220 1 week ago

Holo-3.1 vision-language computer-use agents by H Company. Locate UI elements and drive web, desktop & mobile automation from a screenshot — returns clicks in normalized [0,1000] coords. 0.8B & 4B, instruct & thinking variants, Q4_K_M/Q8_0. Apache 2.0.

vision tools 0.8b 4b
d96d81ac0bfb · 207MB
    Metadata
  • general.architecture
    clip
  • general.file_type
    BF16
  • clip.has_vision_encoder
    true
  • clip.projector_type
    qwen3vl_merger
  • 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
    768
  • clip.vision.image_std
    [0.5, 0.5, 0.5]
  • clip.vision.is_deepstack_layers
    [false, false, false, false, false, ...]
  • clip.vision.patch_size
    16
  • clip.vision.projection_dim
    1024
  • clip.vision.spatial_merge_size
    2
  • Tensor
  • mm.0.bias
    F32
    [3072]
  • mm.0.weight
    BF16
    [3072, 3072]
  • mm.2.bias
    F32
    [1024]
  • mm.2.weight
    BF16
    [3072, 1024]
  • v.blk.0
  • v.blk.0.attn_out.bias
    F32
    [768]
  • v.blk.0.attn_out.weight
    BF16
    [768, 768]
  • v.blk.0.attn_qkv.bias
    F32
    [2304]
  • v.blk.0.attn_qkv.weight
    BF16
    [768, 2304]
  • v.blk.0.ffn_down.bias
    F32
    [768]
  • v.blk.0.ffn_down.weight
    BF16
    [3072, 768]
  • v.blk.0.ffn_up.bias
    F32
    [3072]
  • v.blk.0.ffn_up.weight
    BF16
    [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_out.bias
    F32
    [768]
  • v.blk.1.attn_out.weight
    BF16
    [768, 768]
  • v.blk.1.attn_qkv.bias
    F32
    [2304]
  • v.blk.1.attn_qkv.weight
    BF16
    [768, 2304]
  • v.blk.1.ffn_down.bias
    F32
    [768]
  • v.blk.1.ffn_down.weight
    BF16
    [3072, 768]
  • v.blk.1.ffn_up.bias
    F32
    [3072]
  • v.blk.1.ffn_up.weight
    BF16
    [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_out.bias
    F32
    [768]
  • v.blk.2.attn_out.weight
    BF16
    [768, 768]
  • v.blk.2.attn_qkv.bias
    F32
    [2304]
  • v.blk.2.attn_qkv.weight
    BF16
    [768, 2304]
  • v.blk.2.ffn_down.bias
    F32
    [768]
  • v.blk.2.ffn_down.weight
    BF16
    [3072, 768]
  • v.blk.2.ffn_up.bias
    F32
    [3072]
  • v.blk.2.ffn_up.weight
    BF16
    [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_out.bias
    F32
    [768]
  • v.blk.3.attn_out.weight
    BF16
    [768, 768]
  • v.blk.3.attn_qkv.bias
    F32
    [2304]
  • v.blk.3.attn_qkv.weight
    BF16
    [768, 2304]
  • v.blk.3.ffn_down.bias
    F32
    [768]
  • v.blk.3.ffn_down.weight
    BF16
    [3072, 768]
  • v.blk.3.ffn_up.bias
    F32
    [3072]
  • v.blk.3.ffn_up.weight
    BF16
    [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_out.bias
    F32
    [768]
  • v.blk.4.attn_out.weight
    BF16
    [768, 768]
  • v.blk.4.attn_qkv.bias
    F32
    [2304]
  • v.blk.4.attn_qkv.weight
    BF16
    [768, 2304]
  • v.blk.4.ffn_down.bias
    F32
    [768]
  • v.blk.4.ffn_down.weight
    BF16
    [3072, 768]
  • v.blk.4.ffn_up.bias
    F32
    [3072]
  • v.blk.4.ffn_up.weight
    BF16
    [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_out.bias
    F32
