granite-embedding:latest

92.4K 8 months ago

The IBM Granite Embedding 30M and 278M models models are text-only dense biencoder embedding models, with 30M available in English only and 278M serving multilingual use cases.

embedding 30m 278m
... /
model
27d24c87a53d · 63MB
    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
  • tokenizer.ggml.add_bos_token
    true
  • tokenizer.ggml.add_eos_token
    true
  • tokenizer.ggml.bos_token_id
    0
  • tokenizer.ggml.cls_token_id
    0
  • tokenizer.ggml.eos_token_id
    2
  • tokenizer.ggml.mask_token_id
    50264
  • tokenizer.ggml.merges
    [Ġ t, Ġ a, h e, i n, r e, ...]
  • tokenizer.ggml.model
    gpt2
  • tokenizer.ggml.padding_token_id
    1
  • tokenizer.ggml.pre
    gpt-2
  • tokenizer.ggml.seperator_token_id
    2
  • tokenizer.ggml.token_type
    [3, 3, 3, 3, 1, ...]
  • tokenizer.ggml.token_type_count
    2
  • tokenizer.ggml.tokens
    [<s>, <pad>, </s>, <unk>, ., ...]
  • tokenizer.ggml.unknown_token_id
    3
    • Tensor
  • token_embd.weight
    F16
    [384, 50265]
    • blk.0
  • blk.0.attn_k.bias
    F32
    [384]
  • blk.0.attn_k.weight
    F16
    [384, 384]
  • blk.0.attn_output.bias
    F32
    [384]
  • blk.0.attn_output.weight
    F16
    [384, 384]
  • blk.0.attn_output_norm.bias
    F32
    [384]
  • blk.0.attn_output_norm.weight
    F32
    [384]
  • blk.0.attn_q.bias
    F32
    [384]
  • blk.0.attn_q.weight
    F16
    [384, 384]
  • blk.0.attn_v.bias
    F32
    [384]
  • blk.0.attn_v.weight
    F16
    [384, 384]
  • blk.0.ffn_down.bias
    F32
    [384]
  • blk.0.ffn_down.weight
    F16
    [1536, 384]
  • blk.0.ffn_up.bias
    F32
    [1536]
  • blk.0.ffn_up.weight
    F16
    [384, 1536]
  • blk.0.layer_output_norm.bias
    F32
    [384]
  • blk.0.layer_output_norm.weight
    F32
    [384]
    • blk.1
  • blk.1.attn_k.bias
    F32
    [384]
  • blk.1.attn_k.weight
    F16
    [384, 384]
  • blk.1.attn_output.bias
    F32
    [384]
  • blk.1.attn_output.weight
    F16
    [384, 384]
  • blk.1.attn_output_norm.bias
    F32
    [384]
  • blk.1.attn_output_norm.weight
    F32
    [384]
  • blk.1.attn_q.bias
    F32
    [384]
  • blk.1.attn_q.weight
    F16
    [384, 384]
  • blk.1.attn_v.bias
    F32
    [384]
  • blk.1.attn_v.weight
    F16
    [384, 384]
  • blk.1.ffn_down.bias
    F32
    [384]
  • blk.1.ffn_down.weight
    F16
    [1536, 384]
  • blk.1.ffn_up.bias
    F32
    [1536]
  • blk.1.ffn_up.weight
    F16
    [384, 1536]
  • blk.1.layer_output_norm.bias
    F32
    [384]
  • blk.1.layer_output_norm.weight
    F32
    [384]
    • blk.2
  • blk.2.attn_k.bias
    F32
    [384]
  • blk.2.attn_k.weight
    F16
    [384, 384]
  • blk.2.attn_output.bias
    F32
    [384]
  • blk.2.attn_output.weight
    F16
    [384, 384]
  • blk.2.attn_output_norm.bias
    F32
    [384]
  • blk.2.attn_output_norm.weight
    F32
    [384]
  • blk.2.attn_q.bias
