frob/ mimo-v2.6-pro:1023b-a42b-mxfp4

2 6 hours ago

vision tools thinking 1023b
4f21c653e953 · 2.8GB
    Metadata
  • general.architecture
    clip
  • general.file_type
    BF16
  • clip.audio.attention.head_count
    16
  • clip.audio.attention.layer_norm_epsilon
    1e-05
  • clip.audio.block_count
    24
  • clip.audio.embedding_length
    1024
  • clip.audio.feed_forward_length
    4096
  • clip.audio.local_block_count
    6
  • clip.audio.local_group_size
    4
  • clip.audio.num_mel_bins
    128
  • clip.audio.projection_dim
    6144
  • clip.audio.projector_type
    mimo_audio
  • clip.audio.rvq.codebook_size
    [1024, 1024, 256, 128, 128, ...]
  • clip.audio.rvq.num_quantizers
    20
  • clip.audio.wa_pattern_mode
    [0, -1, 0, -1, 0, ...]
  • clip.audio.window_size
    128
  • clip.has_audio_encoder
    true
  • clip.has_vision_encoder
    true
  • clip.use_silu
    true
  • clip.vision.attention.head_count
    32
  • clip.vision.attention.head_count_kv
    8
  • clip.vision.attention.layer_norm_epsilon
    1e-06
  • clip.vision.block_count
    28
  • clip.vision.embedding_length
    1280
  • clip.vision.feed_forward_length
    4608
  • clip.vision.image_max_pixels
    12845056
  • clip.vision.image_mean
    [0.48145467, 0.4578275, 0.40821072]
  • clip.vision.image_min_pixels
    3136
  • clip.vision.image_size
    560
  • clip.vision.image_std
    [0.26862955, 0.2613026, 0.2757771]
  • clip.vision.patch_size
    16
  • clip.vision.projection_dim
    6144
  • clip.vision.projector_type
    mimovl
  • clip.vision.spatial_merge_size
    2
  • clip.vision.wa_pattern_mode
    [-1, 0, 0, 0, 0, ...]
  • clip.vision.window_size
    64
  • Tensor
    a.blk.0
  • a.blk.0.attn_k.weight
    BF16
    [1024, 1024]
  • a.blk.0.attn_out.bias
    F32
    [1024]
  • a.blk.0.attn_out.weight
    BF16
    [1024, 1024]
  • a.blk.0.attn_q.bias
    F32
    [1024]
  • a.blk.0.attn_q.weight
    BF16
    [1024, 1024]
  • a.blk.0.attn_v.bias
    F32
    [1024]
  • a.blk.0.attn_v.weight
    BF16
    [1024, 1024]
  • a.blk.0.ffn_down.bias
    F32
    [1024]
  • a.blk.0.ffn_down.weight
    BF16
    [4096, 1024]
  • a.blk.0.ffn_up.bias
    F32
    [4096]
  • a.blk.0.ffn_up.weight
    BF16
    [1024, 4096]
  • a.blk.0.ln1.bias
    F32
    [1024]
  • a.blk.0.ln1.weight
    F32
    [1024]
  • a.blk.0.ln2.bias
    F32
    [1024]
  • a.blk.0.ln2.weight
    F32
    [1024]
  • a.blk.1
  • a.blk.1.attn_k.weight
    BF16
    [1024, 1024]
  • a.blk.1.attn_out.bias
    F32
    [1024]
  • a.blk.1.attn_out.weight
    BF16
    [1024, 1024]
  • a.blk.1.attn_q.bias
    F32
    [1024]
  • a.blk.1.attn_q.weight
    BF16
    [1024, 1024]
  • a.blk.1.attn_v.bias
    F32
    [1024]
  • a.blk.1.attn_v.weight
    BF16
    [1024, 1024]
  • a.blk.1.ffn_down.bias
    F32
    [1024]
  • a.blk.1.ffn_down.weight
    BF16
    [4096, 1024]
  • a.blk.1.ffn_up.bias
    F32
    [4096]
  • a.blk.1.ffn_up.weight
    BF16
    [1024, 4096]
  • a.blk.1.ln1.bias
    F32
    [1024]
  • a.blk.1.ln1.weight
    F32
    [1024]
  • a.blk.1.ln2.bias
    F32
    [1024]
  • a.blk.1.ln2.weight
    F32
    [1024]
  • a.blk.2
  • a.blk.2.attn_k.weight
    BF16
    [1024, 1024]
  • a.blk.2.attn_out.bias
    F32
    [1024]
  • a.blk.2.attn_out.weight
    BF16
    [1024, 1024]
  • a.blk.2.attn_q.bias
    F32
    [1024]
  • a.blk.2.attn_q.weight
    BF16
    [1024, 1024]
  • a.blk.2.attn_v.bias
    F32
    [1024]
  • a.blk.2.attn_v.weight
    BF16
    [1024, 1024]
  • a.blk.2.ffn_down.bias
    F32
    [1024]
  • a.blk.2.ffn_down.weight
    BF16
    [4096, 1024]
  • a.blk.2.ffn_up.bias
    F32
    [4096]
  • a.blk.2.ffn_up.weight
    BF16
    [1024, 4096]
  • a.blk.2.ln1.bias
    F32
    [1024]
  • a.blk.2.ln1.weight
    F32
    [1024]
  • a.blk.2.ln2.bias
    F32
    [1024]
  • a.blk.2.ln2.weight
    F32
    [1024]
  • a.blk.3
  • a.blk.3.attn_k.weight
    BF16
    [1024, 1024]
  • a.blk.3.attn_out.bias
    F32
    [1024]
  • a.blk.3.attn_out.weight
    BF16
    [1024, 1024]
  • a.blk.3.attn_q.bias
    F32
    [1024]
  • a.blk.3.attn_q.weight
    BF16
    [1024, 1024]
  • a.blk.3.attn_v.bias
    F32
    [1024]
  • a.blk.3.attn_v.weight
    BF16
    [1024, 1024]
  • a.blk.3.ffn_down.bias
    F32
    [1024]
  • a.blk.3.ffn_down.weight
    BF16
    [4096, 1024]
  • a.blk.3.ffn_up.bias
    F32
    [4096]
  • a.blk.3.ffn_up.weight
    BF16
    [1024, 4096]
  • a.blk.3.ln1.bias
    F32
    [1024]
  • a.blk.3.ln1.weight
    F32
    [1024]
  • a.blk.3.ln2.bias
    F32
    [1024]
  • a.blk.3.ln2.weight
    F32
    [1024]
  • a.blk.4
  • a.blk.4.attn_k.weight
    BF16
    [1024, 1024]
  • a.blk.4.attn_out.bias
    F32
    [1024]
  • a.blk.4.attn_out.weight
    BF16
    [1024, 1024]
  • a.blk.4.attn_q.bias
    F32
    [1024]
  • a.blk.4.attn_q.weight
    BF16
    [1024, 1024]
  • a.blk.4.attn_v.bias
    F32
    [1024]
  • a.blk.4.attn_v.weight
    BF16
    [1024, 1024]
  • a.blk.4.ffn_down.bias
    F32
    [1024]
  • a.blk.4.ffn_down.weight
    BF16
    [4096, 1024]
  • a.blk.4.ffn_up.bias
    F32
    [4096]
  • a.blk.4.ffn_up.weight
    BF16
    [1024, 4096]
  • a.blk.4.ln1.bias
    F32
    [1024]
  • a.blk.4.ln1.weight
    F32
    [1024]
  • a.blk.4.ln2.bias
    F32
    [1024]
  • a.blk.4.ln2.weight
