152 3 weeks ago

vision 258m
4b7b2c4ed316 · 190MB
    Metadata
  • general.architecture
    clip
  • general.file_type
    F16
  • clip.has_vision_encoder
    true
  • clip.projector_type
    idefics3
  • clip.use_gelu
    true
  • clip.vision.attention.head_count
    12
  • clip.vision.attention.layer_norm_epsilon
    1e-06
  • clip.vision.block_count
    12
  • clip.vision.embedding_length
    768
  • clip.vision.feed_forward_length
    3072
  • clip.vision.image_mean
    [0.5, 0.5, 0.5]
  • clip.vision.image_size
    512
  • clip.vision.image_std
    [0.5, 0.5, 0.5]
  • clip.vision.patch_size
    16
  • clip.vision.preproc_image_size
    2048
  • clip.vision.projection_dim
    576
  • clip.vision.projector.scale_factor
    4
  • Tensor
  • mm.model.fc.weight
    F16
    [12288, 576]
  • v.blk.0
  • v.blk.0.attn_k.bias
    F32
    [768]
  • v.blk.0.attn_k.weight
    F16
    [768, 768]
  • v.blk.0.attn_out.bias
    F32
    [768]
  • v.blk.0.attn_out.weight
    F16
    [768, 768]
  • v.blk.0.attn_q.bias
    F32
    [768]
  • v.blk.0.attn_q.weight
    F16
    [768, 768]
  • v.blk.0.attn_v.bias
    F32
    [768]
  • v.blk.0.attn_v.weight
    F16
    [768, 768]
  • v.blk.0.ffn_down.bias
    F32
    [768]
  • v.blk.0.ffn_down.weight
    F16
    [3072, 768]
  • v.blk.0.ffn_up.bias
    F32
    [3072]
  • v.blk.0.ffn_up.weight
    F16
    [768, 3072]
  • v.blk.0.ln1.bias
    F32
    [768]
  • v.blk.0.ln1.weight
    F32
    [768]
  • v.blk.0.ln2.bias
    F32
    [768]
  • v.blk.0.ln2.weight
    F32
    [768]
  • v.blk.1
  • v.blk.1.attn_k.bias
    F32
    [768]
  • v.blk.1.attn_k.weight
    F16
    [768, 768]
  • v.blk.1.attn_out.bias
    F32
    [768]
  • v.blk.1.attn_out.weight
    F16
    [768, 768]
  • v.blk.1.attn_q.bias
    F32
    [768]
  • v.blk.1.attn_q.weight
    F16
    [768, 768]
  • v.blk.1.attn_v.bias
    F32
    [768]
  • v.blk.1.attn_v.weight
    F16
    [768, 768]
  • v.blk.1.ffn_down.bias
    F32
    [768]
  • v.blk.1.ffn_down.weight
    F16
    [3072, 768]
  • v.blk.1.ffn_up.bias
    F32
    [3072]
  • v.blk.1.ffn_up.weight
    F16
    [768, 3072]
  • v.blk.1.ln1.bias
    F32
    [768]
  • v.blk.1.ln1.weight
    F32
    [768]
  • v.blk.1.ln2.bias
    F32
    [768]
  • v.blk.1.ln2.weight
    F32
    [768]
  • v.blk.2
  • v.blk.2.attn_k.bias
    F32
    [768]
  • v.blk.2.attn_k.weight
    F16
    [768, 768]
  • v.blk.2.attn_out.bias
    F32
    [768]
  • v.blk.2.attn_out.weight
    F16
    [768, 768]
  • v.blk.2.attn_q.bias
    F32
    [768]
  • v.blk.2.attn_q.weight
    F16
    [768, 768]
  • v.blk.2.attn_v.bias
    F32
    [768]
  • v.blk.2.attn_v.weight
    F16
    [768, 768]
  • v.blk.2.ffn_down.bias
    F32
    [768]
  • v.blk.2.ffn_down.weight
    F16
    [3072, 768]
  • v.blk.2.ffn_up.bias
    F32
    [3072]
  • v.blk.2.ffn_up.weight
    F16
    [768, 3072]
  • v.blk.2.ln1.bias
    F32
    [768]
  • v.blk.2.ln1.weight
    F32
    [768]
  • v.blk.2.ln2.bias
    F32
    [768]
  • v.blk.2.ln2.weight
    F32
    [768]
  • v.blk.3
  • v.blk.3.attn_k.bias
    F32
    [768]
  • v.blk.3.attn_k.weight
    F16
    [768, 768]
  • v.blk.3.attn_out.bias
    F32
    [768]
  • v.blk.3.attn_out.weight
    F16
    [768, 768]
  • v.blk.3.attn_q.bias
    F32
    [768]
  • v.blk.3.attn_q.weight
    F16
    [768, 768]
