The current, most capable model that runs on a single GPU.

1b 4b 12b 27b

654.4K yesterday

... /
model
377655e65351 · 3.3GB
    Metadata
  • gemma3.attention.head_count
    8
  • gemma3.attention.head_count_kv
    4
  • gemma3.attention.key_length
    256
  • gemma3.attention.sliding_window
    1024
  • gemma3.attention.value_length
    256
  • gemma3.block_count
    34
  • gemma3.context_length
    8192
  • gemma3.embedding_length
    2560
  • gemma3.feed_forward_length
    10240
  • gemma3.vision.attention.head_count
    16
  • gemma3.vision.attention.layer_norm_epsilon
    1e-06
  • gemma3.vision.block_count
    27
  • gemma3.vision.embedding_length
    1152
  • gemma3.vision.feed_forward_length
    4304
  • gemma3.vision.image_size
    896
  • gemma3.vision.num_channels
    3
  • gemma3.vision.patch_size
    14
  • general.architecture
    gemma3
  • general.file_type
    15
  • general.quantization_version
    2
  • tokenizer.ggml.add_bos_token
    true
  • tokenizer.ggml.add_eos_token
    false
  • tokenizer.ggml.add_padding_token
    false
  • tokenizer.ggml.add_unknown_token
    false
  • tokenizer.ggml.bos_token_id
    2
  • tokenizer.ggml.eos_token_id
    1
  • tokenizer.ggml.merges
    [ , ▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁ ▁, , , ▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁ ▁▁, ...]
  • tokenizer.ggml.model
    llama
  • tokenizer.ggml.padding_token_id
    0
  • tokenizer.ggml.pre
    default
  • tokenizer.ggml.scores
    [0, 0, 0, 0, 0, ...]
  • tokenizer.ggml.token_type
    [3, 3, 3, 2, 1, ...]
  • tokenizer.ggml.tokens
    [<pad>, <eos>, <bos>, <unk>, <mask>, ...]
  • tokenizer.ggml.unknown_token_id
    3
  • Tensor
  • token_embd.weight
    Q6_K
    [2560, 262144]
  • blk.0
  • blk.0.attn_k.weight
    Q4_K
    [2560, 1024]
  • blk.0.attn_k_norm.weight
    F32
    [256]
  • blk.0.attn_norm.weight
    F32
    [2560]
  • blk.0.attn_output.weight
    Q4_K
    [2048, 2560]
  • blk.0.attn_q.weight
    Q4_K
    [2560, 2048]
  • blk.0.attn_q_norm.weight
    F32
    [256]
  • blk.0.attn_v.weight
    Q6_K
    [2560, 1024]
  • blk.0.ffn_down.weight
    Q6_K
    [10240, 2560]
  • blk.0.ffn_gate.weight
    Q4_K
    [2560, 10240]
  • blk.0.ffn_norm.weight
    F32
    [2560]
  • blk.0.ffn_up.weight
    Q4_K
    [2560, 10240]
  • blk.0.post_attention_norm.weight
    F32
    [2560]
  • blk.0.post_ffw_norm.weight
    F32
    [2560]
  • blk.1
  • blk.1.attn_k.weight
    Q4_K
    [2560, 1024]
  • blk.1.attn_k_norm.weight
    F32
    [256]
  • blk.1.attn_norm.weight
    F32
    [2560]
  • blk.1.attn_output.weight
    Q4_K
    [2048, 2560]
  • blk.1.attn_q.weight
    Q4_K
    [2560, 2048]
  • blk.1.attn_q_norm.weight
    F32
    [256]
  • blk.1.attn_v.weight
    Q6_K
    [2560, 1024]
  • blk.1.ffn_down.weight
    Q6_K
    [10240, 2560]
  • blk.1.ffn_gate.weight
    Q4_K
    [2560, 10240]
  • blk.1.ffn_norm.weight
    F32
    [2560]
  • blk.1.ffn_up.weight
    Q4_K
    [2560, 10240]
  • blk.1.post_attention_norm.weight
    F32
    [2560]
  • blk.1.post_ffw_norm.weight
    F32
    [2560]
  • blk.2
  • blk.2.attn_k.weight
    Q4_K
    [2560, 1024]
  • blk.2.attn_k_norm.weight
    F32
    [256]
  • blk.2.attn_norm.weight
    F32
    [2560]
  • blk.2.attn_output.weight
    Q4_K
    [2048, 2560]
  • blk.2.attn_q.weight
    Q4_K
    [2560, 2048]
  • blk.2.attn_q_norm.weight
    F32
    [256]
  • blk.2.attn_v.weight
    Q6_K
    [2560, 1024]
  • blk.2.ffn_down.weight
    Q6_K
    [10240, 2560]
  • blk.2.ffn_gate.weight
    Q4_K
    [2560, 10240]
  • blk.2.ffn_norm.weight
    F32
    [2560]
  • blk.2.ffn_up.weight
    Q4_K
    [2560, 10240]
  • blk.2.post_attention_norm.weight
    F32
    [2560]
  • blk.2.post_ffw_norm.weight
    F32
    [2560]
  • blk.3
  • blk.3.attn_k.weight
    Q4_K
    [2560, 1024]
  • blk.3.attn_k_norm.weight
    F32
    [256]
  • blk.3.attn_norm.weight
    F32
    [2560]
  • blk.3.attn_output.weight
    Q4_K
    [2048, 2560]
  • blk.3.attn_q.weight
    Q4_K
    [2560, 2048]
  • blk.3.attn_q_norm.weight
    F32
    [256]
  • blk.3.attn_v.weight
    Q6_K
    [2560, 1024]
  • blk.3.ffn_down.weight
    Q6_K
    [10240, 2560]
  • blk.3.ffn_gate.weight
    Q4_K
    [2560, 10240]
  • blk.3.ffn_norm.weight
    F32
    [2560]
  • blk.3.ffn_up.weight
    Q4_K
    [2560, 10240]
  • blk.3.post_attention_norm.weight
    F32
    [2560]
  • blk.3.post_ffw_norm.weight
    F32
    [2560]
  • blk.4
  • blk.4.attn_k.weight
    Q4_K
    [2560, 1024]
  • blk.4.attn_k_norm.weight
    F32
    [256]
  • blk.4.attn_norm.weight
    F32
    [2560]
  • blk.4.attn_output.weight
    Q4_K
    [2048, 2560]
  • blk.4.attn_q.weight
    Q4_K
    [2560, 2048]
  • blk.4.attn_q_norm.weight
    F32
    [256]
  • blk.4.attn_v.weight
    Q6_K
    [2560, 1024]
  • blk.4.ffn_down.weight
    Q4_K
    [10240, 2560]
  • blk.4.ffn_gate.weight
    Q4_K
    [2560, 10240]
  • blk.4.ffn_norm.weight
    F32
    [2560]
  • blk.4.ffn_up.weight
    Q4_K
    [2560, 10240]
  • blk.4.post_attention_norm.weight
    F32
    [2560]
  • blk.4.post_ffw_norm.weight
    F32
    [2560]
  • blk.5
  • blk.5.attn_k.weight
    Q4_K
    [2560, 1024]
  • blk.5.attn_k_norm.weight
    F32
    [256]
  • blk.5.attn_norm.weight
    F32
    [2560]
  • blk.5.attn_output.weight
    Q4_K
    [2048, 2560]
  • blk.5.attn_q.weight
    Q4_K
    [2560, 2048]
  • blk.5.attn_q_norm.weight
    F32
    [256]
  • blk.5.attn_v.weight
    Q6_K
    [2560, 1024]
  • blk.5.ffn_down.weight
    Q4_K
    [10240, 2560]
  • blk.5.ffn_gate.weight
    Q4_K
    [2560, 10240]
  • blk.5.ffn_norm.weight
    F32
    [2560]
  • blk.5.ffn_up.weight
    Q4_K
    [2560, 10240]
  • blk.5.post_attention_norm.weight
    F32
    [2560]
  • blk.5.post_ffw_norm.weight
    F32
    [2560]
  • blk.6
  • blk.6.attn_k.weight
    Q4_K
    [2560, 1024]
  • blk.6.attn_k_norm.weight
    F32