    F32
    [1024]
  • a.blk.5
  • a.blk.5.attn_k.weight
    BF16
    [1024, 1024]
  • a.blk.5.attn_out.bias
    F32
    [1024]
  • a.blk.5.attn_out.weight
    BF16
    [1024, 1024]
  • a.blk.5.attn_q.bias
    F32
    [1024]
  • a.blk.5.attn_q.weight
    BF16
    [1024, 1024]
  • a.blk.5.attn_v.bias
    F32
    [1024]
  • a.blk.5.attn_v.weight
    BF16
    [1024, 1024]
  • a.blk.5.ffn_down.bias
    F32
    [1024]
  • a.blk.5.ffn_down.weight
    BF16
    [4096, 1024]
  • a.blk.5.ffn_up.bias
    F32
    [4096]
  • a.blk.5.ffn_up.weight
    BF16
    [1024, 4096]
  • a.blk.5.ln1.bias
    F32
    [1024]
  • a.blk.5.ln1.weight
    F32
    [1024]
  • a.blk.5.ln2.bias
    F32
    [1024]
  • a.blk.5.ln2.weight
    F32
    [1024]
  • a.blk.6
  • a.blk.6.attn_k.weight
    BF16
    [1024, 1024]
  • a.blk.6.attn_out.bias
    F32
    [1024]
  • a.blk.6.attn_out.weight
    BF16
    [1024, 1024]
  • a.blk.6.attn_q.bias
    F32
    [1024]
  • a.blk.6.attn_q.weight
    BF16
    [1024, 1024]
  • a.blk.6.attn_v.bias
    F32
    [1024]
  • a.blk.6.attn_v.weight
    BF16
    [1024, 1024]
  • a.blk.6.ffn_down.bias
    F32
    [1024]
  • a.blk.6.ffn_down.weight
    BF16
    [4096, 1024]
  • a.blk.6.ffn_up.bias
    F32
    [4096]
  • a.blk.6.ffn_up.weight
    BF16
    [1024, 4096]
  • a.blk.6.ln1.bias
    F32
    [1024]
  • a.blk.6.ln1.weight
    F32
    [1024]
  • a.blk.6.ln2.bias
    F32
    [1024]
  • a.blk.6.ln2.weight
    F32
    [1024]
  • a.blk.7
  • a.blk.7.attn_k.weight
    BF16
    [1024, 1024]
  • a.blk.7.attn_out.bias
    F32
    [1024]
  • a.blk.7.attn_out.weight
    BF16
    [1024, 1024]
  • a.blk.7.attn_q.bias
    F32
    [1024]
  • a.blk.7.attn_q.weight
    BF16
    [1024, 1024]
  • a.blk.7.attn_v.bias
    F32
    [1024]
  • a.blk.7.attn_v.weight
    BF16
    [1024, 1024]
  • a.blk.7.ffn_down.bias
    F32
    [1024]
  • a.blk.7.ffn_down.weight
    BF16
    [4096, 1024]
  • a.blk.7.ffn_up.bias
    F32
    [4096]
  • a.blk.7.ffn_up.weight
    BF16
    [1024, 4096]
  • a.blk.7.ln1.bias
    F32
    [1024]
  • a.blk.7.ln1.weight
    F32
    [1024]
  • a.blk.7.ln2.bias
    F32
    [1024]
  • a.blk.7.ln2.weight
    F32
    [1024]
  • a.blk.8
  • a.blk.8.attn_k.weight
    BF16
    [1024, 1024]
  • a.blk.8.attn_out.bias
    F32
    [1024]
  • a.blk.8.attn_out.weight
    BF16
    [1024, 1024]
  • a.blk.8.attn_q.bias
    F32
    [1024]
  • a.blk.8.attn_q.weight
    BF16
    [1024, 1024]
  • a.blk.8.attn_v.bias
    F32
    [1024]
  • a.blk.8.attn_v.weight
    BF16
    [1024, 1024]
  • a.blk.8.ffn_down.bias
    F32
    [1024]
  • a.blk.8.ffn_down.weight
    BF16
    [4096, 1024]
  • a.blk.8.ffn_up.bias
    F32
    [4096]
  • a.blk.8.ffn_up.weight
    BF16
    [1024, 4096]
  • a.blk.8.ln1.bias
    F32
    [1024]
  • a.blk.8.ln1.weight
    F32
    [1024]
  • a.blk.8.ln2.bias
    F32
    [1024]
  • a.blk.8.ln2.weight
    F32
    [1024]
  • a.blk.9
  • a.blk.9.attn_k.weight
    BF16
    [1024, 1024]
  • a.blk.9.attn_out.bias
    F32
    [1024]
  • a.blk.9.attn_out.weight
    BF16
    [1024, 1024]
  • a.blk.9.attn_q.bias
    F32
    [1024]
  • a.blk.9.attn_q.weight
    BF16
    [1024, 1024]
  • a.blk.9.attn_v.bias
    F32
    [1024]
  • a.blk.9.attn_v.weight
    BF16
    [1024, 1024]
  • a.blk.9.ffn_down.bias
    F32
    [1024]
  • a.blk.9.ffn_down.weight
    BF16
    [4096, 1024]
  • a.blk.9.ffn_up.bias
    F32
    [4096]
  • a.blk.9.ffn_up.weight
    BF16
    [1024, 4096]
  • a.blk.9.ln1.bias
    F32
    [1024]
  • a.blk.9.ln1.weight
    F32
    [1024]
  • a.blk.9.ln2.bias
    F32
    [1024]
  • a.blk.9.ln2.weight
    F32
    [1024]
  • a.blk.10
  • a.blk.10.attn_k.weight
    BF16
    [1024, 1024]
  • a.blk.10.attn_out.bias
    F32
    [1024]
  • a.blk.10.attn_out.weight
    BF16
    [1024, 1024]
  • a.blk.10.attn_q.bias
    F32
    [1024]
  • a.blk.10.attn_q.weight
    BF16
    [1024, 1024]
  • a.blk.10.attn_v.bias
    F32
    [1024]
  • a.blk.10.attn_v.weight
    BF16
    [1024, 1024]
  • a.blk.10.ffn_down.bias
    F32
    [1024]
  • a.blk.10.ffn_down.weight
    BF16
    [4096, 1024]
  • a.blk.10.ffn_up.bias
    F32
    [4096]
  • a.blk.10.ffn_up.weight
    BF16
    [1024, 4096]
  • a.blk.10.ln1.bias
    F32
    [1024]
  • a.blk.10.ln1.weight
    F32
    [1024]
  • a.blk.10.ln2.bias
    F32
    [1024]
  • a.blk.10.ln2.weight
    F32
    [1024]
  • a.blk.11
  • a.blk.11.attn_k.weight
    BF16
    [1024, 1024]
  • a.blk.11.attn_out.bias
    F32
    [1024]
  • a.blk.11.attn_out.weight
    BF16
    [1024, 1024]
  • a.blk.11.attn_q.bias
    F32
    [1024]
  • a.blk.11.attn_q.weight
    BF16
    [1024, 1024]
  • a.blk.11.attn_v.bias
    F32
    [1024]
  • a.blk.11.attn_v.weight
    BF16
    [1024, 1024]
  • a.blk.11.ffn_down.bias
    F32
    [1024]
  • a.blk.11.ffn_down.weight
    BF16
    [4096, 1024]
  • a.blk.11.ffn_up.bias
    F32
    [4096]
  • a.blk.11.ffn_up.weight
    BF16
    [1024, 4096]
  • a.blk.11.ln1.bias
    F32
    [1024]
  • a.blk.11.ln1.weight
    F32
    [1024]
  • a.blk.11.ln2.bias
    F32
    [1024]
  • a.blk.11.ln2.weight
    F32
    [1024]
  • a.blk.12
  • a.blk.12.attn_k.weight
    BF16
    [1024, 1024]
  • a.blk.12.attn_out.bias
    F32
    [1024]
  • a.blk.12.attn_out.weight
    BF16
    [1024, 1024]