    [768]
  • v.blk.5.attn_out.weight
    BF16
    [768, 768]
  • v.blk.5.attn_qkv.bias
    F32
    [2304]
  • v.blk.5.attn_qkv.weight
    BF16
    [768, 2304]
  • v.blk.5.ffn_down.bias
    F32
    [768]
  • v.blk.5.ffn_down.weight
    BF16
    [3072, 768]
  • v.blk.5.ffn_up.bias
    F32
    [3072]
  • v.blk.5.ffn_up.weight
    BF16
    [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_out.bias
    F32
    [768]
  • v.blk.6.attn_out.weight
    BF16
    [768, 768]
  • v.blk.6.attn_qkv.bias
    F32
    [2304]
  • v.blk.6.attn_qkv.weight
    BF16
    [768, 2304]
  • v.blk.6.ffn_down.bias
    F32
    [768]
  • v.blk.6.ffn_down.weight
    BF16
    [3072, 768]
  • v.blk.6.ffn_up.bias
    F32
    [3072]
  • v.blk.6.ffn_up.weight
    BF16
    [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_out.bias
    F32
    [768]
  • v.blk.7.attn_out.weight
    BF16
    [768, 768]
  • v.blk.7.attn_qkv.bias
    F32
    [2304]
  • v.blk.7.attn_qkv.weight
    BF16
    [768, 2304]
  • v.blk.7.ffn_down.bias
    F32
    [768]
  • v.blk.7.ffn_down.weight
    BF16
    [3072, 768]
  • v.blk.7.ffn_up.bias
    F32
    [3072]
  • v.blk.7.ffn_up.weight
    BF16
    [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_out.bias
    F32
    [768]
  • v.blk.8.attn_out.weight
    BF16
    [768, 768]
  • v.blk.8.attn_qkv.bias
    F32
    [2304]
  • v.blk.8.attn_qkv.weight
    BF16
    [768, 2304]
  • v.blk.8.ffn_down.bias
    F32
    [768]
  • v.blk.8.ffn_down.weight
    BF16
    [3072, 768]
  • v.blk.8.ffn_up.bias
    F32
    [3072]
  • v.blk.8.ffn_up.weight
    BF16
    [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_out.bias
    F32
    [768]
  • v.blk.9.attn_out.weight
    BF16
    [768, 768]
  • v.blk.9.attn_qkv.bias
    F32
    [2304]
  • v.blk.9.attn_qkv.weight
    BF16
    [768, 2304]
  • v.blk.9.ffn_down.bias
    F32
    [768]
  • v.blk.9.ffn_down.weight
    BF16
    [3072, 768]
  • v.blk.9.ffn_up.bias
    F32
    [3072]
  • v.blk.9.ffn_up.weight
    BF16
    [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_out.bias
    F32
    [768]
  • v.blk.10.attn_out.weight
    BF16
    [768, 768]
  • v.blk.10.attn_qkv.bias
    F32
    [2304]
  • v.blk.10.attn_qkv.weight
    BF16
    [768, 2304]
  • v.blk.10.ffn_down.bias
    F32
    [768]
  • v.blk.10.ffn_down.weight
    BF16
    [3072, 768]
  • v.blk.10.ffn_up.bias
    F32
    [3072]
  • v.blk.10.ffn_up.weight
    BF16
    [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_out.bias
    F32
    [768]
  • v.blk.11.attn_out.weight
    BF16
    [768, 768]
  • v.blk.11.attn_qkv.bias
    F32
    [2304]
  • v.blk.11.attn_qkv.weight
    BF16
    [768, 2304]
  • v.blk.11.ffn_down.bias
    F32
    [768]
  • v.blk.11.ffn_down.weight
    BF16
    [3072, 768]
  • v.blk.11.ffn_up.bias
    F32
    [3072]
  • v.blk.11.ffn_up.weight
    BF16
    [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.patch_embd.weight.1
    F32
    [16, 16, 3, 768]
  • v.position_embd.weight
    F32
    [768, 2304]
  • v.post_ln.bias
    F32
    [768]
  • v.post_ln.weight
    F32
    [768]