105 1 month ago

vision
d14a3822e7b2 · 1.8GB
    Metadata
  • general.architecture
    glm4
  • general.file_type
    BF16
  • glm4.attention.head_count
    16
  • glm4.attention.head_count_kv
    8
  • glm4.attention.key_length
    128
  • glm4.attention.layer_norm_rms_epsilon
    1e-05
  • glm4.attention.value_length
    128
  • glm4.block_count
    17
  • glm4.context_length
    131072
  • glm4.embedding_length
    1536
  • glm4.feed_forward_length
    4608
  • glm4.nextn_predict_layers
    1
  • glm4.rope.dimension_count
    128
  • glm4.rope.dimension_sections
    [16, 24, 24, 0]
  • glm4.rope.freq_base
    10000
  • tokenizer.ggml.add_eos_token
    false
  • tokenizer.ggml.add_padding_token
    false
  • tokenizer.ggml.eos_token_id
    59246
  • tokenizer.ggml.eos_token_ids
    [59246, 59253]
  • tokenizer.ggml.merges
    [Ġ Ġ, Ġ ĠĠĠ, ĠĠ ĠĠ, ĠĠĠ Ġ, Ġ t, ...]
  • tokenizer.ggml.model
    gpt2
  • tokenizer.ggml.padding_token_id
    59246
  • tokenizer.ggml.pre
    llama-bpe
  • tokenizer.ggml.scores
    [0, 1, 2, 3, 4, ...]
  • tokenizer.ggml.token_type
    [1, 1, 1, 1, 1, ...]
  • tokenizer.ggml.tokens
    [Ā, ā, Ă, ă, Ą, ...]
  • Tensor
  • token_embd.weight
    BF16
    [1536, 59392]
  • blk.0
  • blk.0.attn_k.weight
    BF16
    [1536, 1024]
  • blk.0.attn_norm.weight
    F32
    [1536]
  • blk.0.attn_output.weight
    BF16
    [2048, 1536]
  • blk.0.attn_q.weight
    BF16
    [1536, 2048]
  • blk.0.attn_v.weight
    BF16
    [1536, 1024]
  • blk.0.ffn_down.weight
    BF16
    [4608, 1536]
  • blk.0.ffn_norm.weight
    F32
    [1536]
  • blk.0.ffn_up.weight
    BF16
    [1536, 9216]
  • blk.0.post_attention_norm.weight
    F32
    [1536]
  • blk.0.post_ffw_norm.weight
    F32
    [1536]
  • blk.1
  • blk.1.attn_k.weight
    BF16
    [1536, 1024]
  • blk.1.attn_norm.weight
    F32
    [1536]
  • blk.1.attn_output.weight
    BF16
    [2048, 1536]
  • blk.1.attn_q.weight
    BF16
    [1536, 2048]
  • blk.1.attn_v.weight
    BF16
    [1536, 1024]
  • blk.1.ffn_down.weight
    BF16
    [4608, 1536]
  • blk.1.ffn_norm.weight
    F32
    [1536]
  • blk.1.ffn_up.weight
    BF16
    [1536, 9216]
  • blk.1.post_attention_norm.weight
    F32
    [1536]
  • blk.1.post_ffw_norm.weight
    F32
    [1536]
  • blk.2
  • blk.2.attn_k.weight
    BF16
    [1536, 1024]
  • blk.2.attn_norm.weight
    F32
    [1536]
  • blk.2.attn_output.weight
    BF16
    [2048, 1536]
  • blk.2.attn_q.weight
    BF16
    [1536, 2048]
  • blk.2.attn_v.weight
    BF16
    [1536, 1024]
  • blk.2.ffn_down.weight
    BF16
    [4608, 1536]
  • blk.2.ffn_norm.weight
    F32
    [1536]
  • blk.2.ffn_up.weight
    BF16
    [1536, 9216]
  • blk.2.post_attention_norm.weight
    F32
    [1536]
  • blk.2.post_ffw_norm.weight
    F32
    [1536]
  • blk.3
  • blk.3.attn_k.weight
    BF16
    [1536, 1024]
  • blk.3.attn_norm.weight
    F32
    [1536]
  • blk.3.attn_output.weight
    BF16
    [2048, 1536]
  • blk.3.attn_q.weight
    BF16
    [1536, 2048]
  • blk.3.attn_v.weight
    BF16
    [1536, 1024]
  • blk.3.ffn_down.weight
    BF16
    [4608, 1536]
  • blk.3.ffn_norm.weight
    F32
    [1536]
  • blk.3.ffn_up.weight
    BF16
    [1536, 9216]
  • blk.3.post_attention_norm.weight
    F32
    [1536]
  • blk.3.post_ffw_norm.weight
    F32
    [1536]
  • blk.4
  • blk.4.attn_k.weight
    BF16
    [1536, 1024]
  • blk.4.attn_norm.weight
    F32
    [1536]
  • blk.4.attn_output.weight
    BF16
    [2048, 1536]
  • blk.4.attn_q.weight
    BF16
    [1536, 2048]