  • a.blk.12.attn_q.bias
    F32
    [1024]
  • a.blk.12.attn_q.weight
    BF16
    [1024, 1024]
  • a.blk.12.attn_v.bias
    F32
    [1024]
  • a.blk.12.attn_v.weight
    BF16
    [1024, 1024]
  • a.blk.12.ffn_down.bias
    F32
    [1024]
  • a.blk.12.ffn_down.weight
    BF16
    [4096, 1024]
  • a.blk.12.ffn_up.bias
    F32
    [4096]
  • a.blk.12.ffn_up.weight
    BF16
    [1024, 4096]
  • a.blk.12.ln1.bias
    F32
    [1024]
  • a.blk.12.ln1.weight
    F32
    [1024]
  • a.blk.12.ln2.bias
    F32
    [1024]
  • a.blk.12.ln2.weight
    F32
    [1024]
  • a.blk.13
  • a.blk.13.attn_k.weight
    BF16
    [1024, 1024]
  • a.blk.13.attn_out.bias
    F32
    [1024]
  • a.blk.13.attn_out.weight
    BF16
    [1024, 1024]
  • a.blk.13.attn_q.bias
    F32
    [1024]
  • a.blk.13.attn_q.weight
    BF16
    [1024, 1024]
  • a.blk.13.attn_v.bias
    F32
    [1024]
  • a.blk.13.attn_v.weight
    BF16
    [1024, 1024]
  • a.blk.13.ffn_down.bias
    F32
    [1024]
  • a.blk.13.ffn_down.weight
    BF16
    [4096, 1024]
  • a.blk.13.ffn_up.bias
    F32
    [4096]
  • a.blk.13.ffn_up.weight
    BF16
    [1024, 4096]
  • a.blk.13.ln1.bias
    F32
    [1024]
  • a.blk.13.ln1.weight
    F32
    [1024]
  • a.blk.13.ln2.bias
    F32
    [1024]
  • a.blk.13.ln2.weight
    F32
    [1024]
  • a.blk.14
  • a.blk.14.attn_k.weight
    BF16
    [1024, 1024]
  • a.blk.14.attn_out.bias
    F32
    [1024]
  • a.blk.14.attn_out.weight
    BF16
    [1024, 1024]
  • a.blk.14.attn_q.bias
    F32
    [1024]
  • a.blk.14.attn_q.weight
    BF16
    [1024, 1024]
  • a.blk.14.attn_v.bias
    F32
    [1024]
  • a.blk.14.attn_v.weight
    BF16
    [1024, 1024]
  • a.blk.14.ffn_down.bias
    F32
    [1024]
  • a.blk.14.ffn_down.weight
    BF16
    [4096, 1024]
  • a.blk.14.ffn_up.bias
    F32
    [4096]
  • a.blk.14.ffn_up.weight
    BF16
    [1024, 4096]
  • a.blk.14.ln1.bias
    F32
    [1024]
  • a.blk.14.ln1.weight
    F32
    [1024]
  • a.blk.14.ln2.bias
    F32
    [1024]
  • a.blk.14.ln2.weight
    F32
    [1024]
  • a.blk.15
  • a.blk.15.attn_k.weight
    BF16
    [1024, 1024]
  • a.blk.15.attn_out.bias
    F32
    [1024]
  • a.blk.15.attn_out.weight
    BF16
    [1024, 1024]
  • a.blk.15.attn_q.bias
    F32
    [1024]
  • a.blk.15.attn_q.weight
    BF16
    [1024, 1024]
  • a.blk.15.attn_v.bias
    F32
    [1024]
  • a.blk.15.attn_v.weight
    BF16
    [1024, 1024]
  • a.blk.15.ffn_down.bias
    F32
    [1024]
  • a.blk.15.ffn_down.weight
    BF16
    [4096, 1024]
  • a.blk.15.ffn_up.bias
    F32
    [4096]
  • a.blk.15.ffn_up.weight
    BF16
    [1024, 4096]
  • a.blk.15.ln1.bias
    F32
    [1024]
  • a.blk.15.ln1.weight
    F32
    [1024]
  • a.blk.15.ln2.bias
    F32
    [1024]
  • a.blk.15.ln2.weight
    F32
    [1024]
  • a.blk.16
  • a.blk.16.attn_k.weight
    BF16
    [1024, 1024]
  • a.blk.16.attn_out.bias
    F32
    [1024]
  • a.blk.16.attn_out.weight
    BF16
    [1024, 1024]
  • a.blk.16.attn_q.bias
    F32
    [1024]
  • a.blk.16.attn_q.weight
    BF16
    [1024, 1024]
  • a.blk.16.attn_v.bias
    F32
    [1024]
  • a.blk.16.attn_v.weight
    BF16
    [1024, 1024]
  • a.blk.16.ffn_down.bias
    F32
    [1024]
  • a.blk.16.ffn_down.weight
    BF16
    [4096, 1024]
  • a.blk.16.ffn_up.bias
    F32
    [4096]
  • a.blk.16.ffn_up.weight
    BF16
    [1024, 4096]
  • a.blk.16.ln1.bias
    F32
    [1024]
  • a.blk.16.ln1.weight
    F32
    [1024]
  • a.blk.16.ln2.bias
    F32
    [1024]
  • a.blk.16.ln2.weight
    F32
    [1024]
  • a.blk.17
  • a.blk.17.attn_k.weight
    BF16
    [1024, 1024]
  • a.blk.17.attn_out.bias
    F32
    [1024]
  • a.blk.17.attn_out.weight
    BF16
    [1024, 1024]
  • a.blk.17.attn_q.bias
    F32
    [1024]
  • a.blk.17.attn_q.weight
    BF16
    [1024, 1024]
  • a.blk.17.attn_v.bias
    F32
    [1024]
  • a.blk.17.attn_v.weight
    BF16
    [1024, 1024]
  • a.blk.17.ffn_down.bias
    F32
    [1024]
  • a.blk.17.ffn_down.weight
    BF16
    [4096, 1024]
  • a.blk.17.ffn_up.bias
    F32
    [4096]
  • a.blk.17.ffn_up.weight
    BF16
    [1024, 4096]
  • a.blk.17.ln1.bias
    F32
    [1024]
  • a.blk.17.ln1.weight
    F32
    [1024]
  • a.blk.17.ln2.bias
    F32
    [1024]
  • a.blk.17.ln2.weight
    F32
    [1024]
  • a.blk.18
  • a.blk.18.attn_k.weight
    BF16
    [1024, 1024]
  • a.blk.18.attn_out.bias
    F32
    [1024]
  • a.blk.18.attn_out.weight
    BF16
    [1024, 1024]
  • a.blk.18.attn_q.bias
    F32
    [1024]
  • a.blk.18.attn_q.weight
    BF16
    [1024, 1024]
  • a.blk.18.attn_v.bias
    F32
    [1024]
  • a.blk.18.attn_v.weight
    BF16
    [1024, 1024]
  • a.blk.18.ffn_down.bias
    F32
    [1024]
  • a.blk.18.ffn_down.weight
    BF16
    [4096, 1024]
  • a.blk.18.ffn_up.bias
    F32
    [4096]
  • a.blk.18.ffn_up.weight
    BF16
    [1024, 4096]
  • a.blk.18.ln1.bias
    F32
    [1024]
  • a.blk.18.ln1.weight
    F32
    [1024]
  • a.blk.18.ln2.bias
    F32
    [1024]
  • a.blk.18.ln2.weight
    F32
    [1024]
  • a.blk.19
  • a.blk.19.attn_k.weight
    BF16
    [1024, 1024]
  • a.blk.19.attn_out.bias
    F32
    [1024]
  • a.blk.19.attn_out.weight
    BF16
    [1024, 1024]
  • a.blk.19.attn_q.bias
    F32
    [1024]
  • a.blk.19.attn_q.weight
    BF16
    [1024, 1024]
  • a.blk.19.attn_v.bias
