laya:322m-multilingual-bf16

8,361 35 minutes ago

Laya is a 421M decision model from Convai Innovations, fine-tuned from ModernBERT-large.

embedding decision
4d8ab9c0f270 · 658MB
    Metadata
  • general.architecture
    modern-bert
  • general.file_type
    BF16
  • modern-bert.attention.causal
    false
  • modern-bert.attention.head_count
    12
  • modern-bert.attention.layer_norm_epsilon
    1e-05
  • modern-bert.attention.layer_norm_rms_epsilon
    1e-05
  • modern-bert.attention.sliding_window
    128
  • modern-bert.attention.sliding_window_pattern
    3
  • modern-bert.block_count
    24
  • modern-bert.classifier.activation
    gelu
  • modern-bert.classifier.pooling_type
    Mean
  • modern-bert.context_length
    8192
  • modern-bert.decision.block_count
    2
  • modern-bert.decision.max_head_tokens
    256
  • modern-bert.decision.temperature.choice
    1
  • modern-bert.decision.temperature.noul
    1
  • modern-bert.decision.temperature.score
    1
  • modern-bert.decision.type
    laya
  • modern-bert.embedding_length
    768
  • modern-bert.feed_forward_length
    [1152, 1152, 1152, 1152, 1152, ...]
  • modern-bert.hidden_activation
    gelu
  • modern-bert.rope.freq_base
    160000
  • modern-bert.rope.freq_base_swa
    160000
  • modern-bert.rope.scaling.type
    none
  • modern-bert.vocab_size
    256000
  • tokenizer.chat_template.systemone
    <bos>{{ type }} question: {{ instructions if instructions is string else instructions | tojson }}<eos>{% for o in options %}<mask> {% if type == 'choice' %}{{ o.key }}{% if o.description %}: {{ o.description if o.description is string else o.description | tojson }}{% endif %}{% elif type == 'score' %}level {{ o.key }}: {{ o.description if o.description is string else o.description | tojson }}{% else %}{{ o.key }}: {% if o.description %}{{ o.description if o.description is string else o.description | tojson }}{% elif o.key == 'true' %}yes, the statement holds{% else %}no, the statement does not hold{% endif %}{% endif %}{% endfor %}<eos>{{ state if state is string else state | tojson }}<eos>
  • tokenizer.chat_templates
    [systemone]
  • tokenizer.ggml.add_bos_token
    true
  • tokenizer.ggml.add_eos_token
    true
  • tokenizer.ggml.add_sep_token
    false
  • tokenizer.ggml.bos_token_id
    2
  • tokenizer.ggml.eos_token_id
    1
  • tokenizer.ggml.mask_token_id
    4
  • tokenizer.ggml.merges
    [ , ▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁ ▁, , , ▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁ ▁▁, ...]
  • tokenizer.ggml.model
    gpt2
  • tokenizer.ggml.padding_token_id
    0
  • tokenizer.ggml.pre
    mmbert
  • tokenizer.ggml.seperator_token_id
    1
  • tokenizer.ggml.token_type
    [3, 3, 3, 3, 3, ...]
  • tokenizer.ggml.token_type_count
    3
  • tokenizer.ggml.tokens
    [<pad>, <eos>, <bos>, <unk>, <mask>, ...]
  • tokenizer.ggml.unknown_token_id
    3
  • Tensor
  • token_embd.weight
    BF16
    [768, 256000]
  • blk.0
  • blk.0.attn_output.weight
    BF16
    [768, 768]
  • blk.0.attn_qkv.weight
    BF16
    [768, 2304]
  • blk.0.ffn_down.weight
    BF16
    [1152, 768]
  • blk.0.ffn_norm.weight
    F32
    [768]
  • blk.0.ffn_up.weight
    BF16
    [768, 2304]
  • blk.1
  • blk.1.attn_norm.weight
    F32
    [768]
  • blk.1.attn_output.weight
    BF16
    [768, 768]
  • blk.1.attn_qkv.weight
    BF16
    [768, 2304]
  • blk.1.ffn_down.weight
    BF16
    [1152, 768]
  • blk.1.ffn_norm.weight
    F32
    [768]
  • blk.1.ffn_up.weight
    BF16
    [768, 2304]
  • blk.2
  • blk.2.attn_norm.weight
    F32
    [768]
  • blk.2.attn_output.weight
    BF16
    [768, 768]
  • blk.2.attn_qkv.weight
    BF16
    [768, 2304]
  • blk.2.ffn_down.weight
    BF16