    [256]
  • blk.6.attn_norm.weight
    F32
    [2560]
  • blk.6.attn_output.weight
    Q4_K
    [2048, 2560]
  • blk.6.attn_q.weight
    Q4_K
    [2560, 2048]
  • blk.6.attn_q_norm.weight
    F32
    [256]
  • blk.6.attn_v.weight
    Q6_K
    [2560, 1024]
  • blk.6.ffn_down.weight
    Q6_K
    [10240, 2560]
  • blk.6.ffn_gate.weight
    Q4_K
    [2560, 10240]
  • blk.6.ffn_norm.weight
    F32
    [2560]
  • blk.6.ffn_up.weight
    Q4_K
    [2560, 10240]
  • blk.6.post_attention_norm.weight
    F32
    [2560]
  • blk.6.post_ffw_norm.weight
    F32
    [2560]
  • blk.7
  • blk.7.attn_k.weight
    Q4_K
    [2560, 1024]
  • blk.7.attn_k_norm.weight
    F32
    [256]
  • blk.7.attn_norm.weight
    F32
    [2560]
  • blk.7.attn_output.weight
    Q4_K
    [2048, 2560]
  • blk.7.attn_q.weight
    Q4_K
    [2560, 2048]
  • blk.7.attn_q_norm.weight
    F32
    [256]
  • blk.7.attn_v.weight
    Q4_K
    [2560, 1024]
  • blk.7.ffn_down.weight
    Q4_K
    [10240, 2560]
  • blk.7.ffn_gate.weight
    Q4_K
    [2560, 10240]
  • blk.7.ffn_norm.weight
    F32
    [2560]
  • blk.7.ffn_up.weight
    Q4_K
    [2560, 10240]
  • blk.7.post_attention_norm.weight
    F32
    [2560]
  • blk.7.post_ffw_norm.weight
    F32
    [2560]
  • blk.8
  • blk.8.attn_k.weight
    Q4_K
    [2560, 1024]
  • blk.8.attn_k_norm.weight
    F32
    [256]
  • blk.8.attn_norm.weight
    F32
    [2560]
  • blk.8.attn_output.weight
    Q4_K
    [2048, 2560]
  • blk.8.attn_q.weight
    Q4_K
    [2560, 2048]
  • blk.8.attn_q_norm.weight
    F32
    [256]
  • blk.8.attn_v.weight
    Q4_K
    [2560, 1024]
  • blk.8.ffn_down.weight
    Q4_K
    [10240, 2560]
  • blk.8.ffn_gate.weight
    Q4_K
    [2560, 10240]
  • blk.8.ffn_norm.weight
    F32
    [2560]
  • blk.8.ffn_up.weight
    Q4_K
    [2560, 10240]
  • blk.8.post_attention_norm.weight
    F32
    [2560]
  • blk.8.post_ffw_norm.weight
    F32
    [2560]
  • blk.9
  • blk.9.attn_k.weight
    Q4_K
    [2560, 1024]
  • blk.9.attn_k_norm.weight
    F32
    [256]
  • blk.9.attn_norm.weight
    F32
    [2560]
  • blk.9.attn_output.weight
    Q4_K
    [2048, 2560]
  • blk.9.attn_q.weight
    Q4_K
    [2560, 2048]
  • blk.9.attn_q_norm.weight
    F32
    [256]
  • blk.9.attn_v.weight
    Q6_K
    [2560, 1024]
  • blk.9.ffn_down.weight
    Q6_K
    [10240, 2560]
  • blk.9.ffn_gate.weight
    Q4_K
    [2560, 10240]
  • blk.9.ffn_norm.weight
    F32
    [2560]
  • blk.9.ffn_up.weight
    Q4_K
    [2560, 10240]
  • blk.9.post_attention_norm.weight
    F32
    [2560]
  • blk.9.post_ffw_norm.weight
    F32
    [2560]
  • blk.10
  • blk.10.attn_k.weight
    Q4_K
    [2560, 1024]
  • blk.10.attn_k_norm.weight
    F32
    [256]
  • blk.10.attn_norm.weight
    F32
    [2560]
  • blk.10.attn_output.weight
    Q4_K
    [2048, 2560]
  • blk.10.attn_q.weight
    Q4_K
    [2560, 2048]
  • blk.10.attn_q_norm.weight
    F32
    [256]
  • blk.10.attn_v.weight
    Q4_K
    [2560, 1024]
  • blk.10.ffn_down.weight
    Q4_K
    [10240, 2560]
  • blk.10.ffn_gate.weight
    Q4_K
    [2560, 10240]
  • blk.10.ffn_norm.weight
    F32
    [2560]
  • blk.10.ffn_up.weight
    Q4_K
    [2560, 10240]
  • blk.10.post_attention_norm.weight
    F32
    [2560]
  • blk.10.post_ffw_norm.weight
    F32
    [2560]
  • blk.11
  • blk.11.attn_k.weight
    Q4_K
    [2560, 1024]
  • blk.11.attn_k_norm.weight
    F32
    [256]
  • blk.11.attn_norm.weight
    F32
    [2560]
  • blk.11.attn_output.weight
    Q4_K
    [2048, 2560]
  • blk.11.attn_q.weight
    Q4_K
    [2560, 2048]
  • blk.11.attn_q_norm.weight
    F32
    [256]
  • blk.11.attn_v.weight
    Q4_K
    [2560, 1024]
  • blk.11.ffn_down.weight
    Q4_K
    [10240, 2560]
  • blk.11.ffn_gate.weight
    Q4_K
    [2560, 10240]
  • blk.11.ffn_norm.weight
    F32
    [2560]
  • blk.11.ffn_up.weight
    Q4_K
    [2560, 10240]
  • blk.11.post_attention_norm.weight
    F32
    [2560]
  • blk.11.post_ffw_norm.weight
    F32
    [2560]
  • blk.12
  • blk.12.attn_k.weight
    Q4_K
    [2560, 1024]
  • blk.12.attn_k_norm.weight
    F32
    [256]
  • blk.12.attn_norm.weight
    F32
    [2560]
  • blk.12.attn_output.weight
    Q4_K
    [2048, 2560]
  • blk.12.attn_q.weight
    Q4_K
    [2560, 2048]
  • blk.12.attn_q_norm.weight
    F32
    [256]
  • blk.12.attn_v.weight
    Q6_K
    [2560, 1024]
  • blk.12.ffn_down.weight
    Q6_K
    [10240, 2560]
  • blk.12.ffn_gate.weight
    Q4_K
    [2560, 10240]
  • blk.12.ffn_norm.weight
    F32
    [2560]
  • blk.12.ffn_up.weight
    Q4_K
    [2560, 10240]
  • blk.12.post_attention_norm.weight
    F32
    [2560]
  • blk.12.post_ffw_norm.weight
    F32
    [2560]
  • blk.13
  • blk.13.attn_k.weight
    Q4_K
    [2560, 1024]
  • blk.13.attn_k_norm.weight
    F32
    [256]
  • blk.13.attn_norm.weight
    F32
    [2560]
  • blk.13.attn_output.weight
    Q4_K
    [2048, 2560]
  • blk.13.attn_q.weight
    Q4_K
    [2560, 2048]
  • blk.13.attn_q_norm.weight
    F32
    [256]
  • blk.13.attn_v.weight
    Q4_K
    [2560, 1024]
  • blk.13.ffn_down.weight
    Q4_K
    [10240, 2560]
  • blk.13.ffn_gate.weight
    Q4_K
    [2560, 10240]
  • blk.13.ffn_norm.weight
    F32
    [2560]
  • blk.13.ffn_up.weight
    Q4_K
    [2560, 10240]
  • blk.13.post_attention_norm.weight
    F32
    [2560]
  • blk.13.post_ffw_norm.weight
    F32
    [2560]
  • blk.14
  • blk.14.attn_k.weight
    Q4_K
    [2560, 1024]
  • blk.14.attn_k_norm.weight
    F32
    [256]
  • blk.14.attn_norm.weight
    F32
    [2560]
  • blk.14.attn_output.weight
    Q4_K
    [2048, 2560]
  • blk.14.attn_q.weight
    Q4_K
    [2560, 2048]
  • blk.14.attn_q_norm.weight
    F32
    [256]
  • blk.14.attn_v.weight
    Q4_K
    [2560, 1024]
  • blk.14.ffn_down.weight
    Q4_K
    [10240, 2560]
  • blk.14.ffn_gate.weight
    Q4_K
    [2560, 10240]
  • blk.14.ffn_norm.weight
    F32
    [2560]
  • blk.14.ffn_up.weight
    Q4_K
