all-minilm

Embedding models on very large sentence level datasets.

Embedding 22M 33M

81.9K Pulls Updated 4 months ago

... /
model
797b70c4edf8 · 46MB
    Metadata
  • general.architecture
    bert
  • general.file_type
    F16
  • bert.attention.causal
    false
  • bert.attention.head_count
    12
  • bert.attention.layer_norm_epsilon
    1e-12
  • bert.block_count
    6
  • bert.context_length
    512
  • bert.embedding_length
    384
  • bert.feed_forward_length
    1536
  • bert.pooling_type
    CLS token pooling
  • tokenizer.ggml.bos_token_id
    101
  • tokenizer.ggml.cls_token_id
    101
  • tokenizer.ggml.eos_token_id
    102
  • tokenizer.ggml.mask_token_id
    103
  • tokenizer.ggml.model
    bert
  • tokenizer.ggml.padding_token_id
    0
  • tokenizer.ggml.scores
    [-1000 -1000 -1000 -1000 -1000 ...]
  • 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
    • Tensors
  • Name
    Type
    Shape
  • token_embd.weight
    F16
    [384 30522]
  • position_embd.weight
    F16
    [384 512]
  • token_embd_norm.weight
    F32
    [384]
  • token_embd_norm.bias
    F32
    [384]
    • blk.0
  • blk.0.attn_q.weight
    F16
    [384 384]
  • blk.0.attn_q.bias
    F32
    [384]
  • blk.0.attn_k.weight
    F16
    [384 384]
  • blk.0.attn_k.bias
    F32
    [384]
  • blk.0.attn_v.weight
    F16
    [384 384]
  • blk.0.attn_v.bias
    F32
    [384]
  • blk.0.attn_output.weight
    F16
    [384 384]
  • blk.0.attn_output.bias
    F32
    [384]
  • blk.0.attn_output_norm.weight
    F32
    [384]
  • blk.0.attn_output_norm.bias
    F32
    [384]
  • blk.0.ffn_up.weight
    F16
    [384 1536]
  • blk.0.ffn_up.bias
    F32
    [1536]
  • blk.0.ffn_down.weight
    F16
    [1536 384]
  • blk.0.ffn_down.bias
    F32
    [384]
  • blk.0.layer_output_norm.weight
    F32
    [384]
  • blk.0.layer_output_norm.bias
    F32
    [384]
    • blk.1
  • blk.1.attn_q.weight
    F16
    [384 384]
  • blk.1.attn_q.bias
    F32
    [384]
  • blk.1.attn_k.weight
    F16
    [384 384]
  • blk.1.attn_k.bias
    F32
    [384]
  • blk.1.attn_v.weight
    F16
    [384 384]
  • blk.1.attn_v.bias
    F32
    [384]
  • blk.1.attn_output.weight
    F16
    [384 384]
  • blk.1.attn_output.bias
    F32
    [384]
  • blk.1.attn_output_norm.weight
    F32
    [384]
  • blk.1.attn_output_norm.bias
    F32
    [384]
  • blk.1.ffn_up.weight
    F16
    [384 1536]
  • blk.1.ffn_up.bias
    F32
    [1536]
  • blk.1.ffn_down.weight
    F16
    [1536 384]
  • blk.1.ffn_down.bias
    F32
    [384]
  • blk.1.layer_output_norm.weight
    F32
    [384]
  • blk.1.layer_output_norm.bias
    F32
    [384]
    • blk.2
  • blk.2.attn_q.weight
    F16
    [384 384]
  • blk.2.attn_q.bias
    F32
    [384]
  • blk.2.attn_k.weight
    F16
    [384 384]
  • blk.2.attn_k.bias
    F32
    [384]
  • blk.2.attn_v.weight
    F16
    [384 384]
  • blk.2.attn_v.bias
    F32
    [384]
  • blk.2.attn_output.weight
    F16
    [384 384]
  • blk.2.attn_output.bias
    F32
    [384]
  • blk.2.attn_output_norm.weight
    F32
    [384]
  • blk.2.attn_output_norm.bias
    F32
    [384]
  • blk.2.ffn_up.weight
    F16
    [384 1536]
  • blk.2.ffn_up.bias
    F32
    [1536]
  • blk.2.ffn_down.weight
    F16
    [1536 384]
  • blk.2.ffn_down.bias
    F32
    [384]
  • blk.2.layer_output_norm.weight
    F32
    [384]
  • blk.2.layer_output_norm.bias
    F32
    [384]
    • blk.3
  • blk.3.attn_q.weight
    F16
    [384 384]
  • blk.3.attn_q.bias
    F32
    [384]
  • blk.3.attn_k.weight
    F16
    [384 384]
  • blk.3.attn_k.bias
    F32
    [384]
  • blk.3.attn_v.weight
    F16
    [384 384]
  • blk.3.attn_v.bias
    F32
    [384]
  • blk.3.attn_output.weight
    F16
    [384 384]
  • blk.3.attn_output.bias
    F32
    [384]
  • blk.3.attn_output_norm.weight
    F32
    [384]
  • blk.3.attn_output_norm.bias
    F32
    [384]
  • blk.3.ffn_up.weight
    F16
    [384 1536]
  • blk.3.ffn_up.bias
    F32
    [1536]
  • blk.3.ffn_down.weight
    F16
    [1536 384]
  • blk.3.ffn_down.bias
    F32
    [384]
  • blk.3.layer_output_norm.weight
    F32
    [384]
  • blk.3.layer_output_norm.bias
    F32
    [384]
    • blk.4
  • blk.4.attn_q.weight
    F16
    [384 384]
  • blk.4.attn_q.bias
    F32
    [384]
  • blk.4.attn_k.weight
    F16
    [384 384]
  • blk.4.attn_k.bias
    F32
    [384]
  • blk.4.attn_v.weight
    F16
    [384 384]
  • blk.4.attn_v.bias
    F32
    [384]
  • blk.4.attn_output.weight
    F16
    [384 384]
  • blk.4.attn_output.bias
    F32
    [384]
  • blk.4.attn_output_norm.weight
    F32
    [384]
  • blk.4.attn_output_norm.bias
    F32
    [384]
  • blk.4.ffn_up.weight
    F16
    [384 1536]
  • blk.4.ffn_up.bias
    F32
    [1536]
  • blk.4.ffn_down.weight
    F16
    [1536 384]
  • blk.4.ffn_down.bias
    F32
    [384]
  • blk.4.layer_output_norm.weight
    F32
    [384]
  • blk.4.layer_output_norm.bias
    F32
    [384]
    • blk.5
  • blk.5.attn_q.weight
    F16
    [384 384]
  • blk.5.attn_q.bias
    F32
    [384]
  • blk.5.attn_k.weight
    F16
    [384 384]
  • blk.5.attn_k.bias
    F32
    [384]
  • blk.5.attn_v.weight
    F16
    [384 384]
  • blk.5.attn_v.bias
    F32
    [384]
  • blk.5.attn_output.weight
    F16
    [384 384]
  • blk.5.attn_output.bias
    F32
    [384]
  • blk.5.attn_output_norm.weight
    F32
    [384]
  • blk.5.attn_output_norm.bias
    F32
    [384]
  • blk.5.ffn_up.weight
    F16
    [384 1536]
  • blk.5.ffn_up.bias
    F32
    [1536]
  • blk.5.ffn_down.weight
    F16
    [1536 384]
  • blk.5.ffn_down.bias
    F32
    [384]
  • blk.5.layer_output_norm.weight
    F32
    [384]
  • blk.5.layer_output_norm.bias
    F32
    [384]
  • token_types.weight
    F32
    [384 2]