    F32
    [384]
  • blk.2.attn_q.weight
    F16
    [384, 384]
  • blk.2.attn_v.bias
    F32
    [384]
  • blk.2.attn_v.weight
    F16
    [384, 384]
  • blk.2.ffn_down.bias
    F32
    [384]
  • blk.2.ffn_down.weight
    F16
    [1536, 384]
  • blk.2.ffn_up.bias
    F32
    [1536]
  • blk.2.ffn_up.weight
    F16
    [384, 1536]
  • blk.2.layer_output_norm.bias
    F32
    [384]
  • blk.2.layer_output_norm.weight
    F32
    [384]
    • blk.3
  • blk.3.attn_k.bias
    F32
    [384]
  • blk.3.attn_k.weight
    F16
    [384, 384]
  • blk.3.attn_output.bias
    F32
    [384]
  • blk.3.attn_output.weight
    F16
    [384, 384]
  • blk.3.attn_output_norm.bias
    F32
    [384]
  • blk.3.attn_output_norm.weight
    F32
    [384]
  • blk.3.attn_q.bias
    F32
    [384]
  • blk.3.attn_q.weight
    F16
    [384, 384]
  • blk.3.attn_v.bias
    F32
    [384]
  • blk.3.attn_v.weight
    F16
    [384, 384]
  • blk.3.ffn_down.bias
    F32
    [384]
  • blk.3.ffn_down.weight
    F16
    [1536, 384]
  • blk.3.ffn_up.bias
    F32
    [1536]
  • blk.3.ffn_up.weight
    F16
    [384, 1536]
  • blk.3.layer_output_norm.bias
    F32
    [384]
  • blk.3.layer_output_norm.weight
    F32
    [384]
    • blk.4
  • blk.4.attn_k.bias
    F32
    [384]
  • blk.4.attn_k.weight
    F16
    [384, 384]
  • blk.4.attn_output.bias
    F32
    [384]
  • blk.4.attn_output.weight
    F16
    [384, 384]
  • blk.4.attn_output_norm.bias
    F32
    [384]
  • blk.4.attn_output_norm.weight
    F32
    [384]
  • blk.4.attn_q.bias
    F32
    [384]
  • blk.4.attn_q.weight
    F16
    [384, 384]
  • blk.4.attn_v.bias
    F32
    [384]
  • blk.4.attn_v.weight
    F16
    [384, 384]
  • blk.4.ffn_down.bias
    F32
    [384]
  • blk.4.ffn_down.weight
    F16
    [1536, 384]
  • blk.4.ffn_up.bias
    F32
    [1536]
  • blk.4.ffn_up.weight
    F16
    [384, 1536]
  • blk.4.layer_output_norm.bias
    F32
    [384]
  • blk.4.layer_output_norm.weight
    F32
    [384]
    • blk.5
  • blk.5.attn_k.bias
    F32
    [384]
  • blk.5.attn_k.weight
    F16
    [384, 384]
  • blk.5.attn_output.bias
    F32
    [384]
  • blk.5.attn_output.weight
    F16
    [384, 384]
  • blk.5.attn_output_norm.bias
    F32
    [384]
  • blk.5.attn_output_norm.weight
    F32
    [384]
  • blk.5.attn_q.bias
    F32
    [384]
  • blk.5.attn_q.weight
    F16
    [384, 384]
  • blk.5.attn_v.bias
    F32
    [384]
  • blk.5.attn_v.weight
    F16
    [384, 384]
  • blk.5.ffn_down.bias
    F32
    [384]
  • blk.5.ffn_down.weight
    F16
    [1536, 384]
  • blk.5.ffn_up.bias
    F32
    [1536]
  • blk.5.ffn_up.weight
    F16
    [384, 1536]
  • blk.5.layer_output_norm.bias
    F32
    [384]
  • blk.5.layer_output_norm.weight
    F32
    [384]
  • position_embd.weight
    F32
    [384, 512]
  • token_embd_norm.bias
    F32
    [384]
  • token_embd_norm.weight
    F32
    [384]
  • token_types.weight
    F32
    [384, 2]