granite-embedding:30m-en-fp16/model

granite-embedding

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

21.8K Pulls Updated 2 months ago

granite-embedding:30m-en-fp16 ... /

model

27d24c87a53d · 63MB

Metadata

bert.attention.causal

false

false
bert.attention.head_count

12

12
bert.attention.layer_norm_epsilon

1e-12

1e-12
bert.block_count

6

6
bert.context_length

512

512
bert.embedding_length

384

384
bert.feed_forward_length

1536

1536
bert.pooling_type

2

2
general.architecture

bert

bert
general.basename

granite-embedding

granite-embedding
general.file_type

1

1
general.finetune

english

english
general.languages

[en]

[en]
general.license

apache-2.0

apache-2.0
general.name

Granite Embedding 30m English

Granite Embedding 30m English
general.quantization_version

2

2
general.size_label

30M

30M
general.tags

[language, granite, embeddings, sentence-similarity]

[language, granite, embeddings, sentence-similarity]
general.type

model

model
tokenizer.ggml.add_bos_token

true

true
tokenizer.ggml.add_eos_token

true

true
tokenizer.ggml.bos_token_id

0

0
tokenizer.ggml.cls_token_id

0

0
tokenizer.ggml.eos_token_id

2

2
tokenizer.ggml.mask_token_id

50264

50264
tokenizer.ggml.merges

[Ġ t, Ġ a, h e, i n, r e, ...]

[Ġ t, Ġ a, h e, i n, r e, ...]
tokenizer.ggml.model

gpt2

gpt2
tokenizer.ggml.padding_token_id

1

1
tokenizer.ggml.pre

gpt-2

gpt-2
tokenizer.ggml.seperator_token_id

2

2
tokenizer.ggml.token_type

[3, 3, 3, 3, 1, ...]

[3, 3, 3, 3, 1, ...]
tokenizer.ggml.token_type_count

2

2
tokenizer.ggml.tokens

[<s>, <pad>, </s>, <unk>, ., ...]

[<s>, <pad>, </s>, <unk>, ., ...]
tokenizer.ggml.unknown_token_id

3

3

Tensor

Name

Type

Shape
token_embd.weight

F16

F16

[384, 50265]

blk.0

blk.0.attn_k.bias

F32

F32

[384]
blk.0.attn_k.weight

F16

F16

[384, 384]
blk.0.attn_output.bias

F32

F32

[384]
blk.0.attn_output.weight

F16

F16

[384, 384]
blk.0.attn_output_norm.bias

F32

F32

[384]
blk.0.attn_output_norm.weight

F32

F32

[384]
blk.0.attn_q.bias

F32

F32

[384]
blk.0.attn_q.weight

F16

F16

[384, 384]
blk.0.attn_v.bias

F32

F32

[384]
blk.0.attn_v.weight

F16

F16

[384, 384]
blk.0.ffn_down.bias

F32

F32

[384]
blk.0.ffn_down.weight

F16

F16

[1536, 384]
blk.0.ffn_up.bias

F32

F32

[1536]
blk.0.ffn_up.weight

F16

F16

[384, 1536]
blk.0.layer_output_norm.bias

F32

F32

[384]
blk.0.layer_output_norm.weight

F32

F32

[384]

blk.1

blk.1.attn_k.bias

F32

F32

[384]
blk.1.attn_k.weight

F16

F16

[384, 384]
blk.1.attn_output.bias

F32

F32

[384]
blk.1.attn_output.weight

F16

F16

[384, 384]
blk.1.attn_output_norm.bias

F32

F32

[384]
blk.1.attn_output_norm.weight

F32

F32

[384]
blk.1.attn_q.bias

F32

F32

[384]
blk.1.attn_q.weight

F16

F16

[384, 384]
blk.1.attn_v.bias

F32

F32

[384]
blk.1.attn_v.weight

F16

F16

[384, 384]
blk.1.ffn_down.bias

F32

F32

[384]
blk.1.ffn_down.weight

F16

F16

[1536, 384]
blk.1.ffn_up.bias

F32

F32

[1536]
blk.1.ffn_up.weight

F16

F16

[384, 1536]
blk.1.layer_output_norm.bias

F32

F32

[384]
blk.1.layer_output_norm.weight

F32

F32

[384]