  • v.blk.3.attn_v.bias
    F32
    [768]
  • v.blk.3.attn_v.weight
    F16
    [768, 768]
  • v.blk.3.ffn_down.bias
    F32
    [768]
  • v.blk.3.ffn_down.weight
    F16
    [3072, 768]
  • v.blk.3.ffn_up.bias
    F32
    [3072]
  • v.blk.3.ffn_up.weight
    F16
    [768, 3072]
  • v.blk.3.ln1.bias
    F32
    [768]
  • v.blk.3.ln1.weight
    F32
    [768]
  • v.blk.3.ln2.bias
    F32
    [768]
  • v.blk.3.ln2.weight
    F32
    [768]
  • v.blk.4
  • v.blk.4.attn_k.bias
    F32
    [768]
  • v.blk.4.attn_k.weight
    F16
    [768, 768]
  • v.blk.4.attn_out.bias
    F32
    [768]
  • v.blk.4.attn_out.weight
    F16
    [768, 768]
  • v.blk.4.attn_q.bias
    F32
    [768]
  • v.blk.4.attn_q.weight
    F16
    [768, 768]
  • v.blk.4.attn_v.bias
    F32
    [768]
  • v.blk.4.attn_v.weight
    F16
    [768, 768]
  • v.blk.4.ffn_down.bias
    F32
    [768]
  • v.blk.4.ffn_down.weight
    F16
    [3072, 768]
  • v.blk.4.ffn_up.bias
    F32
    [3072]
  • v.blk.4.ffn_up.weight
    F16
    [768, 3072]
  • v.blk.4.ln1.bias
    F32
    [768]
  • v.blk.4.ln1.weight
    F32
    [768]
  • v.blk.4.ln2.bias
    F32
    [768]
  • v.blk.4.ln2.weight
    F32
    [768]
  • v.blk.5
  • v.blk.5.attn_k.bias
    F32
    [768]
  • v.blk.5.attn_k.weight
    F16
    [768, 768]
  • v.blk.5.attn_out.bias
    F32
    [768]
  • v.blk.5.attn_out.weight
    F16
    [768, 768]
  • v.blk.5.attn_q.bias
    F32
    [768]
  • v.blk.5.attn_q.weight
    F16
    [768, 768]
  • v.blk.5.attn_v.bias
    F32
    [768]
  • v.blk.5.attn_v.weight
    F16
    [768, 768]
  • v.blk.5.ffn_down.bias
    F32
    [768]
  • v.blk.5.ffn_down.weight
    F16
    [3072, 768]
  • v.blk.5.ffn_up.bias
    F32
    [3072]
  • v.blk.5.ffn_up.weight
    F16
    [768, 3072]
  • v.blk.5.ln1.bias
    F32
    [768]
  • v.blk.5.ln1.weight
    F32
    [768]
  • v.blk.5.ln2.bias
    F32
    [768]
  • v.blk.5.ln2.weight
    F32
    [768]
  • v.blk.6
  • v.blk.6.attn_k.bias
    F32
    [768]
  • v.blk.6.attn_k.weight
    F16
    [768, 768]
  • v.blk.6.attn_out.bias
    F32
    [768]
  • v.blk.6.attn_out.weight
    F16
    [768, 768]
  • v.blk.6.attn_q.bias
    F32
    [768]
  • v.blk.6.attn_q.weight
    F16
    [768, 768]
  • v.blk.6.attn_v.bias
    F32
    [768]
  • v.blk.6.attn_v.weight
    F16
    [768, 768]
  • v.blk.6.ffn_down.bias
    F32
    [768]
  • v.blk.6.ffn_down.weight
    F16
    [3072, 768]
  • v.blk.6.ffn_up.bias
    F32
    [3072]
  • v.blk.6.ffn_up.weight
    F16
    [768, 3072]
  • v.blk.6.ln1.bias
    F32
    [768]
  • v.blk.6.ln1.weight
    F32
    [768]
  • v.blk.6.ln2.bias
    F32
    [768]
  • v.blk.6.ln2.weight
    F32
    [768]
  • v.blk.7
  • v.blk.7.attn_k.bias
    F32
    [768]
  • v.blk.7.attn_k.weight
    F16
    [768, 768]
  • v.blk.7.attn_out.bias
    F32
    [768]
  • v.blk.7.attn_out.weight
    F16
    [768, 768]
  • v.blk.7.attn_q.bias
    F32
    [768]
  • v.blk.7.attn_q.weight
    F16
    [768, 768]
  • v.blk.7.attn_v.bias
    F32
    [768]
  • v.blk.7.attn_v.weight
    F16
    [768, 768]
  • v.blk.7.ffn_down.bias
    F32
    [768]
  • v.blk.7.ffn_down.weight
    F16
    [3072, 768]
  • v.blk.7.ffn_up.bias
    F32
    [3072]
  • v.blk.7.ffn_up.weight
    F16
    [768, 3072]
  • v.blk.7.ln1.bias
    F32
    [768]
  • v.blk.7.ln1.weight