    [1152, 768]
  • blk.2.ffn_norm.weight
    F32
    [768]
  • blk.2.ffn_up.weight
    BF16
    [768, 2304]
  • blk.3
  • blk.3.attn_norm.weight
    F32
    [768]
  • blk.3.attn_output.weight
    BF16
    [768, 768]
  • blk.3.attn_qkv.weight
    BF16
    [768, 2304]
  • blk.3.ffn_down.weight
    BF16
    [1152, 768]
  • blk.3.ffn_norm.weight
    F32
    [768]
  • blk.3.ffn_up.weight
    BF16
    [768, 2304]
  • blk.4
  • blk.4.attn_norm.weight
    F32
    [768]
  • blk.4.attn_output.weight
    BF16
    [768, 768]
  • blk.4.attn_qkv.weight
    BF16
    [768, 2304]
  • blk.4.ffn_down.weight
    BF16
    [1152, 768]
  • blk.4.ffn_norm.weight
    F32
    [768]
  • blk.4.ffn_up.weight
    BF16
    [768, 2304]
  • blk.5
  • blk.5.attn_norm.weight
    F32
    [768]
  • blk.5.attn_output.weight
    BF16
    [768, 768]
  • blk.5.attn_qkv.weight
    BF16
    [768, 2304]
  • blk.5.ffn_down.weight
    BF16
    [1152, 768]
  • blk.5.ffn_norm.weight
    F32
    [768]
  • blk.5.ffn_up.weight
    BF16
    [768, 2304]
  • blk.6
  • blk.6.attn_norm.weight
    F32
    [768]
  • blk.6.attn_output.weight
    BF16
    [768, 768]
  • blk.6.attn_qkv.weight
    BF16
    [768, 2304]
  • blk.6.ffn_down.weight
    BF16
    [1152, 768]
  • blk.6.ffn_norm.weight
    F32
    [768]
  • blk.6.ffn_up.weight
    BF16
    [768, 2304]
  • blk.7
  • blk.7.attn_norm.weight
    F32
    [768]
  • blk.7.attn_output.weight
    BF16
    [768, 768]
  • blk.7.attn_qkv.weight
    BF16
    [768, 2304]
  • blk.7.ffn_down.weight
    BF16
    [1152, 768]
  • blk.7.ffn_norm.weight
    F32
    [768]
  • blk.7.ffn_up.weight
    BF16
    [768, 2304]
  • blk.8
  • blk.8.attn_norm.weight
    F32
    [768]
  • blk.8.attn_output.weight
    BF16
    [768, 768]
  • blk.8.attn_qkv.weight
    BF16
    [768, 2304]
  • blk.8.ffn_down.weight
    BF16
    [1152, 768]
  • blk.8.ffn_norm.weight
    F32
    [768]
  • blk.8.ffn_up.weight
    BF16
    [768, 2304]
  • blk.9
  • blk.9.attn_norm.weight
    F32
    [768]
  • blk.9.attn_output.weight
    BF16
    [768, 768]
  • blk.9.attn_qkv.weight
    BF16
    [768, 2304]
  • blk.9.ffn_down.weight
    BF16
    [1152, 768]
  • blk.9.ffn_norm.weight
    F32
    [768]
  • blk.9.ffn_up.weight
    BF16
    [768, 2304]
  • blk.10
  • blk.10.attn_norm.weight
    F32
    [768]
  • blk.10.attn_output.weight
    BF16
    [768, 768]
  • blk.10.attn_qkv.weight
    BF16
    [768, 2304]
  • blk.10.ffn_down.weight
    BF16
    [1152, 768]
  • blk.10.ffn_norm.weight
    F32
    [768]
  • blk.10.ffn_up.weight
    BF16
    [768, 2304]
  • blk.11
  • blk.11.attn_norm.weight
    F32
    [768]
  • blk.11.attn_output.weight
    BF16
    [768, 768]
  • blk.11.attn_qkv.weight
    BF16
    [768, 2304]
  • blk.11.ffn_down.weight
    BF16
    [1152, 768]
  • blk.11.ffn_norm.weight
    F32
    [768]
  • blk.11.ffn_up.weight
    BF16
    [768, 2304]
  • blk.12
  • blk.12.attn_norm.weight
    F32
    [768]
  • blk.12.attn_output.weight
    BF16
    [768, 768]
  • blk.12.attn_qkv.weight
    BF16
    [768, 2304]
  • blk.12.ffn_down.weight
    BF16
    [1152, 768]
  • blk.12.ffn_norm.weight
    F32
    [768]
  • blk.12.ffn_up.weight
    BF16
    [768, 2304]
  • blk.13
  • blk.13.attn_norm.weight
    F32
    [768]
  • blk.13.attn_output.weight
    BF16
    [768, 768]
  • blk.13.attn_qkv.weight
    BF16
    [768, 2304]
  • blk.13.ffn_down.weight
    BF16
    [1152, 768]
  • blk.13.ffn_norm.weight
    F32
    [768]
  • blk.13.ffn_up.weight
    BF16
    [768, 2304]
  • blk.14
  • blk.14.attn_norm.weight
    F32
    [768]
  • blk.14.attn_output.weight
    BF16
    [768, 768]
  • blk.14.attn_qkv.weight
    BF16
    [768, 2304]
  • blk.14.ffn_down.weight
    BF16
    [1152, 768]
  • blk.14.ffn_norm.weight
    F32