    F32
    [1024]
  • a.blk.19.attn_v.weight
    BF16
    [1024, 1024]
  • a.blk.19.ffn_down.bias
    F32
    [1024]
  • a.blk.19.ffn_down.weight
    BF16
    [4096, 1024]
  • a.blk.19.ffn_up.bias
    F32
    [4096]
  • a.blk.19.ffn_up.weight
    BF16
    [1024, 4096]
  • a.blk.19.ln1.bias
    F32
    [1024]
  • a.blk.19.ln1.weight
    F32
    [1024]
  • a.blk.19.ln2.bias
    F32
    [1024]
  • a.blk.19.ln2.weight
    F32
    [1024]
  • a.blk.20
  • a.blk.20.attn_k.weight
    BF16
    [1024, 1024]
  • a.blk.20.attn_out.bias
    F32
    [1024]
  • a.blk.20.attn_out.weight
    BF16
    [1024, 1024]
  • a.blk.20.attn_q.bias
    F32
    [1024]
  • a.blk.20.attn_q.weight
    BF16
    [1024, 1024]
  • a.blk.20.attn_v.bias
    F32
    [1024]
  • a.blk.20.attn_v.weight
    BF16
    [1024, 1024]
  • a.blk.20.ffn_down.bias
    F32
    [1024]
  • a.blk.20.ffn_down.weight
    BF16
    [4096, 1024]
  • a.blk.20.ffn_up.bias
    F32
    [4096]
  • a.blk.20.ffn_up.weight
    BF16
    [1024, 4096]
  • a.blk.20.ln1.bias
    F32
    [1024]
  • a.blk.20.ln1.weight
    F32
    [1024]
  • a.blk.20.ln2.bias
    F32
    [1024]
  • a.blk.20.ln2.weight
    F32
    [1024]
  • a.blk.21
  • a.blk.21.attn_k.weight
    BF16
    [1024, 1024]
  • a.blk.21.attn_out.bias
    F32
    [1024]
  • a.blk.21.attn_out.weight
    BF16
    [1024, 1024]
  • a.blk.21.attn_q.bias
    F32
    [1024]
  • a.blk.21.attn_q.weight
    BF16
    [1024, 1024]
  • a.blk.21.attn_v.bias
    F32
    [1024]
  • a.blk.21.attn_v.weight
    BF16
    [1024, 1024]
  • a.blk.21.ffn_down.bias
    F32
    [1024]
  • a.blk.21.ffn_down.weight
    BF16
    [4096, 1024]
  • a.blk.21.ffn_up.bias
    F32
    [4096]
  • a.blk.21.ffn_up.weight
    BF16
    [1024, 4096]
  • a.blk.21.ln1.bias
    F32
    [1024]
  • a.blk.21.ln1.weight
    F32
    [1024]
  • a.blk.21.ln2.bias
    F32
    [1024]
  • a.blk.21.ln2.weight
    F32
    [1024]
  • a.blk.22
  • a.blk.22.attn_k.weight
    BF16
    [1024, 1024]
  • a.blk.22.attn_out.bias
    F32
    [1024]
  • a.blk.22.attn_out.weight
    BF16
    [1024, 1024]
  • a.blk.22.attn_q.bias
    F32
    [1024]
  • a.blk.22.attn_q.weight
    BF16
    [1024, 1024]
  • a.blk.22.attn_v.bias
    F32
    [1024]
  • a.blk.22.attn_v.weight
    BF16
    [1024, 1024]
  • a.blk.22.ffn_down.bias
    F32
    [1024]
  • a.blk.22.ffn_down.weight
    BF16
    [4096, 1024]
  • a.blk.22.ffn_up.bias
    F32
    [4096]
  • a.blk.22.ffn_up.weight
    BF16
    [1024, 4096]
  • a.blk.22.ln1.bias
    F32
    [1024]
  • a.blk.22.ln1.weight
    F32
    [1024]
  • a.blk.22.ln2.bias
    F32
    [1024]
  • a.blk.22.ln2.weight
    F32
    [1024]
  • a.blk.23
  • a.blk.23.attn_k.weight
    BF16
    [1024, 1024]
  • a.blk.23.attn_out.bias
    F32
    [1024]
  • a.blk.23.attn_out.weight
    BF16
    [1024, 1024]
  • a.blk.23.attn_q.bias
    F32
    [1024]
  • a.blk.23.attn_q.weight
    BF16
    [1024, 1024]
  • a.blk.23.attn_v.bias
    F32
    [1024]
  • a.blk.23.attn_v.weight
    BF16
    [1024, 1024]
  • a.blk.23.ffn_down.bias
    F32
    [1024]
  • a.blk.23.ffn_down.weight
    BF16
    [4096, 1024]
  • a.blk.23.ffn_up.bias
    F32
    [4096]
  • a.blk.23.ffn_up.weight
    BF16
    [1024, 4096]
  • a.blk.23.ln1.bias
    F32
    [1024]
  • a.blk.23.ln1.weight
    F32
    [1024]
  • a.blk.23.ln2.bias
    F32
    [1024]
  • a.blk.23.ln2.weight
    F32
    [1024]
  • a.conv1d.1.bias
    F32
    [1, 1024]
  • a.conv1d.1.weight
    F32
    [3, 128, 1024]
  • a.conv1d.2.bias
    F32
    [1, 1024]
  • a.conv1d.2.weight
    F32
    [3, 1024, 1024]
  • a.downsample.conv.weight
    F32
    [2, 1024, 1024]
  • a.downsample.norm.bias
    F32
    [1024]
  • a.downsample.norm.weight
    F32
    [1024]
  • a.post_ln.bias
    F32
    [1024]
  • a.post_ln.weight
    F32
    [1024]
  • a.rvq.codebook.weight
    F32
    [1024, 1024, 20]
  • mm.0.weight
    BF16
    [5120, 5120]
  • mm.2.weight
    BF16
    [5120, 6144]
  • mm.a.code_embd.weight
    F32
    [1024, 1280, 20]
  • mm.a.local_blk.0.attn_k.bias
    F32
    [1024]
  • mm.a.local_blk.0.attn_k.weight
    BF16
    [1024, 1024]
  • mm.a.local_blk.0.attn_out.weight
    BF16
    [1024, 1024]
  • mm.a.local_blk.0.attn_q.bias
    F32
    [1024]
  • mm.a.local_blk.0.attn_q.weight
    BF16
    [1024, 1024]
  • mm.a.local_blk.0.attn_v.bias
    F32
    [1024]
  • mm.a.local_blk.0.attn_v.weight
    BF16
    [1024, 1024]
  • mm.a.local_blk.0.ffn_down.weight
    BF16
    [4096, 1024]
  • mm.a.local_blk.0.ffn_gate.weight
    BF16
    [1024, 4096]
  • mm.a.local_blk.0.ffn_up.weight
    BF16
    [1024, 4096]
  • mm.a.local_blk.0.ln1.weight
    F32
    [1024]
  • mm.a.local_blk.0.ln2.weight
    F32
    [1024]
  • mm.a.local_blk.1.attn_k.bias
    F32
    [1024]
  • mm.a.local_blk.1.attn_k.weight
    BF16
    [1024, 1024]
  • mm.a.local_blk.1.attn_out.weight
    BF16
    [1024, 1024]
  • mm.a.local_blk.1.attn_q.bias
    F32
    [1024]
  • mm.a.local_blk.1.attn_q.weight
    BF16
    [1024, 1024]
  • mm.a.local_blk.1.attn_v.bias
    F32
    [1024]
  • mm.a.local_blk.1.attn_v.weight
    BF16
    [1024, 1024]
  • mm.a.local_blk.1.ffn_down.weight
    BF16
    [4096, 1024]
  • mm.a.local_blk.1.ffn_gate.weight
    BF16
    [1024, 4096]
  • mm.a.local_blk.1.ffn_up.weight