    [2560, 10240]
  • blk.14.post_attention_norm.weight
    F32
    [2560]
  • blk.14.post_ffw_norm.weight
    F32
    [2560]
  • blk.15
  • blk.15.attn_k.weight
    Q4_K
    [2560, 1024]
  • blk.15.attn_k_norm.weight
    F32
    [256]
  • blk.15.attn_norm.weight
    F32
    [2560]
  • blk.15.attn_output.weight
    Q4_K
    [2048, 2560]
  • blk.15.attn_q.weight
    Q4_K
    [2560, 2048]
  • blk.15.attn_q_norm.weight
    F32
    [256]
  • blk.15.attn_v.weight
    Q6_K
    [2560, 1024]
  • blk.15.ffn_down.weight
    Q6_K
    [10240, 2560]
  • blk.15.ffn_gate.weight
    Q4_K
    [2560, 10240]
  • blk.15.ffn_norm.weight
    F32
    [2560]
  • blk.15.ffn_up.weight
    Q4_K
    [2560, 10240]
  • blk.15.post_attention_norm.weight
    F32
    [2560]
  • blk.15.post_ffw_norm.weight
    F32
    [2560]
  • blk.16
  • blk.16.attn_k.weight
    Q4_K
    [2560, 1024]
  • blk.16.attn_k_norm.weight
    F32
    [256]
  • blk.16.attn_norm.weight
    F32
    [2560]
  • blk.16.attn_output.weight
    Q4_K
    [2048, 2560]
  • blk.16.attn_q.weight
    Q4_K
    [2560, 2048]
  • blk.16.attn_q_norm.weight
    F32
    [256]
  • blk.16.attn_v.weight
    Q4_K
    [2560, 1024]
  • blk.16.ffn_down.weight
    Q4_K
    [10240, 2560]
  • blk.16.ffn_gate.weight
    Q4_K
    [2560, 10240]
  • blk.16.ffn_norm.weight
    F32
    [2560]
  • blk.16.ffn_up.weight
    Q4_K
    [2560, 10240]
  • blk.16.post_attention_norm.weight
    F32
    [2560]
  • blk.16.post_ffw_norm.weight
    F32
    [2560]
  • blk.17
  • blk.17.attn_k.weight
    Q4_K
    [2560, 1024]
  • blk.17.attn_k_norm.weight
    F32
    [256]
  • blk.17.attn_norm.weight
    F32
    [2560]
  • blk.17.attn_output.weight
    Q4_K
    [2048, 2560]
  • blk.17.attn_q.weight
    Q4_K
    [2560, 2048]
  • blk.17.attn_q_norm.weight
    F32
    [256]
  • blk.17.attn_v.weight
    Q4_K
    [2560, 1024]
  • blk.17.ffn_down.weight
    Q4_K
    [10240, 2560]
  • blk.17.ffn_gate.weight
    Q4_K
    [2560, 10240]
  • blk.17.ffn_norm.weight
    F32
    [2560]
  • blk.17.ffn_up.weight
    Q4_K
    [2560, 10240]
  • blk.17.post_attention_norm.weight
    F32
    [2560]
  • blk.17.post_ffw_norm.weight
    F32
    [2560]
  • blk.18
  • blk.18.attn_k.weight
    Q4_K
    [2560, 1024]
  • blk.18.attn_k_norm.weight
    F32
    [256]
  • blk.18.attn_norm.weight
    F32
    [2560]
  • blk.18.attn_output.weight
    Q4_K
    [2048, 2560]
  • blk.18.attn_q.weight
    Q4_K
    [2560, 2048]
  • blk.18.attn_q_norm.weight
    F32
    [256]
  • blk.18.attn_v.weight
    Q6_K
    [2560, 1024]
  • blk.18.ffn_down.weight
    Q6_K
    [10240, 2560]
  • blk.18.ffn_gate.weight
    Q4_K
    [2560, 10240]
  • blk.18.ffn_norm.weight
    F32
    [2560]
  • blk.18.ffn_up.weight
    Q4_K
    [2560, 10240]
  • blk.18.post_attention_norm.weight
    F32
    [2560]
  • blk.18.post_ffw_norm.weight
    F32
    [2560]
  • blk.19
  • blk.19.attn_k.weight
    Q4_K
    [2560, 1024]
  • blk.19.attn_k_norm.weight
    F32
    [256]
  • blk.19.attn_norm.weight
    F32
    [2560]
  • blk.19.attn_output.weight
    Q4_K
    [2048, 2560]
  • blk.19.attn_q.weight
    Q4_K
    [2560, 2048]
  • blk.19.attn_q_norm.weight
    F32
    [256]
  • blk.19.attn_v.weight
    Q4_K
    [2560, 1024]
  • blk.19.ffn_down.weight
    Q4_K
    [10240, 2560]
  • blk.19.ffn_gate.weight
    Q4_K
    [2560, 10240]
  • blk.19.ffn_norm.weight
    F32
    [2560]
  • blk.19.ffn_up.weight
    Q4_K
    [2560, 10240]
  • blk.19.post_attention_norm.weight
    F32
    [2560]
  • blk.19.post_ffw_norm.weight
    F32
    [2560]
  • blk.20
  • blk.20.attn_k.weight
    Q4_K
    [2560, 1024]
  • blk.20.attn_k_norm.weight
    F32
    [256]
  • blk.20.attn_norm.weight
    F32
    [2560]
  • blk.20.attn_output.weight
    Q4_K
    [2048, 2560]
  • blk.20.attn_q.weight
    Q4_K
    [2560, 2048]
  • blk.20.attn_q_norm.weight
    F32
    [256]
  • blk.20.attn_v.weight
    Q4_K
    [2560, 1024]
  • blk.20.ffn_down.weight
    Q4_K
    [10240, 2560]
  • blk.20.ffn_gate.weight
    Q4_K
    [2560, 10240]
  • blk.20.ffn_norm.weight
    F32
    [2560]
  • blk.20.ffn_up.weight
    Q4_K
    [2560, 10240]
  • blk.20.post_attention_norm.weight
    F32
    [2560]
  • blk.20.post_ffw_norm.weight
    F32
    [2560]
  • blk.21
  • blk.21.attn_k.weight
    Q4_K
    [2560, 1024]
  • blk.21.attn_k_norm.weight
    F32
    [256]
  • blk.21.attn_norm.weight
    F32
    [2560]
  • blk.21.attn_output.weight
    Q4_K
    [2048, 2560]
  • blk.21.attn_q.weight
    Q4_K
    [2560, 2048]
  • blk.21.attn_q_norm.weight
    F32
    [256]
  • blk.21.attn_v.weight
    Q6_K
    [2560, 1024]
  • blk.21.ffn_down.weight
    Q6_K
    [10240, 2560]
  • blk.21.ffn_gate.weight
    Q4_K
    [2560, 10240]
  • blk.21.ffn_norm.weight
    F32
    [2560]
  • blk.21.ffn_up.weight
    Q4_K
    [2560, 10240]
  • blk.21.post_attention_norm.weight
    F32
    [2560]
  • blk.21.post_ffw_norm.weight
    F32
    [2560]
  • blk.22
  • blk.22.attn_k.weight
    Q4_K
    [2560, 1024]
  • blk.22.attn_k_norm.weight
    F32
    [256]
  • blk.22.attn_norm.weight
    F32
    [2560]
  • blk.22.attn_output.weight
    Q4_K
    [2048, 2560]
  • blk.22.attn_q.weight
    Q4_K
    [2560, 2048]
  • blk.22.attn_q_norm.weight
    F32
    [256]
  • blk.22.attn_v.weight
    Q4_K
    [2560, 1024]
  • blk.22.ffn_down.weight
    Q4_K
    [10240, 2560]
  • blk.22.ffn_gate.weight
    Q4_K
    [2560, 10240]
  • blk.22.ffn_norm.weight
    F32
    [2560]
  • blk.22.ffn_up.weight
    Q4_K
    [2560, 10240]
  • blk.22.post_attention_norm.weight
    F32
    [2560]
  • blk.22.post_ffw_norm.weight
    F32
    [2560]
  • blk.23
  • blk.23.attn_k.weight
    Q4_K
    [2560, 1024]
  • blk.23.attn_k_norm.weight
    F32
    [256]
  • blk.23.attn_norm.weight
    F32
    [2560]
  • blk.23.attn_output.weight
    Q4_K
    [2048, 2560]
  • blk.23.attn_q.weight
    Q4_K
    [2560, 2048]