    [768]
  • blk.14.ffn_up.weight
    BF16
    [768, 2304]
  • blk.15
  • blk.15.attn_norm.weight
    F32
    [768]
  • blk.15.attn_output.weight
    BF16
    [768, 768]
  • blk.15.attn_qkv.weight
    BF16
    [768, 2304]
  • blk.15.ffn_down.weight
    BF16
    [1152, 768]
  • blk.15.ffn_norm.weight
    F32
    [768]
  • blk.15.ffn_up.weight
    BF16
    [768, 2304]
  • blk.16
  • blk.16.attn_norm.weight
    F32
    [768]
  • blk.16.attn_output.weight
    BF16
    [768, 768]
  • blk.16.attn_qkv.weight
    BF16
    [768, 2304]
  • blk.16.ffn_down.weight
    BF16
    [1152, 768]
  • blk.16.ffn_norm.weight
    F32
    [768]
  • blk.16.ffn_up.weight
    BF16
    [768, 2304]
  • blk.17
  • blk.17.attn_norm.weight
    F32
    [768]
  • blk.17.attn_output.weight
    BF16
    [768, 768]
  • blk.17.attn_qkv.weight
    BF16
    [768, 2304]
  • blk.17.ffn_down.weight
    BF16
    [1152, 768]
  • blk.17.ffn_norm.weight
    F32
    [768]
  • blk.17.ffn_up.weight
    BF16
    [768, 2304]
  • blk.18
  • blk.18.attn_norm.weight
    F32
    [768]
  • blk.18.attn_output.weight
    BF16
    [768, 768]
  • blk.18.attn_qkv.weight
    BF16
    [768, 2304]
  • blk.18.ffn_down.weight
    BF16
    [1152, 768]
  • blk.18.ffn_norm.weight
    F32
    [768]
  • blk.18.ffn_up.weight
    BF16
    [768, 2304]
  • blk.19
  • blk.19.attn_norm.weight
    F32
    [768]
  • blk.19.attn_output.weight
    BF16
    [768, 768]
  • blk.19.attn_qkv.weight
    BF16
    [768, 2304]
  • blk.19.ffn_down.weight
    BF16
    [1152, 768]
  • blk.19.ffn_norm.weight
    F32
    [768]
  • blk.19.ffn_up.weight
    BF16
    [768, 2304]
  • blk.20
  • blk.20.attn_norm.weight
    F32
    [768]
  • blk.20.attn_output.weight
    BF16
    [768, 768]
  • blk.20.attn_qkv.weight
    BF16
    [768, 2304]
  • blk.20.ffn_down.weight
    BF16
    [1152, 768]
  • blk.20.ffn_norm.weight
    F32
    [768]
  • blk.20.ffn_up.weight
    BF16
    [768, 2304]
  • blk.21
  • blk.21.attn_norm.weight
    F32
    [768]
  • blk.21.attn_output.weight
    BF16
    [768, 768]
  • blk.21.attn_qkv.weight
    BF16
    [768, 2304]
  • blk.21.ffn_down.weight
    BF16
    [1152, 768]
  • blk.21.ffn_norm.weight
    F32
    [768]
  • blk.21.ffn_up.weight
    BF16
    [768, 2304]
  • blk.22
  • blk.22.attn_norm.bias
    F32
    [768]
  • blk.22.attn_norm.weight
    F32
    [768]
  • blk.22.attn_output.bias
    F32
    [768]
  • blk.22.attn_output.weight
    BF16
    [768, 768]
  • blk.22.attn_qkv.bias
    F32
    [2304]
  • blk.22.attn_qkv.weight
    BF16
    [768, 2304]
  • blk.22.ffn_down.bias
    F32
    [768]
  • blk.22.ffn_down.weight
    BF16
    [3072, 768]
  • blk.22.ffn_norm.bias
    F32
    [768]
  • blk.22.ffn_norm.weight
    F32
    [768]
  • blk.22.ffn_up.bias
    F32
    [3072]
  • blk.22.ffn_up.weight
    BF16
    [768, 3072]
  • blk.23
  • blk.23.attn_norm.bias
    F32
    [768]
  • blk.23.attn_norm.weight
    F32
    [768]
  • blk.23.attn_output.bias
    F32
    [768]
  • blk.23.attn_output.weight
    BF16
    [768, 768]
  • blk.23.attn_qkv.bias
    F32
    [2304]
  • blk.23.attn_qkv.weight
    BF16
    [768, 2304]
  • blk.23.ffn_down.bias
    F32
    [768]
  • blk.23.ffn_down.weight
    BF16
    [3072, 768]
  • blk.23.ffn_norm.bias
    F32
    [768]
  • blk.23.ffn_norm.weight
    F32
    [768]
  • blk.23.ffn_up.bias
    F32
    [3072]
  • blk.23.ffn_up.weight
    BF16
    [768, 3072]
  • cls.bias
    F32
    [768]
  • cls.norm.bias
    F32
    [768]
  • cls.norm.weight
    F32
    [768]
  • cls.output.bias
    F32
    [1]
  • cls.output.weight
    BF16
    [768, 1]
  • cls.weight
    BF16
    [768, 768]
  • token_embd_norm.weight
    F32
    [768]
  • token_types.weight
    F32
    [768, 3]
  • output_norm.weight
    F32
    [768]