    BF16
    [1024, 4096]
  • mm.a.local_blk.1.ln1.weight
    F32
    [1024]
  • mm.a.local_blk.1.ln2.weight
    F32
    [1024]
  • mm.a.local_blk.2.attn_k.bias
    F32
    [1024]
  • mm.a.local_blk.2.attn_k.weight
    BF16
    [1024, 1024]
  • mm.a.local_blk.2.attn_out.weight
    BF16
    [1024, 1024]
  • mm.a.local_blk.2.attn_q.bias
    F32
    [1024]
  • mm.a.local_blk.2.attn_q.weight
    BF16
    [1024, 1024]
  • mm.a.local_blk.2.attn_v.bias
    F32
    [1024]
  • mm.a.local_blk.2.attn_v.weight
    BF16
    [1024, 1024]
  • mm.a.local_blk.2.ffn_down.weight
    BF16
    [4096, 1024]
  • mm.a.local_blk.2.ffn_gate.weight
    BF16
    [1024, 4096]
  • mm.a.local_blk.2.ffn_up.weight
    BF16
    [1024, 4096]
  • mm.a.local_blk.2.ln1.weight
    F32
    [1024]
  • mm.a.local_blk.2.ln2.weight
    F32
    [1024]
  • mm.a.local_blk.3.attn_k.bias
    F32
    [1024]
  • mm.a.local_blk.3.attn_k.weight
    BF16
    [1024, 1024]
  • mm.a.local_blk.3.attn_out.weight
    BF16
    [1024, 1024]
  • mm.a.local_blk.3.attn_q.bias
    F32
    [1024]
  • mm.a.local_blk.3.attn_q.weight
    BF16
    [1024, 1024]
  • mm.a.local_blk.3.attn_v.bias
    F32
    [1024]
  • mm.a.local_blk.3.attn_v.weight
    BF16
    [1024, 1024]
  • mm.a.local_blk.3.ffn_down.weight
    BF16
    [4096, 1024]
  • mm.a.local_blk.3.ffn_gate.weight
    BF16
    [1024, 4096]
  • mm.a.local_blk.3.ffn_up.weight
    BF16
    [1024, 4096]
  • mm.a.local_blk.3.ln1.weight
    F32
    [1024]
  • mm.a.local_blk.3.ln2.weight
    F32
    [1024]
  • mm.a.local_blk.4.attn_k.bias
    F32
    [1024]
  • mm.a.local_blk.4.attn_k.weight
    BF16
    [1024, 1024]
  • mm.a.local_blk.4.attn_out.weight
    BF16
    [1024, 1024]
  • mm.a.local_blk.4.attn_q.bias
    F32
    [1024]
  • mm.a.local_blk.4.attn_q.weight
    BF16
    [1024, 1024]
  • mm.a.local_blk.4.attn_v.bias
    F32
    [1024]
  • mm.a.local_blk.4.attn_v.weight
    BF16
    [1024, 1024]
  • mm.a.local_blk.4.ffn_down.weight
    BF16
    [4096, 1024]
  • mm.a.local_blk.4.ffn_gate.weight
    BF16
    [1024, 4096]
  • mm.a.local_blk.4.ffn_up.weight
    BF16
    [1024, 4096]
  • mm.a.local_blk.4.ln1.weight
    F32
    [1024]
  • mm.a.local_blk.4.ln2.weight
    F32
    [1024]
  • mm.a.local_blk.5.attn_k.bias
    F32
    [1024]
  • mm.a.local_blk.5.attn_k.weight
    BF16
    [1024, 1024]
  • mm.a.local_blk.5.attn_out.weight
    BF16
    [1024, 1024]
  • mm.a.local_blk.5.attn_q.bias
    F32
    [1024]
  • mm.a.local_blk.5.attn_q.weight
    BF16
    [1024, 1024]
  • mm.a.local_blk.5.attn_v.bias
    F32
    [1024]
  • mm.a.local_blk.5.attn_v.weight
    BF16
    [1024, 1024]
  • mm.a.local_blk.5.ffn_down.weight
    BF16
    [4096, 1024]
  • mm.a.local_blk.5.ffn_gate.weight
    BF16
    [1024, 4096]
  • mm.a.local_blk.5.ffn_up.weight
    BF16
    [1024, 4096]
  • mm.a.local_blk.5.ln1.weight
    F32
    [1024]
  • mm.a.local_blk.5.ln2.weight
    F32
    [1024]
  • mm.a.local_norm.weight
    F32
    [1024]
  • mm.a.mlp.1.weight
    BF16
    [4096, 16384]
  • mm.a.mlp.2.weight
    BF16
    [16384, 6144]
  • v.blk.0
  • v.blk.0.attn_out.bias
    F32
    [1280]
  • v.blk.0.attn_out.weight
    BF16
    [2048, 1280]
  • v.blk.0.attn_qkv.bias
    F32
    [3072]
  • v.blk.0.attn_qkv.weight
    BF16
    [1280, 3072]
  • v.blk.0.ffn_down.bias
    F32
    [1280]
  • v.blk.0.ffn_down.weight
    BF16
    [4608, 1280]
  • v.blk.0.ffn_gate.bias
    F32
    [4608]
  • v.blk.0.ffn_gate.weight
    BF16
    [1280, 4608]
  • v.blk.0.ffn_up.bias
    F32
    [4608]
  • v.blk.0.ffn_up.weight
    BF16
    [1280, 4608]
  • v.blk.0.ln1.weight
    F32
    [1280]
  • v.blk.0.ln2.weight
    F32
    [1280]
  • v.blk.1
  • v.blk.1.attn_out.bias
    F32
    [1280]
  • v.blk.1.attn_out.weight
    BF16
    [2048, 1280]
  • v.blk.1.attn_qkv.bias
    F32
    [3072]
  • v.blk.1.attn_qkv.weight
    BF16
    [1280, 3072]
  • v.blk.1.attn_sinks
    F32
    [32]
  • v.blk.1.ffn_down.bias
    F32
    [1280]
  • v.blk.1.ffn_down.weight
    BF16
    [4608, 1280]
  • v.blk.1.ffn_gate.bias
    F32
    [4608]
  • v.blk.1.ffn_gate.weight
    BF16
    [1280, 4608]
  • v.blk.1.ffn_up.bias
    F32
    [4608]
  • v.blk.1.ffn_up.weight
    BF16
    [1280, 4608]
  • v.blk.1.ln1.weight
    F32
    [1280]
  • v.blk.1.ln2.weight
    F32
    [1280]
  • v.blk.2
  • v.blk.2.attn_out.bias
    F32
    [1280]
  • v.blk.2.attn_out.weight
    BF16
    [2048, 1280]
  • v.blk.2.attn_qkv.bias
    F32
    [3072]
  • v.blk.2.attn_qkv.weight
    BF16
    [1280, 3072]
  • v.blk.2.attn_sinks
    F32
    [32]
  • v.blk.2.ffn_down.bias
    F32
    [1280]
  • v.blk.2.ffn_down.weight
    BF16
    [4608, 1280]
  • v.blk.2.ffn_gate.bias
    F32
    [4608]
  • v.blk.2.ffn_gate.weight
    BF16
    [1280, 4608]
  • v.blk.2.ffn_up.bias
    F32
    [4608]
  • v.blk.2.ffn_up.weight
    BF16
    [1280, 4608]
  • v.blk.2.ln1.weight
    F32
    [1280]
  • v.blk.2.ln2.weight
    F32
    [1280]
  • v.blk.3
  • v.blk.3.attn_out.bias
    F32
    [1280]
  • v.blk.3.attn_out.weight
    BF16
    [2048, 1280]
  • v.blk.3.attn_qkv.bias
    F32
    [3072]
  • v.blk.3.attn_qkv.weight
    BF16
    [1280, 3072]
  • v.blk.3.attn_sinks
    F32
    [32]
  • v.blk.3.ffn_down.bias
    F32
    [1280]
  • v.blk.3.ffn_down.weight
    BF16