  • blk.23.attn_q_norm.weight
    F32
    [256]
  • blk.23.attn_v.weight
    Q4_K
    [2560, 1024]
  • blk.23.ffn_down.weight
    Q4_K
    [10240, 2560]
  • blk.23.ffn_gate.weight
    Q4_K
    [2560, 10240]
  • blk.23.ffn_norm.weight
    F32
    [2560]
  • blk.23.ffn_up.weight
    Q4_K
    [2560, 10240]
  • blk.23.post_attention_norm.weight
    F32
    [2560]
  • blk.23.post_ffw_norm.weight
    F32
    [2560]
  • blk.24
  • blk.24.attn_k.weight
    Q4_K
    [2560, 1024]
  • blk.24.attn_k_norm.weight
    F32
    [256]
  • blk.24.attn_norm.weight
    F32
    [2560]
  • blk.24.attn_output.weight
    Q4_K
    [2048, 2560]
  • blk.24.attn_q.weight
    Q4_K
    [2560, 2048]
  • blk.24.attn_q_norm.weight
    F32
    [256]
  • blk.24.attn_v.weight
    Q6_K
    [2560, 1024]
  • blk.24.ffn_down.weight
    Q6_K
    [10240, 2560]
  • blk.24.ffn_gate.weight
    Q4_K
    [2560, 10240]
  • blk.24.ffn_norm.weight
    F32
    [2560]
  • blk.24.ffn_up.weight
    Q4_K
    [2560, 10240]
  • blk.24.post_attention_norm.weight
    F32
    [2560]
  • blk.24.post_ffw_norm.weight
    F32
    [2560]
  • blk.25
  • blk.25.attn_k.weight
    Q4_K
    [2560, 1024]
  • blk.25.attn_k_norm.weight
    F32
    [256]
  • blk.25.attn_norm.weight
    F32
    [2560]
  • blk.25.attn_output.weight
    Q4_K
    [2048, 2560]
  • blk.25.attn_q.weight
    Q4_K
    [2560, 2048]
  • blk.25.attn_q_norm.weight
    F32
    [256]
  • blk.25.attn_v.weight
    Q4_K
    [2560, 1024]
  • blk.25.ffn_down.weight
    Q4_K
    [10240, 2560]
  • blk.25.ffn_gate.weight
    Q4_K
    [2560, 10240]
  • blk.25.ffn_norm.weight
    F32
    [2560]
  • blk.25.ffn_up.weight
    Q4_K
    [2560, 10240]
  • blk.25.post_attention_norm.weight
    F32
    [2560]
  • blk.25.post_ffw_norm.weight
    F32
    [2560]
  • blk.26
  • blk.26.attn_k.weight
    Q4_K
    [2560, 1024]
  • blk.26.attn_k_norm.weight
    F32
    [256]
  • blk.26.attn_norm.weight
    F32
    [2560]
  • blk.26.attn_output.weight
    Q4_K
    [2048, 2560]
  • blk.26.attn_q.weight
    Q4_K
    [2560, 2048]
  • blk.26.attn_q_norm.weight
    F32
    [256]
  • blk.26.attn_v.weight
    Q4_K
    [2560, 1024]
  • blk.26.ffn_down.weight
    Q4_K
    [10240, 2560]
  • blk.26.ffn_gate.weight
    Q4_K
    [2560, 10240]
  • blk.26.ffn_norm.weight
    F32
    [2560]
  • blk.26.ffn_up.weight
    Q4_K
    [2560, 10240]
  • blk.26.post_attention_norm.weight
    F32
    [2560]
  • blk.26.post_ffw_norm.weight
    F32
    [2560]
  • blk.27
  • blk.27.attn_k.weight
    Q4_K
    [2560, 1024]
  • blk.27.attn_k_norm.weight
    F32
    [256]
  • blk.27.attn_norm.weight
    F32
    [2560]
  • blk.27.attn_output.weight
    Q4_K
    [2048, 2560]
  • blk.27.attn_q.weight
    Q4_K
    [2560, 2048]
  • blk.27.attn_q_norm.weight
    F32
    [256]
  • blk.27.attn_v.weight
    Q6_K
    [2560, 1024]
  • blk.27.ffn_down.weight
    Q6_K
    [10240, 2560]
  • blk.27.ffn_gate.weight
    Q4_K
    [2560, 10240]
  • blk.27.ffn_norm.weight
    F32
    [2560]
  • blk.27.ffn_up.weight
    Q4_K
    [2560, 10240]
  • blk.27.post_attention_norm.weight
    F32
    [2560]
  • blk.27.post_ffw_norm.weight
    F32
    [2560]
  • blk.28
  • blk.28.attn_k.weight
    Q4_K
    [2560, 1024]
  • blk.28.attn_k_norm.weight
    F32
    [256]
  • blk.28.attn_norm.weight
    F32
    [2560]
  • blk.28.attn_output.weight
    Q4_K
    [2048, 2560]
  • blk.28.attn_q.weight
    Q4_K
    [2560, 2048]
  • blk.28.attn_q_norm.weight
    F32
    [256]
  • blk.28.attn_v.weight
    Q4_K
    [2560, 1024]
  • blk.28.ffn_down.weight
    Q4_K
    [10240, 2560]
  • blk.28.ffn_gate.weight
    Q4_K
    [2560, 10240]
  • blk.28.ffn_norm.weight
    F32
    [2560]
  • blk.28.ffn_up.weight
    Q4_K
    [2560, 10240]
  • blk.28.post_attention_norm.weight
    F32
    [2560]
  • blk.28.post_ffw_norm.weight
    F32
    [2560]
  • blk.29
  • blk.29.attn_k.weight
    Q4_K
    [2560, 1024]
  • blk.29.attn_k_norm.weight
    F32
    [256]
  • blk.29.attn_norm.weight
    F32
    [2560]
  • blk.29.attn_output.weight
    Q4_K
    [2048, 2560]
  • blk.29.attn_q.weight
    Q4_K
    [2560, 2048]
  • blk.29.attn_q_norm.weight
    F32
    [256]
  • blk.29.attn_v.weight
    Q4_K
    [2560, 1024]
  • blk.29.ffn_down.weight
    Q6_K
    [10240, 2560]
  • blk.29.ffn_gate.weight
    Q4_K
    [2560, 10240]
  • blk.29.ffn_norm.weight
    F32
    [2560]
  • blk.29.ffn_up.weight
    Q4_K
    [2560, 10240]
  • blk.29.post_attention_norm.weight
    F32
    [2560]
  • blk.29.post_ffw_norm.weight
    F32
    [2560]
  • blk.30
  • blk.30.attn_k.weight
    Q4_K
    [2560, 1024]
  • blk.30.attn_k_norm.weight
    F32
    [256]
  • blk.30.attn_norm.weight
    F32
    [2560]
  • blk.30.attn_output.weight
    Q4_K
    [2048, 2560]
  • blk.30.attn_q.weight
    Q4_K
    [2560, 2048]
  • blk.30.attn_q_norm.weight
    F32
    [256]
  • blk.30.attn_v.weight
    Q6_K
    [2560, 1024]
  • blk.30.ffn_down.weight
    Q6_K
    [10240, 2560]
  • blk.30.ffn_gate.weight
    Q4_K
    [2560, 10240]
  • blk.30.ffn_norm.weight
    F32
    [2560]
  • blk.30.ffn_up.weight
    Q4_K
    [2560, 10240]
  • blk.30.post_attention_norm.weight
    F32
    [2560]
  • blk.30.post_ffw_norm.weight
    F32
    [2560]
  • blk.31
  • blk.31.attn_k.weight
    Q4_K
    [2560, 1024]
  • blk.31.attn_k_norm.weight
    F32
    [256]
  • blk.31.attn_norm.weight
    F32
    [2560]
  • blk.31.attn_output.weight
    Q4_K
    [2048, 2560]
  • blk.31.attn_q.weight
    Q4_K
    [2560, 2048]
  • blk.31.attn_q_norm.weight
    F32
    [256]
  • blk.31.attn_v.weight
    Q4_K
    [2560, 1024]
  • blk.31.ffn_down.weight
    Q6_K
    [10240, 2560]
  • blk.31.ffn_gate.weight
    Q4_K
    [2560, 10240]
  • blk.31.ffn_norm.weight
    F32
    [2560]
  • blk.31.ffn_up.weight
    Q4_K
    [2560, 10240]
  • blk.31.post_attention_norm.weight
    F32
    [2560]
  • blk.31.post_ffw_norm.weight
    F32
    [2560]