  • blk.4.attn_v.weight
    BF16
    [1536, 1024]
  • blk.4.ffn_down.weight
    BF16
    [4608, 1536]
  • blk.4.ffn_norm.weight
    F32
    [1536]
  • blk.4.ffn_up.weight
    BF16
    [1536, 9216]
  • blk.4.post_attention_norm.weight
    F32
    [1536]
  • blk.4.post_ffw_norm.weight
    F32
    [1536]
  • blk.5
  • blk.5.attn_k.weight
    BF16
    [1536, 1024]
  • blk.5.attn_norm.weight
    F32
    [1536]
  • blk.5.attn_output.weight
    BF16
    [2048, 1536]
  • blk.5.attn_q.weight
    BF16
    [1536, 2048]
  • blk.5.attn_v.weight
    BF16
    [1536, 1024]
  • blk.5.ffn_down.weight
    BF16
    [4608, 1536]
  • blk.5.ffn_norm.weight
    F32
    [1536]
  • blk.5.ffn_up.weight
    BF16
    [1536, 9216]
  • blk.5.post_attention_norm.weight
    F32
    [1536]
  • blk.5.post_ffw_norm.weight
    F32
    [1536]
  • blk.6
  • blk.6.attn_k.weight
    BF16
    [1536, 1024]
  • blk.6.attn_norm.weight
    F32
    [1536]
  • blk.6.attn_output.weight
    BF16
    [2048, 1536]
  • blk.6.attn_q.weight
    BF16
    [1536, 2048]
  • blk.6.attn_v.weight
    BF16
    [1536, 1024]
  • blk.6.ffn_down.weight
    BF16
    [4608, 1536]
  • blk.6.ffn_norm.weight
    F32
    [1536]
  • blk.6.ffn_up.weight
    BF16
    [1536, 9216]
  • blk.6.post_attention_norm.weight
    F32
    [1536]
  • blk.6.post_ffw_norm.weight
    F32
    [1536]
  • blk.7
  • blk.7.attn_k.weight
    BF16
    [1536, 1024]
  • blk.7.attn_norm.weight
    F32
    [1536]
  • blk.7.attn_output.weight
    BF16
    [2048, 1536]
  • blk.7.attn_q.weight
    BF16
    [1536, 2048]
  • blk.7.attn_v.weight
    BF16
    [1536, 1024]
  • blk.7.ffn_down.weight
    BF16
    [4608, 1536]
  • blk.7.ffn_norm.weight
    F32
    [1536]
  • blk.7.ffn_up.weight
    BF16
    [1536, 9216]
  • blk.7.post_attention_norm.weight
    F32
    [1536]
  • blk.7.post_ffw_norm.weight
    F32
    [1536]
  • blk.8
  • blk.8.attn_k.weight
    BF16
    [1536, 1024]
  • blk.8.attn_norm.weight
    F32
    [1536]
  • blk.8.attn_output.weight
    BF16
    [2048, 1536]
  • blk.8.attn_q.weight
    BF16
    [1536, 2048]
  • blk.8.attn_v.weight
    BF16
    [1536, 1024]
  • blk.8.ffn_down.weight
    BF16
    [4608, 1536]
  • blk.8.ffn_norm.weight
    F32
    [1536]
  • blk.8.ffn_up.weight
    BF16
    [1536, 9216]
  • blk.8.post_attention_norm.weight
    F32
    [1536]
  • blk.8.post_ffw_norm.weight
    F32
    [1536]
  • blk.9
  • blk.9.attn_k.weight
    BF16
    [1536, 1024]
  • blk.9.attn_norm.weight
    F32
    [1536]
  • blk.9.attn_output.weight
    BF16
    [2048, 1536]
  • blk.9.attn_q.weight
    BF16
    [1536, 2048]
  • blk.9.attn_v.weight
    BF16
    [1536, 1024]
  • blk.9.ffn_down.weight
    BF16
    [4608, 1536]
  • blk.9.ffn_norm.weight
    F32
    [1536]
  • blk.9.ffn_up.weight
    BF16
    [1536, 9216]
  • blk.9.post_attention_norm.weight
    F32
    [1536]
  • blk.9.post_ffw_norm.weight
    F32
    [1536]
  • blk.10
  • blk.10.attn_k.weight
    BF16
    [1536, 1024]
  • blk.10.attn_norm.weight
    F32
    [1536]
  • blk.10.attn_output.weight
    BF16
    [2048, 1536]
  • blk.10.attn_q.weight
    BF16
    [1536, 2048]
  • blk.10.attn_v.weight
    BF16
    [1536, 1024]
  • blk.10.ffn_down.weight
    BF16
    [4608, 1536]
  • blk.10.ffn_norm.weight
    F32
    [1536]
  • blk.10.ffn_up.weight
    BF16
    [1536, 9216]
  • blk.10.post_attention_norm.weight
    F32
    [1536]
  • blk.10.post_ffw_norm.weight
    F32
    [1536]
  • blk.11
  • blk.11.attn_k.weight
    BF16
    [1536, 1024]
  • blk.11.attn_norm.weight
    F32