    [4608, 1280]
  • v.blk.3.ffn_gate.bias
    F32
    [4608]
  • v.blk.3.ffn_gate.weight
    BF16
    [1280, 4608]
  • v.blk.3.ffn_up.bias
    F32
    [4608]
  • v.blk.3.ffn_up.weight
    BF16
    [1280, 4608]
  • v.blk.3.ln1.weight
    F32
    [1280]
  • v.blk.3.ln2.weight
    F32
    [1280]
  • v.blk.4
  • v.blk.4.attn_out.bias
    F32
    [1280]
  • v.blk.4.attn_out.weight
    BF16
    [2048, 1280]
  • v.blk.4.attn_qkv.bias
    F32
    [3072]
  • v.blk.4.attn_qkv.weight
    BF16
    [1280, 3072]
  • v.blk.4.attn_sinks
    F32
    [32]
  • v.blk.4.ffn_down.bias
    F32
    [1280]
  • v.blk.4.ffn_down.weight
    BF16
    [4608, 1280]
  • v.blk.4.ffn_gate.bias
    F32
    [4608]
  • v.blk.4.ffn_gate.weight
    BF16
    [1280, 4608]
  • v.blk.4.ffn_up.bias
    F32
    [4608]
  • v.blk.4.ffn_up.weight
    BF16
    [1280, 4608]
  • v.blk.4.ln1.weight
    F32
    [1280]
  • v.blk.4.ln2.weight
    F32
    [1280]
  • v.blk.5
  • v.blk.5.attn_out.bias
    F32
    [1280]
  • v.blk.5.attn_out.weight
    BF16
    [2048, 1280]
  • v.blk.5.attn_qkv.bias
    F32
    [3072]
  • v.blk.5.attn_qkv.weight
    BF16
    [1280, 3072]
  • v.blk.5.attn_sinks
    F32
    [32]
  • v.blk.5.ffn_down.bias
    F32
    [1280]
  • v.blk.5.ffn_down.weight
    BF16
    [4608, 1280]
  • v.blk.5.ffn_gate.bias
    F32
    [4608]
  • v.blk.5.ffn_gate.weight
    BF16
    [1280, 4608]
  • v.blk.5.ffn_up.bias
    F32
    [4608]
  • v.blk.5.ffn_up.weight
    BF16
    [1280, 4608]
  • v.blk.5.ln1.weight
    F32
    [1280]
  • v.blk.5.ln2.weight
    F32
    [1280]
  • v.blk.6
  • v.blk.6.attn_out.bias
    F32
    [1280]
  • v.blk.6.attn_out.weight
    BF16
    [2048, 1280]
  • v.blk.6.attn_qkv.bias
    F32
    [3072]
  • v.blk.6.attn_qkv.weight
    BF16
    [1280, 3072]
  • v.blk.6.attn_sinks
    F32
    [32]
  • v.blk.6.ffn_down.bias
    F32
    [1280]
  • v.blk.6.ffn_down.weight
    BF16
    [4608, 1280]
  • v.blk.6.ffn_gate.bias
    F32
    [4608]
  • v.blk.6.ffn_gate.weight
    BF16
    [1280, 4608]
  • v.blk.6.ffn_up.bias
    F32
    [4608]
  • v.blk.6.ffn_up.weight
    BF16
    [1280, 4608]
  • v.blk.6.ln1.weight
    F32
    [1280]
  • v.blk.6.ln2.weight
    F32
    [1280]
  • v.blk.7
  • v.blk.7.attn_out.bias
    F32
    [1280]
  • v.blk.7.attn_out.weight
    BF16
    [2048, 1280]
  • v.blk.7.attn_qkv.bias
    F32
    [3072]
  • v.blk.7.attn_qkv.weight
    BF16
    [1280, 3072]
  • v.blk.7.attn_sinks
    F32
    [32]
  • v.blk.7.ffn_down.bias
    F32
    [1280]
  • v.blk.7.ffn_down.weight
    BF16
    [4608, 1280]
  • v.blk.7.ffn_gate.bias
    F32
    [4608]
  • v.blk.7.ffn_gate.weight
    BF16
    [1280, 4608]
  • v.blk.7.ffn_up.bias
    F32
    [4608]
  • v.blk.7.ffn_up.weight
    BF16
    [1280, 4608]
  • v.blk.7.ln1.weight
    F32
    [1280]
  • v.blk.7.ln2.weight
    F32
    [1280]
  • v.blk.8
  • v.blk.8.attn_out.bias
    F32
    [1280]
  • v.blk.8.attn_out.weight
    BF16
    [2048, 1280]
  • v.blk.8.attn_qkv.bias
    F32
    [3072]
  • v.blk.8.attn_qkv.weight
    BF16
    [1280, 3072]
  • v.blk.8.attn_sinks
    F32
    [32]
  • v.blk.8.ffn_down.bias
    F32
    [1280]
  • v.blk.8.ffn_down.weight
    BF16
    [4608, 1280]
  • v.blk.8.ffn_gate.bias
    F32
    [4608]
  • v.blk.8.ffn_gate.weight
    BF16
    [1280, 4608]
  • v.blk.8.ffn_up.bias
    F32
    [4608]
  • v.blk.8.ffn_up.weight
    BF16
    [1280, 4608]
  • v.blk.8.ln1.weight
    F32
    [1280]
  • v.blk.8.ln2.weight
    F32
    [1280]
  • v.blk.9
  • v.blk.9.attn_out.bias
    F32
    [1280]
  • v.blk.9.attn_out.weight
    BF16
    [2048, 1280]
  • v.blk.9.attn_qkv.bias
    F32
    [3072]
  • v.blk.9.attn_qkv.weight
    BF16
    [1280, 3072]
  • v.blk.9.ffn_down.bias
    F32
    [1280]
  • v.blk.9.ffn_down.weight
    BF16
    [4608, 1280]
  • v.blk.9.ffn_gate.bias
    F32
    [4608]
  • v.blk.9.ffn_gate.weight
    BF16
    [1280, 4608]
  • v.blk.9.ffn_up.bias
    F32
    [4608]
  • v.blk.9.ffn_up.weight
    BF16
    [1280, 4608]
  • v.blk.9.ln1.weight
    F32
    [1280]
  • v.blk.9.ln2.weight
    F32
    [1280]
  • v.blk.10
  • v.blk.10.attn_out.bias
    F32
    [1280]
  • v.blk.10.attn_out.weight
    BF16
    [2048, 1280]
  • v.blk.10.attn_qkv.bias
    F32
    [3072]
  • v.blk.10.attn_qkv.weight
    BF16
    [1280, 3072]
  • v.blk.10.attn_sinks
    F32
    [32]
  • v.blk.10.ffn_down.bias
    F32
    [1280]
  • v.blk.10.ffn_down.weight
    BF16
    [4608, 1280]
  • v.blk.10.ffn_gate.bias
    F32
    [4608]
  • v.blk.10.ffn_gate.weight
    BF16
    [1280, 4608]
  • v.blk.10.ffn_up.bias
    F32
    [4608]
  • v.blk.10.ffn_up.weight
    BF16
    [1280, 4608]
  • v.blk.10.ln1.weight
    F32
    [1280]
  • v.blk.10.ln2.weight
    F32
    [1280]
  • v.blk.11
  • v.blk.11.attn_out.bias
    F32
    [1280]
  • v.blk.11.attn_out.weight
    BF16
    [2048, 1280]
  • v.blk.11.attn_qkv.bias
    F32
    [3072]
  • v.blk.11.attn_qkv.weight
    BF16
    [1280, 3072]
  • v.blk.11.attn_sinks
    F32
    [32]
  • v.blk.11.ffn_down.bias
    F32
    [1280]
  • v.blk.11.ffn_down.weight
    BF16
    [4608, 1280]
  • v.blk.11.ffn_gate.bias
    F32
    [4608]
  • v.blk.11.ffn_gate.weight
    BF16
    [1280, 4608]
  • v.blk.11.ffn_up.bias
    F32
    [4608]
  • v.blk.11.ffn_up.weight
    BF16
    [1280, 4608]