  • blk.32
  • blk.32.attn_k.weight
    Q4_K
    [2560, 1024]
  • blk.32.attn_k_norm.weight
    F32
    [256]
  • blk.32.attn_norm.weight
    F32
    [2560]
  • blk.32.attn_output.weight
    Q4_K
    [2048, 2560]
  • blk.32.attn_q.weight
    Q4_K
    [2560, 2048]
  • blk.32.attn_q_norm.weight
    F32
    [256]
  • blk.32.attn_v.weight
    Q4_K
    [2560, 1024]
  • blk.32.ffn_down.weight
    Q6_K
    [10240, 2560]
  • blk.32.ffn_gate.weight
    Q4_K
    [2560, 10240]
  • blk.32.ffn_norm.weight
    F32
    [2560]
  • blk.32.ffn_up.weight
    Q4_K
    [2560, 10240]
  • blk.32.post_attention_norm.weight
    F32
    [2560]
  • blk.32.post_ffw_norm.weight
    F32
    [2560]
  • blk.33
  • blk.33.attn_k.weight
    Q4_K
    [2560, 1024]
  • blk.33.attn_k_norm.weight
    F32
    [256]
  • blk.33.attn_norm.weight
    F32
    [2560]
  • blk.33.attn_output.weight
    Q4_K
    [2048, 2560]
  • blk.33.attn_q.weight
    Q4_K
    [2560, 2048]
  • blk.33.attn_q_norm.weight
    F32
    [256]
  • blk.33.attn_v.weight
    Q6_K
    [2560, 1024]
  • blk.33.ffn_down.weight
    Q6_K
    [10240, 2560]
  • blk.33.ffn_gate.weight
    Q4_K
    [2560, 10240]
  • blk.33.ffn_norm.weight
    F32
    [2560]
  • blk.33.ffn_up.weight
    Q4_K
    [2560, 10240]
  • blk.33.post_attention_norm.weight
    F32
    [2560]
  • blk.33.post_ffw_norm.weight
    F32
    [2560]
  • mm.mm_input_projection.weight
    F16
    [2560, 1152]
  • mm.mm_soft_emb_norm.weight
    F32
    [1152]
  • v.blk.0
  • v.blk.0.attn_k.bias
    F32
    [1152]
  • v.blk.0.attn_k.weight
    F16
    [1152, 1152]
  • v.blk.0.attn_output.bias
    F32
    [1152]
  • v.blk.0.attn_output.weight
    F16
    [1152, 1152]
  • v.blk.0.attn_q.bias
    F32
    [1152]
  • v.blk.0.attn_q.weight
    F16
    [1152, 1152]
  • v.blk.0.attn_v.bias
    F32
    [1152]
  • v.blk.0.attn_v.weight
    F16
    [1152, 1152]
  • v.blk.0.layer_norm1.bias
    F32
    [1152]
  • v.blk.0.layer_norm1.weight
    F32
    [1152]
  • v.blk.0.layer_norm2.bias
    F32
    [1152]
  • v.blk.0.layer_norm2.weight
    F32
    [1152]
  • v.blk.0.mlp.fc1.bias
    F32
    [4304]
  • v.blk.0.mlp.fc1.weight
    F16
    [1152, 4304]
  • v.blk.0.mlp.fc2.bias
    F32
    [1152]
  • v.blk.0.mlp.fc2.weight
    F16
    [4304, 1152]
  • v.blk.1
  • v.blk.1.attn_k.bias
    F32
    [1152]
  • v.blk.1.attn_k.weight
    F16
    [1152, 1152]
  • v.blk.1.attn_output.bias
    F32
    [1152]
  • v.blk.1.attn_output.weight
    F16
    [1152, 1152]
  • v.blk.1.attn_q.bias
    F32
    [1152]
  • v.blk.1.attn_q.weight
    F16
    [1152, 1152]
  • v.blk.1.attn_v.bias
    F32
    [1152]
  • v.blk.1.attn_v.weight
    F16
    [1152, 1152]
  • v.blk.1.layer_norm1.bias
    F32
    [1152]
  • v.blk.1.layer_norm1.weight
    F32
    [1152]
  • v.blk.1.layer_norm2.bias
    F32
    [1152]
  • v.blk.1.layer_norm2.weight
    F32
    [1152]
  • v.blk.1.mlp.fc1.bias
    F32
    [4304]
  • v.blk.1.mlp.fc1.weight
    F16
    [1152, 4304]
  • v.blk.1.mlp.fc2.bias
    F32
    [1152]
  • v.blk.1.mlp.fc2.weight
    F16
    [4304, 1152]
  • v.blk.2
  • v.blk.2.attn_k.bias
    F32
    [1152]
  • v.blk.2.attn_k.weight
    F16
    [1152, 1152]
  • v.blk.2.attn_output.bias
    F32
    [1152]
  • v.blk.2.attn_output.weight
    F16
    [1152, 1152]
  • v.blk.2.attn_q.bias
    F32
    [1152]
  • v.blk.2.attn_q.weight
    F16
    [1152, 1152]
  • v.blk.2.attn_v.bias
    F32
    [1152]
  • v.blk.2.attn_v.weight
    F16
    [1152, 1152]
  • v.blk.2.layer_norm1.bias
    F32
    [1152]
  • v.blk.2.layer_norm1.weight
    F32
    [1152]
  • v.blk.2.layer_norm2.bias
    F32
    [1152]
  • v.blk.2.layer_norm2.weight
    F32
    [1152]
  • v.blk.2.mlp.fc1.bias
    F32
    [4304]
  • v.blk.2.mlp.fc1.weight
    F16
    [1152, 4304]
  • v.blk.2.mlp.fc2.bias
    F32
    [1152]
  • v.blk.2.mlp.fc2.weight
    F16
    [4304, 1152]
  • v.blk.3
  • v.blk.3.attn_k.bias
    F32
    [1152]
  • v.blk.3.attn_k.weight
    F16
    [1152, 1152]
  • v.blk.3.attn_output.bias
    F32
    [1152]
  • v.blk.3.attn_output.weight
    F16
    [1152, 1152]
  • v.blk.3.attn_q.bias
    F32
    [1152]
  • v.blk.3.attn_q.weight
    F16
    [1152, 1152]
  • v.blk.3.attn_v.bias
    F32
    [1152]
  • v.blk.3.attn_v.weight
    F16
    [1152, 1152]
  • v.blk.3.layer_norm1.bias
    F32
    [1152]
  • v.blk.3.layer_norm1.weight
    F32
    [1152]
  • v.blk.3.layer_norm2.bias
    F32
    [1152]
  • v.blk.3.layer_norm2.weight
    F32
    [1152]
  • v.blk.3.mlp.fc1.bias
    F32
    [4304]
  • v.blk.3.mlp.fc1.weight
    F16
    [1152, 4304]
  • v.blk.3.mlp.fc2.bias
    F32
    [1152]
  • v.blk.3.mlp.fc2.weight
    F16
    [4304, 1152]
  • v.blk.4
  • v.blk.4.attn_k.bias
    F32
    [1152]
  • v.blk.4.attn_k.weight
    F16
    [1152, 1152]
  • v.blk.4.attn_output.bias
    F32
    [1152]
  • v.blk.4.attn_output.weight
    F16
    [1152, 1152]
  • v.blk.4.attn_q.bias
    F32
    [1152]
  • v.blk.4.attn_q.weight
    F16
    [1152, 1152]
  • v.blk.4.attn_v.bias
    F32
    [1152]
  • v.blk.4.attn_v.weight
    F16
    [1152, 1152]
  • v.blk.4.layer_norm1.bias
    F32
    [1152]
  • v.blk.4.layer_norm1.weight
    F32
    [1152]
  • v.blk.4.layer_norm2.bias
    F32
    [1152]
  • v.blk.4.layer_norm2.weight
    F32
    [1152]
  • v.blk.4.mlp.fc1.bias
    F32
    [4304]
  • v.blk.4.mlp.fc1.weight
    F16
    [1152, 4304]
  • v.blk.4.mlp.fc2.bias
    F32
    [1152]
  • v.blk.4.mlp.fc2.weight
    F16
    [4304, 1152]
  • v.blk.5
  • v.blk.5.attn_k.bias
    F32
    [1152]
  • v.blk.5.attn_k.weight
    F16
    [1152, 1152]
  • v.blk.5.attn_output.bias
    F32
    [1152]
  • v.blk.5.attn_output.weight
    F16
    [1152, 1152]
  • v.blk.5.attn_q.bias
    F32
    [1152]
  • v.blk.5.attn_q.weight
    F16
    [1152, 1152]
  • v.blk.5.attn_v.bias
    F32
    [1152]
  • v.blk.5.attn_v.weight
    F16
    [1152, 1152]
  • v.blk.5.layer_norm1.bias