    [1536]
  • blk.11.attn_output.weight
    BF16
    [2048, 1536]
  • blk.11.attn_q.weight
    BF16
    [1536, 2048]
  • blk.11.attn_v.weight
    BF16
    [1536, 1024]
  • blk.11.ffn_down.weight
    BF16
    [4608, 1536]
  • blk.11.ffn_norm.weight
    F32
    [1536]
  • blk.11.ffn_up.weight
    BF16
    [1536, 9216]
  • blk.11.post_attention_norm.weight
    F32
    [1536]
  • blk.11.post_ffw_norm.weight
    F32
    [1536]
  • blk.12
  • blk.12.attn_k.weight
    BF16
    [1536, 1024]
  • blk.12.attn_norm.weight
    F32
    [1536]
  • blk.12.attn_output.weight
    BF16
    [2048, 1536]
  • blk.12.attn_q.weight
    BF16
    [1536, 2048]
  • blk.12.attn_v.weight
    BF16
    [1536, 1024]
  • blk.12.ffn_down.weight
    BF16
    [4608, 1536]
  • blk.12.ffn_norm.weight
    F32
    [1536]
  • blk.12.ffn_up.weight
    BF16
    [1536, 9216]
  • blk.12.post_attention_norm.weight
    F32
    [1536]
  • blk.12.post_ffw_norm.weight
    F32
    [1536]
  • blk.13
  • blk.13.attn_k.weight
    BF16
    [1536, 1024]
  • blk.13.attn_norm.weight
    F32
    [1536]
  • blk.13.attn_output.weight
    BF16
    [2048, 1536]
  • blk.13.attn_q.weight
    BF16
    [1536, 2048]
  • blk.13.attn_v.weight
    BF16
    [1536, 1024]
  • blk.13.ffn_down.weight
    BF16
    [4608, 1536]
  • blk.13.ffn_norm.weight
    F32
    [1536]
  • blk.13.ffn_up.weight
    BF16
    [1536, 9216]
  • blk.13.post_attention_norm.weight
    F32
    [1536]
  • blk.13.post_ffw_norm.weight
    F32
    [1536]
  • blk.14
  • blk.14.attn_k.weight
    BF16
    [1536, 1024]
  • blk.14.attn_norm.weight
    F32
    [1536]
  • blk.14.attn_output.weight
    BF16
    [2048, 1536]
  • blk.14.attn_q.weight
    BF16
    [1536, 2048]
  • blk.14.attn_v.weight
    BF16
    [1536, 1024]
  • blk.14.ffn_down.weight
    BF16
    [4608, 1536]
  • blk.14.ffn_norm.weight
    F32
    [1536]
  • blk.14.ffn_up.weight
    BF16
    [1536, 9216]
  • blk.14.post_attention_norm.weight
    F32
    [1536]
  • blk.14.post_ffw_norm.weight
    F32
    [1536]
  • blk.15
  • blk.15.attn_k.weight
    BF16
    [1536, 1024]
  • blk.15.attn_norm.weight
    F32
    [1536]
  • blk.15.attn_output.weight
    BF16
    [2048, 1536]
  • blk.15.attn_q.weight
    BF16
    [1536, 2048]
  • blk.15.attn_v.weight
    BF16
    [1536, 1024]
  • blk.15.ffn_down.weight
    BF16
    [4608, 1536]
  • blk.15.ffn_norm.weight
    F32
    [1536]
  • blk.15.ffn_up.weight
    BF16
    [1536, 9216]
  • blk.15.post_attention_norm.weight
    F32
    [1536]
  • blk.15.post_ffw_norm.weight
    F32
    [1536]
  • blk.16
  • blk.16.attn_k.weight
    BF16
    [1536, 1024]
  • blk.16.attn_norm.weight
    F32
    [1536]
  • blk.16.attn_output.weight
    BF16
    [2048, 1536]
  • blk.16.attn_q.weight
    BF16
    [1536, 2048]
  • blk.16.attn_v.weight
    BF16
    [1536, 1024]
  • blk.16.ffn_down.weight
    BF16
    [4608, 1536]
  • blk.16.ffn_norm.weight
    F32
    [1536]
  • blk.16.ffn_up.weight
    BF16
    [1536, 9216]
  • blk.16.nextn.eh_proj.weight
    BF16
    [3072, 1536]
  • blk.16.nextn.embed_tokens.weight
    BF16
    [1536, 59392]
  • blk.16.nextn.enorm.weight
    F32
    [1536]
  • blk.16.nextn.hnorm.weight
    F32
    [1536]
  • blk.16.nextn.shared_head_head.weight
    BF16
    [1536, 59392]
  • blk.16.nextn.shared_head_norm.weight
    F32
    [1536]
  • blk.16.post_attention_norm.weight
    F32
    [1536]
  • blk.16.post_ffw_norm.weight
    F32
    [1536]
  • output.weight
    BF16
    [1536, 59392]
  • output_norm.weight
    F32
    [1536]