
    I32
    [4096, 14336]
    • blk.29
  • blk.29.attn_k.weight
    (!unknown_type 22!)
    [4096, 1024]
  • blk.29.attn_norm.weight
    F32
    [4096]
  • blk.29.attn_output.weight
    (!unknown_type 21!)
    [4096, 4096]
  • blk.29.attn_q.weight
    (!unknown_type 22!)
    [4096, 4096]
  • blk.29.attn_v.weight
    Q4_K
    [4096, 1024]
  • blk.29.ffn_down.weight
    I32
    [14336, 4096]
  • blk.29.ffn_gate.weight
    I32
    [4096, 14336]
  • blk.29.ffn_norm.weight
    F32
    [4096]
  • blk.29.ffn_up.weight
    I32
    [4096, 14336]
    • blk.30
  • blk.30.attn_k.weight
    (!unknown_type 22!)
    [4096, 1024]
  • blk.30.attn_norm.weight
    F32
    [4096]
  • blk.30.attn_output.weight
    (!unknown_type 21!)
    [4096, 4096]
  • blk.30.attn_q.weight
    (!unknown_type 22!)
    [4096, 4096]
  • blk.30.attn_v.weight
    Q4_K
    [4096, 1024]
  • blk.30.ffn_down.weight
    I32
    [14336, 4096]
  • blk.30.ffn_gate.weight
    I32
    [4096, 14336]
  • blk.30.ffn_norm.weight
    F32
    [4096]
  • blk.30.ffn_up.weight
    I32
    [4096, 14336]
    • blk.31
  • blk.31.attn_k.weight
    (!unknown_type 22!)
    [4096, 1024]
  • blk.31.attn_norm.weight
    F32
    [4096]
  • blk.31.attn_output.weight
    (!unknown_type 21!)
    [4096, 4096]
  • blk.31.attn_q.weight
    (!unknown_type 22!)
    [4096, 4096]
  • blk.31.attn_v.weight
    Q4_K
    [4096, 1024]
  • blk.31.ffn_down.weight
    I32
    [14336, 4096]
  • blk.31.ffn_gate.weight
    I32
    [4096, 14336]
  • blk.31.ffn_norm.weight
    F32
    [4096]
  • blk.31.ffn_up.weight
    I32
    [4096, 14336]
  • output.weight
    Q5_K
    [4096, 128256]
  • output_norm.weight
    F32
    [4096]