    F32
    [768]
  • v.blk.7.ln2.bias
    F32
    [768]
  • v.blk.7.ln2.weight
    F32
    [768]
  • v.blk.8
  • v.blk.8.attn_k.bias
    F32
    [768]
  • v.blk.8.attn_k.weight
    F16
    [768, 768]
  • v.blk.8.attn_out.bias
    F32
    [768]
  • v.blk.8.attn_out.weight
    F16
    [768, 768]
  • v.blk.8.attn_q.bias
    F32
    [768]
  • v.blk.8.attn_q.weight
    F16
    [768, 768]
  • v.blk.8.attn_v.bias
    F32
    [768]
  • v.blk.8.attn_v.weight
    F16
    [768, 768]
  • v.blk.8.ffn_down.bias
    F32
    [768]
  • v.blk.8.ffn_down.weight
    F16
    [3072, 768]
  • v.blk.8.ffn_up.bias
    F32
    [3072]
  • v.blk.8.ffn_up.weight
    F16
    [768, 3072]
  • v.blk.8.ln1.bias
    F32
    [768]
  • v.blk.8.ln1.weight
    F32
    [768]
  • v.blk.8.ln2.bias
    F32
    [768]
  • v.blk.8.ln2.weight
    F32
    [768]
  • v.blk.9
  • v.blk.9.attn_k.bias
    F32
    [768]
  • v.blk.9.attn_k.weight
    F16
    [768, 768]
  • v.blk.9.attn_out.bias
    F32
    [768]
  • v.blk.9.attn_out.weight
    F16
    [768, 768]
  • v.blk.9.attn_q.bias
    F32
    [768]
  • v.blk.9.attn_q.weight
    F16
    [768, 768]
  • v.blk.9.attn_v.bias
    F32
    [768]
  • v.blk.9.attn_v.weight
    F16
    [768, 768]
  • v.blk.9.ffn_down.bias
    F32
    [768]
  • v.blk.9.ffn_down.weight
    F16
    [3072, 768]
  • v.blk.9.ffn_up.bias
    F32
    [3072]
  • v.blk.9.ffn_up.weight
    F16
    [768, 3072]
  • v.blk.9.ln1.bias
    F32
    [768]
  • v.blk.9.ln1.weight
    F32
    [768]
  • v.blk.9.ln2.bias
    F32
    [768]
  • v.blk.9.ln2.weight
    F32
    [768]
  • v.blk.10
  • v.blk.10.attn_k.bias
    F32
    [768]
  • v.blk.10.attn_k.weight
    F16
    [768, 768]
  • v.blk.10.attn_out.bias
    F32
    [768]
  • v.blk.10.attn_out.weight
    F16
    [768, 768]
  • v.blk.10.attn_q.bias
    F32
    [768]
  • v.blk.10.attn_q.weight
    F16
    [768, 768]
  • v.blk.10.attn_v.bias
    F32
    [768]
  • v.blk.10.attn_v.weight
    F16
    [768, 768]
  • v.blk.10.ffn_down.bias
    F32
    [768]
  • v.blk.10.ffn_down.weight
    F16
    [3072, 768]
  • v.blk.10.ffn_up.bias
    F32
    [3072]
  • v.blk.10.ffn_up.weight
    F16
    [768, 3072]
  • v.blk.10.ln1.bias
    F32
    [768]
  • v.blk.10.ln1.weight
    F32
    [768]
  • v.blk.10.ln2.bias
    F32
    [768]
  • v.blk.10.ln2.weight
    F32
    [768]
  • v.blk.11
  • v.blk.11.attn_k.bias
    F32
    [768]
  • v.blk.11.attn_k.weight
    F16
    [768, 768]
  • v.blk.11.attn_out.bias
    F32
    [768]
  • v.blk.11.attn_out.weight
    F16
    [768, 768]
  • v.blk.11.attn_q.bias
    F32
    [768]
  • v.blk.11.attn_q.weight
    F16
    [768, 768]
  • v.blk.11.attn_v.bias
    F32
    [768]
  • v.blk.11.attn_v.weight
    F16
    [768, 768]
  • v.blk.11.ffn_down.bias
    F32
    [768]
  • v.blk.11.ffn_down.weight
    F16
    [3072, 768]
  • v.blk.11.ffn_up.bias
    F32
    [3072]
  • v.blk.11.ffn_up.weight
    F16
    [768, 3072]
  • v.blk.11.ln1.bias
    F32
    [768]
  • v.blk.11.ln1.weight
    F32
    [768]
  • v.blk.11.ln2.bias
    F32
    [768]
  • v.blk.11.ln2.weight
    F32
    [768]
  • v.patch_embd.bias
    F32
    [768]
  • v.patch_embd.weight
    F32
    [16, 16, 3, 768]
  • v.position_embd.weight
    F32
    [768, 1024]
  • v.post_ln.bias
    F32
    [768]
  • v.post_ln.weight
    F32
    [768]