    F32
    [1152]
  • v.blk.5.layer_norm1.weight
    F32
    [1152]
  • v.blk.5.layer_norm2.bias
    F32
    [1152]
  • v.blk.5.layer_norm2.weight
    F32
    [1152]
  • v.blk.5.mlp.fc1.bias
    F32
    [4304]
  • v.blk.5.mlp.fc1.weight
    F16
    [1152, 4304]
  • v.blk.5.mlp.fc2.bias
    F32
    [1152]
  • v.blk.5.mlp.fc2.weight
    F16
    [4304, 1152]
  • v.blk.6
  • v.blk.6.attn_k.bias
    F32
    [1152]
  • v.blk.6.attn_k.weight
    F16
    [1152, 1152]
  • v.blk.6.attn_output.bias
    F32
    [1152]
  • v.blk.6.attn_output.weight
    F16
    [1152, 1152]
  • v.blk.6.attn_q.bias
    F32
    [1152]
  • v.blk.6.attn_q.weight
    F16
    [1152, 1152]
  • v.blk.6.attn_v.bias
    F32
    [1152]
  • v.blk.6.attn_v.weight
    F16
    [1152, 1152]
  • v.blk.6.layer_norm1.bias
    F32
    [1152]
  • v.blk.6.layer_norm1.weight
    F32
    [1152]
  • v.blk.6.layer_norm2.bias
    F32
    [1152]
  • v.blk.6.layer_norm2.weight
    F32
    [1152]
  • v.blk.6.mlp.fc1.bias
    F32
    [4304]
  • v.blk.6.mlp.fc1.weight
    F16
    [1152, 4304]
  • v.blk.6.mlp.fc2.bias
    F32
    [1152]
  • v.blk.6.mlp.fc2.weight
    F16
    [4304, 1152]
  • v.blk.7
  • v.blk.7.attn_k.bias
    F32
    [1152]
  • v.blk.7.attn_k.weight
    F16
    [1152, 1152]
  • v.blk.7.attn_output.bias
    F32
    [1152]
  • v.blk.7.attn_output.weight
    F16
    [1152, 1152]
  • v.blk.7.attn_q.bias
    F32
    [1152]
  • v.blk.7.attn_q.weight
    F16
    [1152, 1152]
  • v.blk.7.attn_v.bias
    F32
    [1152]
  • v.blk.7.attn_v.weight
    F16
    [1152, 1152]
  • v.blk.7.layer_norm1.bias
    F32
    [1152]
  • v.blk.7.layer_norm1.weight
    F32
    [1152]
  • v.blk.7.layer_norm2.bias
    F32
    [1152]
  • v.blk.7.layer_norm2.weight
    F32
    [1152]
  • v.blk.7.mlp.fc1.bias
    F32
    [4304]
  • v.blk.7.mlp.fc1.weight
    F16
    [1152, 4304]
  • v.blk.7.mlp.fc2.bias
    F32
    [1152]
  • v.blk.7.mlp.fc2.weight
    F16
    [4304, 1152]
  • v.blk.8
  • v.blk.8.attn_k.bias
    F32
    [1152]
  • v.blk.8.attn_k.weight
    F16
    [1152, 1152]
  • v.blk.8.attn_output.bias
    F32
    [1152]
  • v.blk.8.attn_output.weight
    F16
    [1152, 1152]
  • v.blk.8.attn_q.bias
    F32
    [1152]
  • v.blk.8.attn_q.weight
    F16
    [1152, 1152]
  • v.blk.8.attn_v.bias
    F32
    [1152]
  • v.blk.8.attn_v.weight
    F16
    [1152, 1152]
  • v.blk.8.layer_norm1.bias
    F32
    [1152]
  • v.blk.8.layer_norm1.weight
    F32
    [1152]
  • v.blk.8.layer_norm2.bias
    F32
    [1152]
  • v.blk.8.layer_norm2.weight
    F32
    [1152]
  • v.blk.8.mlp.fc1.bias
    F32
    [4304]
  • v.blk.8.mlp.fc1.weight
    F16
    [1152, 4304]
  • v.blk.8.mlp.fc2.bias
    F32
    [1152]
  • v.blk.8.mlp.fc2.weight
    F16
    [4304, 1152]
  • v.blk.9
  • v.blk.9.attn_k.bias
    F32
    [1152]
  • v.blk.9.attn_k.weight
    F16
    [1152, 1152]
  • v.blk.9.attn_output.bias
    F32
    [1152]
  • v.blk.9.attn_output.weight
    F16
    [1152, 1152]
  • v.blk.9.attn_q.bias
    F32
    [1152]
  • v.blk.9.attn_q.weight
    F16
    [1152, 1152]
  • v.blk.9.attn_v.bias
    F32
    [1152]
  • v.blk.9.attn_v.weight
    F16
    [1152, 1152]
  • v.blk.9.layer_norm1.bias
    F32
    [1152]
  • v.blk.9.layer_norm1.weight
    F32
    [1152]
  • v.blk.9.layer_norm2.bias
    F32
    [1152]
  • v.blk.9.layer_norm2.weight
    F32
    [1152]
  • v.blk.9.mlp.fc1.bias
    F32
    [4304]
  • v.blk.9.mlp.fc1.weight
    F16
    [1152, 4304]
  • v.blk.9.mlp.fc2.bias
    F32
    [1152]
  • v.blk.9.mlp.fc2.weight
    F16
    [4304, 1152]
  • v.blk.10
  • v.blk.10.attn_k.bias
    F32
    [1152]
  • v.blk.10.attn_k.weight
    F16
    [1152, 1152]
  • v.blk.10.attn_output.bias
    F32
    [1152]
  • v.blk.10.attn_output.weight
    F16
    [1152, 1152]
  • v.blk.10.attn_q.bias
    F32
    [1152]
  • v.blk.10.attn_q.weight
    F16
    [1152, 1152]
  • v.blk.10.attn_v.bias
    F32
    [1152]
  • v.blk.10.attn_v.weight
    F16
    [1152, 1152]
  • v.blk.10.layer_norm1.bias
    F32
    [1152]
  • v.blk.10.layer_norm1.weight
    F32
    [1152]
  • v.blk.10.layer_norm2.bias
    F32
    [1152]
  • v.blk.10.layer_norm2.weight
    F32
    [1152]
  • v.blk.10.mlp.fc1.bias
    F32
    [4304]
  • v.blk.10.mlp.fc1.weight
    F16
    [1152, 4304]
  • v.blk.10.mlp.fc2.bias
    F32
    [1152]
  • v.blk.10.mlp.fc2.weight
    F16
    [4304, 1152]
  • v.blk.11
  • v.blk.11.attn_k.bias
    F32
    [1152]
  • v.blk.11.attn_k.weight
    F16
    [1152, 1152]
  • v.blk.11.attn_output.bias
    F32
    [1152]
  • v.blk.11.attn_output.weight
    F16
    [1152, 1152]
  • v.blk.11.attn_q.bias
    F32
    [1152]
  • v.blk.11.attn_q.weight
    F16
    [1152, 1152]
  • v.blk.11.attn_v.bias
    F32
    [1152]
  • v.blk.11.attn_v.weight
    F16
    [1152, 1152]
  • v.blk.11.layer_norm1.bias
    F32
    [1152]
  • v.blk.11.layer_norm1.weight
    F32
    [1152]
  • v.blk.11.layer_norm2.bias
    F32
    [1152]
  • v.blk.11.layer_norm2.weight
    F32
    [1152]
  • v.blk.11.mlp.fc1.bias
    F32
    [4304]
  • v.blk.11.mlp.fc1.weight
    F16
    [1152, 4304]
  • v.blk.11.mlp.fc2.bias
    F32
    [1152]
  • v.blk.11.mlp.fc2.weight
    F16
    [4304, 1152]
  • v.blk.12
  • v.blk.12.attn_k.bias
    F32
    [1152]
  • v.blk.12.attn_k.weight
    F16
    [1152, 1152]
  • v.blk.12.attn_output.bias
    F32
    [1152]
  • v.blk.12.attn_output.weight
    F16
    [1152, 1152]
  • v.blk.12.attn_q.bias
    F32
    [1152]
  • v.blk.12.attn_q.weight
    F16
    [1152, 1152]
  • v.blk.12.attn_v.bias
    F32
    [1152]
  • v.blk.12.attn_v.weight
    F16
    [1152, 1152]
  • v.blk.12.layer_norm1.bias
    F32
    [1152]
  • v.blk.12.layer_norm1.weight
    F32
    [1152]
  • v.blk.12.layer_norm2.bias
    F32
    [1152]
  • v.blk.12.layer_norm2.weight
    F32
    [1152]
  • v.blk.12.mlp.fc1.bias
    F32
    [4304]
  • v.blk.12.mlp.fc1.weight
    F16
    [1152, 4304]