  • v.blk.11.ln1.weight
    F32
    [1280]
  • v.blk.11.ln2.weight
    F32
    [1280]
  • v.blk.12
  • v.blk.12.attn_out.bias
    F32
    [1280]
  • v.blk.12.attn_out.weight
    BF16
    [2048, 1280]
  • v.blk.12.attn_qkv.bias
    F32
    [3072]
  • v.blk.12.attn_qkv.weight
    BF16
    [1280, 3072]
  • v.blk.12.attn_sinks
    F32
    [32]
  • v.blk.12.ffn_down.bias
    F32
    [1280]
  • v.blk.12.ffn_down.weight
    BF16
    [4608, 1280]
  • v.blk.12.ffn_gate.bias
    F32
    [4608]
  • v.blk.12.ffn_gate.weight
    BF16
    [1280, 4608]
  • v.blk.12.ffn_up.bias
    F32
    [4608]
  • v.blk.12.ffn_up.weight
    BF16
    [1280, 4608]
  • v.blk.12.ln1.weight
    F32
    [1280]
  • v.blk.12.ln2.weight
    F32
    [1280]
  • v.blk.13
  • v.blk.13.attn_out.bias
    F32
    [1280]
  • v.blk.13.attn_out.weight
    BF16
    [2048, 1280]
  • v.blk.13.attn_qkv.bias
    F32
    [3072]
  • v.blk.13.attn_qkv.weight
    BF16
    [1280, 3072]
  • v.blk.13.attn_sinks
    F32
    [32]
  • v.blk.13.ffn_down.bias
    F32
    [1280]
  • v.blk.13.ffn_down.weight
    BF16
    [4608, 1280]
  • v.blk.13.ffn_gate.bias
    F32
    [4608]
  • v.blk.13.ffn_gate.weight
    BF16
    [1280, 4608]
  • v.blk.13.ffn_up.bias
    F32
    [4608]
  • v.blk.13.ffn_up.weight
    BF16
    [1280, 4608]
  • v.blk.13.ln1.weight
    F32
    [1280]
  • v.blk.13.ln2.weight
    F32
    [1280]
  • v.blk.14
  • v.blk.14.attn_out.bias
    F32
    [1280]
  • v.blk.14.attn_out.weight
    BF16
    [2048, 1280]
  • v.blk.14.attn_qkv.bias
    F32
    [3072]
  • v.blk.14.attn_qkv.weight
    BF16
    [1280, 3072]
  • v.blk.14.attn_sinks
    F32
    [32]
  • v.blk.14.ffn_down.bias
    F32
    [1280]
  • v.blk.14.ffn_down.weight
    BF16
    [4608, 1280]
  • v.blk.14.ffn_gate.bias
    F32
    [4608]
  • v.blk.14.ffn_gate.weight
    BF16
    [1280, 4608]
  • v.blk.14.ffn_up.bias
    F32
    [4608]
  • v.blk.14.ffn_up.weight
    BF16
    [1280, 4608]
  • v.blk.14.ln1.weight
    F32
    [1280]
  • v.blk.14.ln2.weight
    F32
    [1280]
  • v.blk.15
  • v.blk.15.attn_out.bias
    F32
    [1280]
  • v.blk.15.attn_out.weight
    BF16
    [2048, 1280]
  • v.blk.15.attn_qkv.bias
    F32
    [3072]
  • v.blk.15.attn_qkv.weight
    BF16
    [1280, 3072]
  • v.blk.15.attn_sinks
    F32
    [32]
  • v.blk.15.ffn_down.bias
    F32
    [1280]
  • v.blk.15.ffn_down.weight
    BF16
    [4608, 1280]
  • v.blk.15.ffn_gate.bias
    F32
    [4608]
  • v.blk.15.ffn_gate.weight
    BF16
    [1280, 4608]
  • v.blk.15.ffn_up.bias
    F32
    [4608]
  • v.blk.15.ffn_up.weight
    BF16
    [1280, 4608]
  • v.blk.15.ln1.weight
    F32
    [1280]
  • v.blk.15.ln2.weight
    F32
    [1280]
  • v.blk.16
  • v.blk.16.attn_out.bias
    F32
    [1280]
  • v.blk.16.attn_out.weight
    BF16
    [2048, 1280]
  • v.blk.16.attn_qkv.bias
    F32
    [3072]
  • v.blk.16.attn_qkv.weight
    BF16
    [1280, 3072]
  • v.blk.16.attn_sinks
    F32
    [32]
  • v.blk.16.ffn_down.bias
    F32
    [1280]
  • v.blk.16.ffn_down.weight
    BF16
    [4608, 1280]
  • v.blk.16.ffn_gate.bias
    F32
    [4608]
  • v.blk.16.ffn_gate.weight
    BF16
    [1280, 4608]
  • v.blk.16.ffn_up.bias
    F32
    [4608]
  • v.blk.16.ffn_up.weight
    BF16
    [1280, 4608]
  • v.blk.16.ln1.weight
    F32
    [1280]
  • v.blk.16.ln2.weight
    F32
    [1280]
  • v.blk.17
  • v.blk.17.attn_out.bias
    F32
    [1280]
  • v.blk.17.attn_out.weight
    BF16
    [2048, 1280]
  • v.blk.17.attn_qkv.bias
    F32
    [3072]
  • v.blk.17.attn_qkv.weight
    BF16
    [1280, 3072]
  • v.blk.17.attn_sinks
    F32
    [32]
  • v.blk.17.ffn_down.bias
    F32
    [1280]
  • v.blk.17.ffn_down.weight
    BF16
    [4608, 1280]
  • v.blk.17.ffn_gate.bias
    F32
    [4608]
  • v.blk.17.ffn_gate.weight
    BF16
    [1280, 4608]
  • v.blk.17.ffn_up.bias
    F32
    [4608]
  • v.blk.17.ffn_up.weight
    BF16
    [1280, 4608]
  • v.blk.17.ln1.weight
    F32
    [1280]
  • v.blk.17.ln2.weight
    F32
    [1280]
  • v.blk.18
  • v.blk.18.attn_out.bias
    F32
    [1280]
  • v.blk.18.attn_out.weight
    BF16
    [2048, 1280]
  • v.blk.18.attn_qkv.bias
    F32
    [3072]
  • v.blk.18.attn_qkv.weight
    BF16
    [1280, 3072]
  • v.blk.18.ffn_down.bias
    F32
    [1280]
  • v.blk.18.ffn_down.weight
    BF16
    [4608, 1280]
  • v.blk.18.ffn_gate.bias
    F32
    [4608]
  • v.blk.18.ffn_gate.weight
    BF16
    [1280, 4608]
  • v.blk.18.ffn_up.bias
    F32
    [4608]
  • v.blk.18.ffn_up.weight
    BF16
    [1280, 4608]
  • v.blk.18.ln1.weight
    F32
    [1280]
  • v.blk.18.ln2.weight
    F32
    [1280]
  • v.blk.19
  • v.blk.19.attn_out.bias
    F32
    [1280]
  • v.blk.19.attn_out.weight
    BF16
    [2048, 1280]
  • v.blk.19.attn_qkv.bias
    F32
    [3072]
  • v.blk.19.attn_qkv.weight
    BF16
    [1280, 3072]
  • v.blk.19.attn_sinks
    F32
    [32]
  • v.blk.19.ffn_down.bias
    F32
    [1280]
  • v.blk.19.ffn_down.weight
    BF16
    [4608, 1280]
  • v.blk.19.ffn_gate.bias
    F32
    [4608]
  • v.blk.19.ffn_gate.weight
    BF16
    [1280, 4608]
  • v.blk.19.ffn_up.bias
    F32
    [4608]
  • v.blk.19.ffn_up.weight
    BF16
    [1280, 4608]
  • v.blk.19.ln1.weight
    F32
    [1280]
  • v.blk.19.ln2.weight
    F32
    [1280]
  • v.blk.20
  • v.blk.20.attn_out.bias