blk.2

blk.2.attn_k.bias

F32

F32

[384]
blk.2.attn_k.weight

F16

F16

[384, 384]
blk.2.attn_output.bias

F32

F32

[384]
blk.2.attn_output.weight

F16

F16

[384, 384]
blk.2.attn_output_norm.bias

F32

F32

[384]
blk.2.attn_output_norm.weight

F32

F32

[384]
blk.2.attn_q.bias

F32

F32

[384]
blk.2.attn_q.weight

F16

F16

[384, 384]
blk.2.attn_v.bias

F32

F32

[384]
blk.2.attn_v.weight

F16

F16

[384, 384]
blk.2.ffn_down.bias

F32

F32

[384]
blk.2.ffn_down.weight

F16

F16

[1536, 384]
blk.2.ffn_up.bias

F32

F32

[1536]
blk.2.ffn_up.weight

F16

F16

[384, 1536]
blk.2.layer_output_norm.bias

F32

F32

[384]
blk.2.layer_output_norm.weight

F32

F32

[384]

blk.3

blk.3.attn_k.bias

F32

F32

[384]
blk.3.attn_k.weight

F16

F16

[384, 384]
blk.3.attn_output.bias

F32

F32

[384]
blk.3.attn_output.weight

F16

F16

[384, 384]
blk.3.attn_output_norm.bias

F32

F32

[384]
blk.3.attn_output_norm.weight

F32

F32

[384]
blk.3.attn_q.bias

F32

F32

[384]
blk.3.attn_q.weight

F16

F16

[384, 384]
blk.3.attn_v.bias

F32

F32

[384]
blk.3.attn_v.weight

F16

F16

[384, 384]
blk.3.ffn_down.bias

F32

F32

[384]
blk.3.ffn_down.weight

F16

F16

[1536, 384]
blk.3.ffn_up.bias

F32

F32

[1536]
blk.3.ffn_up.weight

F16

F16

[384, 1536]
blk.3.layer_output_norm.bias

F32

F32

[384]
blk.3.layer_output_norm.weight

F32

F32

[384]

blk.4

blk.4.attn_k.bias

F32

F32

[384]
blk.4.attn_k.weight

F16

F16

[384, 384]
blk.4.attn_output.bias

F32

F32

[384]
blk.4.attn_output.weight

F16

F16

[384, 384]
blk.4.attn_output_norm.bias

F32

F32

[384]
blk.4.attn_output_norm.weight

F32

F32

[384]
blk.4.attn_q.bias

F32

F32

[384]
blk.4.attn_q.weight

F16

F16

[384, 384]
blk.4.attn_v.bias

F32

F32

[384]
blk.4.attn_v.weight

F16

F16

[384, 384]
blk.4.ffn_down.bias

F32

F32

[384]
blk.4.ffn_down.weight

F16

F16

[1536, 384]
blk.4.ffn_up.bias

F32

F32

[1536]
blk.4.ffn_up.weight

F16

F16

[384, 1536]
blk.4.layer_output_norm.bias

F32

F32

[384]
blk.4.layer_output_norm.weight

F32

F32

[384]

blk.5

blk.5.attn_k.bias

F32

F32

[384]
blk.5.attn_k.weight

F16

F16

[384, 384]
blk.5.attn_output.bias

F32

F32

[384]
blk.5.attn_output.weight

F16

F16

[384, 384]
blk.5.attn_output_norm.bias

F32

F32

[384]
blk.5.attn_output_norm.weight

F32

F32

[384]
blk.5.attn_q.bias

F32

F32

[384]
blk.5.attn_q.weight

F16

F16

[384, 384]
blk.5.attn_v.bias

F32

F32

[384]
blk.5.attn_v.weight

F16

F16

[384, 384]
blk.5.ffn_down.bias

F32

F32

[384]
blk.5.ffn_down.weight

F16

F16

[1536, 384]
blk.5.ffn_up.bias

F32

F32

[1536]
blk.5.ffn_up.weight

F16

F16

[384, 1536]
blk.5.layer_output_norm.bias

F32

F32

[384]
blk.5.layer_output_norm.weight

F32

F32

[384]

position_embd.weight

F32

F32

[384, 512]
token_embd_norm.bias

F32

F32

[384]
token_embd_norm.weight

F32

F32

[384]
token_types.weight

F32

F32

[384, 2]