  • v.blk.12.mlp.fc2.bias
    F32
    [1152]
  • v.blk.12.mlp.fc2.weight
    F16
    [4304, 1152]
  • v.blk.13
  • v.blk.13.attn_k.bias
    F32
    [1152]
  • v.blk.13.attn_k.weight
    F16
    [1152, 1152]
  • v.blk.13.attn_output.bias
    F32
    [1152]
  • v.blk.13.attn_output.weight
    F16
    [1152, 1152]
  • v.blk.13.attn_q.bias
    F32
    [1152]
  • v.blk.13.attn_q.weight
    F16
    [1152, 1152]
  • v.blk.13.attn_v.bias
    F32
    [1152]
  • v.blk.13.attn_v.weight
    F16
    [1152, 1152]
  • v.blk.13.layer_norm1.bias
    F32
    [1152]
  • v.blk.13.layer_norm1.weight
    F32
    [1152]
  • v.blk.13.layer_norm2.bias
    F32
    [1152]
  • v.blk.13.layer_norm2.weight
    F32
    [1152]
  • v.blk.13.mlp.fc1.bias
    F32
    [4304]
  • v.blk.13.mlp.fc1.weight
    F16
    [1152, 4304]
  • v.blk.13.mlp.fc2.bias
    F32
    [1152]
  • v.blk.13.mlp.fc2.weight
    F16
    [4304, 1152]
  • v.blk.14
  • v.blk.14.attn_k.bias
    F32
    [1152]
  • v.blk.14.attn_k.weight
    F16
    [1152, 1152]
  • v.blk.14.attn_output.bias
    F32
    [1152]
  • v.blk.14.attn_output.weight
    F16
    [1152, 1152]
  • v.blk.14.attn_q.bias
    F32
    [1152]
  • v.blk.14.attn_q.weight
    F16
    [1152, 1152]
  • v.blk.14.attn_v.bias
    F32
    [1152]
  • v.blk.14.attn_v.weight
    F16
    [1152, 1152]
  • v.blk.14.layer_norm1.bias
    F32
    [1152]
  • v.blk.14.layer_norm1.weight
    F32
    [1152]
  • v.blk.14.layer_norm2.bias
    F32
    [1152]
  • v.blk.14.layer_norm2.weight
    F32
    [1152]
  • v.blk.14.mlp.fc1.bias
    F32
    [4304]
  • v.blk.14.mlp.fc1.weight
    F16
    [1152, 4304]
  • v.blk.14.mlp.fc2.bias
    F32
    [1152]
  • v.blk.14.mlp.fc2.weight
    F16
    [4304, 1152]
  • v.blk.15
  • v.blk.15.attn_k.bias
    F32
    [1152]
  • v.blk.15.attn_k.weight
    F16
    [1152, 1152]
  • v.blk.15.attn_output.bias
    F32
    [1152]
  • v.blk.15.attn_output.weight
    F16
    [1152, 1152]
  • v.blk.15.attn_q.bias
    F32
    [1152]
  • v.blk.15.attn_q.weight
    F16
    [1152, 1152]
  • v.blk.15.attn_v.bias
    F32
    [1152]
  • v.blk.15.attn_v.weight
    F16
    [1152, 1152]
  • v.blk.15.layer_norm1.bias
    F32
    [1152]
  • v.blk.15.layer_norm1.weight
    F32
    [1152]
  • v.blk.15.layer_norm2.bias
    F32
    [1152]
  • v.blk.15.layer_norm2.weight
    F32
    [1152]
  • v.blk.15.mlp.fc1.bias
    F32
    [4304]
  • v.blk.15.mlp.fc1.weight
    F16
    [1152, 4304]
  • v.blk.15.mlp.fc2.bias
    F32
    [1152]
  • v.blk.15.mlp.fc2.weight
    F16
    [4304, 1152]
  • v.blk.16
  • v.blk.16.attn_k.bias
    F32
    [1152]
  • v.blk.16.attn_k.weight
    F16
    [1152, 1152]
  • v.blk.16.attn_output.bias
    F32
    [1152]
  • v.blk.16.attn_output.weight
    F16
    [1152, 1152]
  • v.blk.16.attn_q.bias
    F32
    [1152]
  • v.blk.16.attn_q.weight
    F16
    [1152, 1152]
  • v.blk.16.attn_v.bias
    F32
    [1152]
  • v.blk.16.attn_v.weight
    F16
    [1152, 1152]
  • v.blk.16.layer_norm1.bias
    F32
    [1152]
  • v.blk.16.layer_norm1.weight
    F32
    [1152]
  • v.blk.16.layer_norm2.bias
    F32
    [1152]
  • v.blk.16.layer_norm2.weight
    F32
    [1152]
  • v.blk.16.mlp.fc1.bias
    F32
    [4304]
  • v.blk.16.mlp.fc1.weight
    F16
    [1152, 4304]
  • v.blk.16.mlp.fc2.bias
    F32
    [1152]
  • v.blk.16.mlp.fc2.weight
    F16
    [4304, 1152]
  • v.blk.17
  • v.blk.17.attn_k.bias
    F32
    [1152]
  • v.blk.17.attn_k.weight
    F16
    [1152, 1152]
  • v.blk.17.attn_output.bias
    F32
    [1152]
  • v.blk.17.attn_output.weight
    F16
    [1152, 1152]
  • v.blk.17.attn_q.bias
    F32
    [1152]
  • v.blk.17.attn_q.weight
    F16
    [1152, 1152]
  • v.blk.17.attn_v.bias
    F32
    [1152]
  • v.blk.17.attn_v.weight
    F16
    [1152, 1152]
  • v.blk.17.layer_norm1.bias
    F32
    [1152]
  • v.blk.17.layer_norm1.weight
    F32
    [1152]
  • v.blk.17.layer_norm2.bias
    F32
    [1152]
  • v.blk.17.layer_norm2.weight
    F32
    [1152]
  • v.blk.17.mlp.fc1.bias
    F32
    [4304]
  • v.blk.17.mlp.fc1.weight
    F16
    [1152, 4304]
  • v.blk.17.mlp.fc2.bias
    F32
    [1152]
  • v.blk.17.mlp.fc2.weight
    F16
    [4304, 1152]
  • v.blk.18
  • v.blk.18.attn_k.bias
    F32
    [1152]
  • v.blk.18.attn_k.weight
    F16
    [1152, 1152]
  • v.blk.18.attn_output.bias
    F32
    [1152]
  • v.blk.18.attn_output.weight
    F16
    [1152, 1152]
  • v.blk.18.attn_q.bias
    F32
    [1152]
  • v.blk.18.attn_q.weight
    F16
    [1152, 1152]
  • v.blk.18.attn_v.bias
    F32
    [1152]
  • v.blk.18.attn_v.weight
    F16
    [1152, 1152]
  • v.blk.18.layer_norm1.bias
    F32
    [1152]
  • v.blk.18.layer_norm1.weight
    F32
    [1152]
  • v.blk.18.layer_norm2.bias
    F32
    [1152]
  • v.blk.18.layer_norm2.weight
    F32
    [1152]
  • v.blk.18.mlp.fc1.bias
    F32
    [4304]
  • v.blk.18.mlp.fc1.weight
    F16
    [1152, 4304]
  • v.blk.18.mlp.fc2.bias
    F32
    [1152]
  • v.blk.18.mlp.fc2.weight
    F16
    [4304, 1152]
  • v.blk.19
  • v.blk.19.attn_k.bias
    F32
    [1152]
  • v.blk.19.attn_k.weight
    F16
    [1152, 1152]
  • v.blk.19.attn_output.bias
    F32
    [1152]
  • v.blk.19.attn_output.weight
    F16
    [1152, 1152]
  • v.blk.19.attn_q.bias
    F32
    [1152]
  • v.blk.19.attn_q.weight
    F16
    [1152, 1152]
  • v.blk.19.attn_v.bias
    F32
    [1152]
  • v.blk.19.attn_v.weight
    F16
    [1152, 1152]
  • v.blk.19.layer_norm1.bias
    F32
    [1152]
  • v.blk.19.layer_norm1.weight
    F32
    [1152]
  • v.blk.19.layer_norm2.bias
    F32
    [1152]
  • v.blk.19.layer_norm2.weight
    F32
    [1152]
  • v.blk.19.mlp.fc1.bias
    F32
    [4304]
  • v.blk.19.mlp.fc1.weight
    F16
    [1152, 4304]
  • v.blk.19.mlp.fc2.bias
    F32
    [1152]
  • v.blk.19.mlp.fc2.weight
    F16
    [4304, 1152]
  • v.blk.20
  • v.blk.20.attn_k.bias
    F32
    [1152]
  • v.blk.20.attn_k.weight
    F16