    F32
    [1280]
  • v.blk.20.attn_out.weight
    BF16
    [2048, 1280]
  • v.blk.20.attn_qkv.bias
    F32
    [3072]
  • v.blk.20.attn_qkv.weight
    BF16
    [1280, 3072]
  • v.blk.20.attn_sinks
    F32
    [32]
  • v.blk.20.ffn_down.bias
    F32
    [1280]
  • v.blk.20.ffn_down.weight
    BF16
    [4608, 1280]
  • v.blk.20.ffn_gate.bias
    F32
    [4608]
  • v.blk.20.ffn_gate.weight
    BF16
    [1280, 4608]
  • v.blk.20.ffn_up.bias
    F32
    [4608]
  • v.blk.20.ffn_up.weight
    BF16
    [1280, 4608]
  • v.blk.20.ln1.weight
    F32
    [1280]
  • v.blk.20.ln2.weight
    F32
    [1280]
  • v.blk.21
  • v.blk.21.attn_out.bias
    F32
    [1280]
  • v.blk.21.attn_out.weight
    BF16
    [2048, 1280]
  • v.blk.21.attn_qkv.bias
    F32
    [3072]
  • v.blk.21.attn_qkv.weight
    BF16
    [1280, 3072]
  • v.blk.21.attn_sinks
    F32
    [32]
  • v.blk.21.ffn_down.bias
    F32
    [1280]
  • v.blk.21.ffn_down.weight
    BF16
    [4608, 1280]
  • v.blk.21.ffn_gate.bias
    F32
    [4608]
  • v.blk.21.ffn_gate.weight
    BF16
    [1280, 4608]
  • v.blk.21.ffn_up.bias
    F32
    [4608]
  • v.blk.21.ffn_up.weight
    BF16
    [1280, 4608]
  • v.blk.21.ln1.weight
    F32
    [1280]
  • v.blk.21.ln2.weight
    F32
    [1280]
  • v.blk.22
  • v.blk.22.attn_out.bias
    F32
    [1280]
  • v.blk.22.attn_out.weight
    BF16
    [2048, 1280]
  • v.blk.22.attn_qkv.bias
    F32
    [3072]
  • v.blk.22.attn_qkv.weight
    BF16
    [1280, 3072]
  • v.blk.22.attn_sinks
    F32
    [32]
  • v.blk.22.ffn_down.bias
    F32
    [1280]
  • v.blk.22.ffn_down.weight
    BF16
    [4608, 1280]
  • v.blk.22.ffn_gate.bias
    F32
    [4608]
  • v.blk.22.ffn_gate.weight
    BF16
    [1280, 4608]
  • v.blk.22.ffn_up.bias
    F32
    [4608]
  • v.blk.22.ffn_up.weight
    BF16
    [1280, 4608]
  • v.blk.22.ln1.weight
    F32
    [1280]
  • v.blk.22.ln2.weight
    F32
    [1280]
  • v.blk.23
  • v.blk.23.attn_out.bias
    F32
    [1280]
  • v.blk.23.attn_out.weight
    BF16
    [2048, 1280]
  • v.blk.23.attn_qkv.bias
    F32
    [3072]
  • v.blk.23.attn_qkv.weight
    BF16
    [1280, 3072]
  • v.blk.23.attn_sinks
    F32
    [32]
  • v.blk.23.ffn_down.bias
    F32
    [1280]
  • v.blk.23.ffn_down.weight
    BF16
    [4608, 1280]
  • v.blk.23.ffn_gate.bias
    F32
    [4608]
  • v.blk.23.ffn_gate.weight
    BF16
    [1280, 4608]
  • v.blk.23.ffn_up.bias
    F32
    [4608]
  • v.blk.23.ffn_up.weight
    BF16
    [1280, 4608]
  • v.blk.23.ln1.weight
    F32
    [1280]
  • v.blk.23.ln2.weight
    F32
    [1280]
  • v.blk.24
  • v.blk.24.attn_out.bias
    F32
    [1280]
  • v.blk.24.attn_out.weight
    BF16
    [2048, 1280]
  • v.blk.24.attn_qkv.bias
    F32
    [3072]
  • v.blk.24.attn_qkv.weight
    BF16
    [1280, 3072]
  • v.blk.24.attn_sinks
    F32
    [32]
  • v.blk.24.ffn_down.bias
    F32
    [1280]
  • v.blk.24.ffn_down.weight
    BF16
    [4608, 1280]
  • v.blk.24.ffn_gate.bias
    F32
    [4608]
  • v.blk.24.ffn_gate.weight
    BF16
    [1280, 4608]
  • v.blk.24.ffn_up.bias
    F32
    [4608]
  • v.blk.24.ffn_up.weight
    BF16
    [1280, 4608]
  • v.blk.24.ln1.weight
    F32
    [1280]
  • v.blk.24.ln2.weight
    F32
    [1280]
  • v.blk.25
  • v.blk.25.attn_out.bias
    F32
    [1280]
  • v.blk.25.attn_out.weight
    BF16
    [2048, 1280]
  • v.blk.25.attn_qkv.bias
    F32
    [3072]
  • v.blk.25.attn_qkv.weight
    BF16
    [1280, 3072]
  • v.blk.25.attn_sinks
    F32
    [32]
  • v.blk.25.ffn_down.bias
    F32
    [1280]
  • v.blk.25.ffn_down.weight
    BF16
    [4608, 1280]
  • v.blk.25.ffn_gate.bias
    F32
    [4608]
  • v.blk.25.ffn_gate.weight
    BF16
    [1280, 4608]
  • v.blk.25.ffn_up.bias
    F32
    [4608]
  • v.blk.25.ffn_up.weight
    BF16
    [1280, 4608]
  • v.blk.25.ln1.weight
    F32
    [1280]
  • v.blk.25.ln2.weight
    F32
    [1280]
  • v.blk.26
  • v.blk.26.attn_out.bias
    F32
    [1280]
  • v.blk.26.attn_out.weight
    BF16
    [2048, 1280]
  • v.blk.26.attn_qkv.bias
    F32
    [3072]
  • v.blk.26.attn_qkv.weight
    BF16
    [1280, 3072]
  • v.blk.26.attn_sinks
    F32
    [32]
  • v.blk.26.ffn_down.bias
    F32
    [1280]
  • v.blk.26.ffn_down.weight
    BF16
    [4608, 1280]
  • v.blk.26.ffn_gate.bias
    F32
    [4608]
  • v.blk.26.ffn_gate.weight
    BF16
    [1280, 4608]
  • v.blk.26.ffn_up.bias
    F32
    [4608]
  • v.blk.26.ffn_up.weight
    BF16
    [1280, 4608]
  • v.blk.26.ln1.weight
    F32
    [1280]
  • v.blk.26.ln2.weight
    F32
    [1280]
  • v.blk.27
  • v.blk.27.attn_out.bias
    F32
    [1280]
  • v.blk.27.attn_out.weight
    BF16
    [2048, 1280]
  • v.blk.27.attn_qkv.bias
    F32
    [3072]
  • v.blk.27.attn_qkv.weight
    BF16
    [1280, 3072]
  • v.blk.27.ffn_down.bias
    F32
    [1280]
  • v.blk.27.ffn_down.weight
    BF16
    [4608, 1280]
  • v.blk.27.ffn_gate.bias
    F32
    [4608]
  • v.blk.27.ffn_gate.weight
    BF16
    [1280, 4608]
  • v.blk.27.ffn_up.bias
    F32
    [4608]
  • v.blk.27.ffn_up.weight
    BF16
    [1280, 4608]
  • v.blk.27.ln1.weight
    F32
    [1280]
  • v.blk.27.ln2.weight
    F32
    [1280]
  • v.patch_embd.weight
    F32
    [16, 16, 3, 1280]
  • v.patch_embd.weight.1
    F32
    [16, 16, 3, 1280]
  • v.post_ln.weight
    F32
    [1280]