hotchpotch/ bekko-embedding-v1-a8m:bf16

5 2 hours ago

an ultra-compact multilingual text embedding model

embedding
... /
model
f72934935c21 · 218MB
    Metadata
  • general.architecture
    modern-bert
  • general.file_type
    BF16
  • modern-bert.attention.causal
    false
  • modern-bert.attention.head_count
    6
  • 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
    4
  • modern-bert.context_length
    8192
  • modern-bert.embedding_length
    384
  • modern-bert.feed_forward_length
    1152
  • modern-bert.hidden_activation
    gelu
  • modern-bert.pooling_type
    Mean
  • 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.ggml.add_bos_token
    true
  • tokenizer.ggml.add_eos_token
    true
  • tokenizer.ggml.add_sep_token
    false
  • tokenizer.ggml.add_space_prefix
    true
  • tokenizer.ggml.bos_token_id
    2
  • tokenizer.ggml.eos_token_id
    1
  • tokenizer.ggml.mask_token_id
    4
  • tokenizer.ggml.model
    llama
  • tokenizer.ggml.padding_token_id
    0
  • tokenizer.ggml.pre
    default
  • tokenizer.ggml.scores
    [-1000, -1000, -1000, -1000, -1000, ...]
  • tokenizer.ggml.seperator_token_id
    1
  • tokenizer.ggml.token_type
    [3, 3, 3, 3, 3, ...]
  • tokenizer.ggml.tokens
    [<pad>, <eos>, <bos>, <unk>, <mask>, ...]
  • tokenizer.ggml.unknown_token_id
    3
    • Tensor
  • token_embd.weight
    BF16
    [384, 256000]
    • blk.0
  • blk.0.attn_output.weight
    BF16
    [384, 384]
  • blk.0.attn_qkv.weight
    BF16
    [384, 1152]
  • blk.0.ffn_down.weight
    BF16
    [1152, 384]
  • blk.0.ffn_norm.weight
    F32
    [384]
  • blk.0.ffn_up.weight
    BF16
    [384, 2304]
    • blk.1
  • blk.1.attn_norm.weight
    F32
    [384]
  • blk.1.attn_output.weight
    BF16
    [384, 384]
  • blk.1.attn_qkv.weight
    BF16
    [384, 1152]
  • blk.1.ffn_down.weight
    BF16
    [1152, 384]
  • blk.1.ffn_norm.weight
    F32
    [384]
  • blk.1.ffn_up.weight
    BF16
    [384, 2304]
    • blk.2
  • blk.2.attn_norm.weight
    F32
    [384]
  • blk.2.attn_output.weight
    BF16
    [384, 384]
  • blk.2.attn_qkv.weight
    BF16
    [384, 1152]
  • blk.2.ffn_down.weight
    BF16
    [1152, 384]
  • blk.2.ffn_norm.weight
    F32
    [384]
  • blk.2.ffn_up.weight
    BF16
    [384, 2304]
    • blk.3
  • blk.3.attn_norm.weight
    F32
    [384]
  • blk.3.attn_output.weight
    BF16
    [384, 384]
  • blk.3.attn_qkv.weight
    BF16
    [384, 1152]
  • blk.3.ffn_down.weight
    BF16
    [1152, 384]
  • blk.3.ffn_norm.weight
    F32
    [384]
  • blk.3.ffn_up.weight
    BF16
    [384, 2304]
  • token_embd_norm.weight
    F32
    [384]
  • output_norm.weight
    F32
    [384]