    [1152, 1152]
  • v.blk.20.attn_output.bias
    F32
    [1152]
  • v.blk.20.attn_output.weight
    F16
    [1152, 1152]
  • v.blk.20.attn_q.bias
    F32
    [1152]
  • v.blk.20.attn_q.weight
    F16
    [1152, 1152]
  • v.blk.20.attn_v.bias
    F32
    [1152]
  • v.blk.20.attn_v.weight
    F16
    [1152, 1152]
  • v.blk.20.layer_norm1.bias
    F32
    [1152]
  • v.blk.20.layer_norm1.weight
    F32
    [1152]
  • v.blk.20.layer_norm2.bias
    F32
    [1152]
  • v.blk.20.layer_norm2.weight
    F32
    [1152]
  • v.blk.20.mlp.fc1.bias
    F32
    [4304]
  • v.blk.20.mlp.fc1.weight
    F16
    [1152, 4304]
  • v.blk.20.mlp.fc2.bias
    F32
    [1152]
  • v.blk.20.mlp.fc2.weight
    F16
    [4304, 1152]
  • v.blk.21
  • v.blk.21.attn_k.bias
    F32
    [1152]
  • v.blk.21.attn_k.weight
    F16
    [1152, 1152]
  • v.blk.21.attn_output.bias
    F32
    [1152]
  • v.blk.21.attn_output.weight
    F16
    [1152, 1152]
  • v.blk.21.attn_q.bias
    F32
    [1152]
  • v.blk.21.attn_q.weight
    F16
    [1152, 1152]
  • v.blk.21.attn_v.bias
    F32
    [1152]
  • v.blk.21.attn_v.weight
    F16
    [1152, 1152]
  • v.blk.21.layer_norm1.bias
    F32
    [1152]
  • v.blk.21.layer_norm1.weight
    F32
    [1152]
  • v.blk.21.layer_norm2.bias
    F32
    [1152]
  • v.blk.21.layer_norm2.weight
    F32
    [1152]
  • v.blk.21.mlp.fc1.bias
    F32
    [4304]
  • v.blk.21.mlp.fc1.weight
    F16
    [1152, 4304]
  • v.blk.21.mlp.fc2.bias
    F32
    [1152]
  • v.blk.21.mlp.fc2.weight
    F16
    [4304, 1152]
  • v.blk.22
  • v.blk.22.attn_k.bias
    F32
    [1152]
  • v.blk.22.attn_k.weight
    F16
    [1152, 1152]
  • v.blk.22.attn_output.bias
    F32
    [1152]
  • v.blk.22.attn_output.weight
    F16
    [1152, 1152]
  • v.blk.22.attn_q.bias
    F32
    [1152]
  • v.blk.22.attn_q.weight
    F16
    [1152, 1152]
  • v.blk.22.attn_v.bias
    F32
    [1152]
  • v.blk.22.attn_v.weight
    F16
    [1152, 1152]
  • v.blk.22.layer_norm1.bias
    F32
    [1152]
  • v.blk.22.layer_norm1.weight
    F32
    [1152]
  • v.blk.22.layer_norm2.bias
    F32
    [1152]
  • v.blk.22.layer_norm2.weight
    F32
    [1152]
  • v.blk.22.mlp.fc1.bias
    F32
    [4304]
  • v.blk.22.mlp.fc1.weight
    F16
    [1152, 4304]
  • v.blk.22.mlp.fc2.bias
    F32
    [1152]
  • v.blk.22.mlp.fc2.weight
    F16
    [4304, 1152]
  • v.blk.23
  • v.blk.23.attn_k.bias
    F32
    [1152]
  • v.blk.23.attn_k.weight
    F16
    [1152, 1152]
  • v.blk.23.attn_output.bias
    F32
    [1152]
  • v.blk.23.attn_output.weight
    F16
    [1152, 1152]
  • v.blk.23.attn_q.bias
    F32
    [1152]
  • v.blk.23.attn_q.weight
    F16
    [1152, 1152]
  • v.blk.23.attn_v.bias
    F32
    [1152]
  • v.blk.23.attn_v.weight
    F16
    [1152, 1152]
  • v.blk.23.layer_norm1.bias
    F32
    [1152]
  • v.blk.23.layer_norm1.weight
    F32
    [1152]
  • v.blk.23.layer_norm2.bias
    F32
    [1152]
  • v.blk.23.layer_norm2.weight
    F32
    [1152]
  • v.blk.23.mlp.fc1.bias
    F32
    [4304]
  • v.blk.23.mlp.fc1.weight
    F16
    [1152, 4304]
  • v.blk.23.mlp.fc2.bias
    F32
    [1152]
  • v.blk.23.mlp.fc2.weight
    F16
    [4304, 1152]
  • v.blk.24
  • v.blk.24.attn_k.bias
    F32
    [1152]
  • v.blk.24.attn_k.weight
    F16
    [1152, 1152]
  • v.blk.24.attn_output.bias
    F32
    [1152]
  • v.blk.24.attn_output.weight
    F16
    [1152, 1152]
  • v.blk.24.attn_q.bias
    F32
    [1152]
  • v.blk.24.attn_q.weight
    F16
    [1152, 1152]
  • v.blk.24.attn_v.bias
    F32
    [1152]
  • v.blk.24.attn_v.weight
    F16
    [1152, 1152]
  • v.blk.24.layer_norm1.bias
    F32
    [1152]
  • v.blk.24.layer_norm1.weight
    F32
    [1152]
  • v.blk.24.layer_norm2.bias
    F32
    [1152]
  • v.blk.24.layer_norm2.weight
    F32
    [1152]
  • v.blk.24.mlp.fc1.bias
    F32
    [4304]
  • v.blk.24.mlp.fc1.weight
    F16
    [1152, 4304]
  • v.blk.24.mlp.fc2.bias
    F32
    [1152]
  • v.blk.24.mlp.fc2.weight
    F16
    [4304, 1152]
  • v.blk.25
  • v.blk.25.attn_k.bias
    F32
    [1152]
  • v.blk.25.attn_k.weight
    F16
    [1152, 1152]
  • v.blk.25.attn_output.bias
    F32
    [1152]
  • v.blk.25.attn_output.weight
    F16
    [1152, 1152]
  • v.blk.25.attn_q.bias
    F32
    [1152]
  • v.blk.25.attn_q.weight
    F16
    [1152, 1152]
  • v.blk.25.attn_v.bias
    F32
    [1152]
  • v.blk.25.attn_v.weight
    F16
    [1152, 1152]
  • v.blk.25.layer_norm1.bias
    F32
    [1152]
  • v.blk.25.layer_norm1.weight
    F32
    [1152]
  • v.blk.25.layer_norm2.bias
    F32
    [1152]
  • v.blk.25.layer_norm2.weight
    F32
    [1152]
  • v.blk.25.mlp.fc1.bias
    F32
    [4304]
  • v.blk.25.mlp.fc1.weight
    F16
    [1152, 4304]
  • v.blk.25.mlp.fc2.bias
    F32
    [1152]
  • v.blk.25.mlp.fc2.weight
    F16
    [4304, 1152]
  • v.blk.26
  • v.blk.26.attn_k.bias
    F32
    [1152]
  • v.blk.26.attn_k.weight
    F16
    [1152, 1152]
  • v.blk.26.attn_output.bias
    F32
    [1152]
  • v.blk.26.attn_output.weight
    F16
    [1152, 1152]
  • v.blk.26.attn_q.bias
    F32
    [1152]
  • v.blk.26.attn_q.weight
    F16
    [1152, 1152]
  • v.blk.26.attn_v.bias
    F32
    [1152]
  • v.blk.26.attn_v.weight
    F16
    [1152, 1152]
  • v.blk.26.layer_norm1.bias
    F32
    [1152]
  • v.blk.26.layer_norm1.weight
    F32
    [1152]
  • v.blk.26.layer_norm2.bias
    F32
    [1152]
  • v.blk.26.layer_norm2.weight
    F32
    [1152]
  • v.blk.26.mlp.fc1.bias
    F32
    [4304]
  • v.blk.26.mlp.fc1.weight
    F16
    [1152, 4304]
  • v.blk.26.mlp.fc2.bias
    F32
    [1152]
  • v.blk.26.mlp.fc2.weight
    F16
    [4304, 1152]
  • v.patch_embedding.bias
    F32
    [1152]
  • v.patch_embedding.weight
    F16
    [14, 14, 3, 1152]
  • v.position_embedding.weight
    F16
    [1152, 4096]
  • v.post_layernorm.bias
    F32
    [1152]
  • v.post_layernorm.weight
    F32
    [1152]
  • output_norm.weight
    F32
    [2560]