jina/
jina-embeddings-v2-small-en:latest

2,603 1 year ago

Text embedding model (small) for input of size up to 8192 tokens

embedding
... /
model
465b654b2a5a · 66MB
    Metadata
  • general.architecture
    jina-bert-v2
  • general.file_type
    F16
  • jina-bert-v2.attention.causal
    false
  • jina-bert-v2.attention.head_count
    8
  • jina-bert-v2.attention.layer_norm_epsilon
    1e-12
  • jina-bert-v2.block_count
    4
  • jina-bert-v2.context_length
    8192
  • jina-bert-v2.embedding_length
    512
  • jina-bert-v2.feed_forward_length
    2048
  • jina-bert-v2.pooling_type
    Mean
  • tokenizer.ggml.add_bos_token
    true
  • tokenizer.ggml.add_eos_token
    true
  • tokenizer.ggml.cls_token_id
    101
  • tokenizer.ggml.mask_token_id
    103
  • tokenizer.ggml.model
    bert
  • tokenizer.ggml.padding_token_id
    0
  • tokenizer.ggml.pre
    jina-v2-en
  • tokenizer.ggml.seperator_token_id
    102
  • tokenizer.ggml.token_type
    [3, 1, 1, 1, 1, ...]
  • tokenizer.ggml.token_type_count
    2
  • tokenizer.ggml.tokens
    [[PAD], [unused0], [unused1], [unused2], [unused3], ...]
  • tokenizer.ggml.unknown_token_id
    100
    • Tensor
  • token_embd.weight
    F16
    [512, 30528]
    • blk.0
  • blk.0.attn_k.bias
    F32
    [512]
  • blk.0.attn_k.weight
    F16
    [512, 512]
  • blk.0.attn_output.bias
    F32
    [512]
  • blk.0.attn_output.weight
    F16
    [512, 512]
  • blk.0.attn_output_norm.bias
    F32
    [512]
  • blk.0.attn_output_norm.weight
    F32
    [512]
  • blk.0.attn_q.bias
    F32
    [512]
  • blk.0.attn_q.weight
    F16
    [512, 512]
  • blk.0.attn_v.bias
    F32
    [512]
  • blk.0.attn_v.weight
    F16
    [512, 512]
  • blk.0.ffn_down.bias
    F32
    [512]
  • blk.0.ffn_down.weight
    F16
    [2048, 512]
  • blk.0.ffn_gate.weight
    F16
    [512, 2048]
  • blk.0.ffn_up.weight
    F16
    [512, 2048]
  • blk.0.layer_output_norm.bias
    F32
    [512]
  • blk.0.layer_output_norm.weight
    F32
    [512]
    • blk.1
  • blk.1.attn_k.bias
    F32
    [512]
  • blk.1.attn_k.weight
    F16
    [512, 512]
  • blk.1.attn_output.bias
    F32
    [512]
  • blk.1.attn_output.weight
    F16
    [512, 512]
  • blk.1.attn_output_norm.bias
    F32
    [512]
  • blk.1.attn_output_norm.weight
    F32
    [512]
  • blk.1.attn_q.bias
    F32
    [512]
  • blk.1.attn_q.weight
    F16
    [512, 512]
  • blk.1.attn_v.bias
    F32
    [512]
  • blk.1.attn_v.weight
    F16
    [512, 512]
  • blk.1.ffn_down.bias
    F32
    [512]
  • blk.1.ffn_down.weight
    F16
    [2048, 512]
  • blk.1.ffn_gate.weight
    F16
    [512, 2048]
  • blk.1.ffn_up.weight
    F16
    [512, 2048]
  • blk.1.layer_output_norm.bias
    F32
    [512]
  • blk.1.layer_output_norm.weight
    F32
    [512]
    • blk.2
  • blk.2.attn_k.bias
    F32
    [512]
  • blk.2.attn_k.weight
    F16
    [512, 512]
  • blk.2.attn_output.bias
    F32
    [512]
  • blk.2.attn_output.weight
    F16
    [512, 512]
  • blk.2.attn_output_norm.bias
    F32
    [512]
  • blk.2.attn_output_norm.weight
    F32
    [512]
  • blk.2.attn_q.bias
    F32
    [512]
  • blk.2.attn_q.weight
    F16
    [512, 512]
  • blk.2.attn_v.bias
    F32
    [512]
  • blk.2.attn_v.weight
    F16
    [512, 512]
  • blk.2.ffn_down.bias
    F32
    [512]
  • blk.2.ffn_down.weight
    F16
    [2048, 512]
  • blk.2.ffn_gate.weight
    F16
    [512, 2048]
  • blk.2.ffn_up.weight
    F16
    [512, 2048]
  • blk.2.layer_output_norm.bias
    F32
    [512]
  • blk.2.layer_output_norm.weight
    F32
    [512]
    • blk.3
  • blk.3.attn_k.bias
    F32
    [512]
  • blk.3.attn_k.weight
    F16
    [512, 512]
  • blk.3.attn_output.bias
    F32
    [512]
  • blk.3.attn_output.weight
    F16
    [512, 512]
  • blk.3.attn_output_norm.bias
    F32
    [512]
  • blk.3.attn_output_norm.weight
    F32
    [512]
  • blk.3.attn_q.bias
    F32
    [512]
  • blk.3.attn_q.weight
    F16
    [512, 512]
  • blk.3.attn_v.bias
    F32
    [512]
  • blk.3.attn_v.weight
    F16
    [512, 512]
  • blk.3.ffn_down.bias
    F32
    [512]
  • blk.3.ffn_down.weight
    F16
    [2048, 512]
  • blk.3.ffn_gate.weight
    F16
    [512, 2048]
  • blk.3.ffn_up.weight
    F16
    [512, 2048]
  • blk.3.layer_output_norm.bias
    F32
    [512]
  • blk.3.layer_output_norm.weight
    F32
    [512]
  • token_embd_norm.bias
    F32
    [512]
  • token_embd_norm.weight
    F32
    [512]
  • token_types.weight
    F32
    [512, 2]