toshk0/nomic-embed-text-v2-moe:Q6

toshk0/

nomic-embed-text-v2-moe:Q6_K

393 Downloads Updated 2 months ago

nomic-embed-text-v2-moe is a SoTA multilingual MoE text embedding model

embedding

nomic-embed-text-v2-moe:Q6_K ... /

model

e74dafe6932f · 397MB

Metadata

general.architecture

nomic-bert-moe

nomic-bert-moe
general.file_type

Q6_K

Q6_K
nomic-bert-moe.attention.causal

false

false
nomic-bert-moe.attention.head_count

12

12
nomic-bert-moe.attention.layer_norm_epsilon

1e-05

1e-05
nomic-bert-moe.block_count

12

12
nomic-bert-moe.context_length

512

512
nomic-bert-moe.embedding_length

768

768
nomic-bert-moe.expert_count

8

8
nomic-bert-moe.expert_used_count

2

2
nomic-bert-moe.feed_forward_length

3072

3072
nomic-bert-moe.moe_every_n_layers

2

2
nomic-bert-moe.pooling_type

Mean

Mean
nomic-bert-moe.rope.freq_base

10000

10000
tokenizer.ggml.add_bos_token

true

true
tokenizer.ggml.add_eos_token

true

true
tokenizer.ggml.add_space_prefix

true

true
tokenizer.ggml.bos_token_id

0

0
tokenizer.ggml.eos_token_id

2

2
tokenizer.ggml.mask_token_id

250001

250001
tokenizer.ggml.model

t5

t5
tokenizer.ggml.padding_token_id

1

1
tokenizer.ggml.pre

default

default
tokenizer.ggml.precompiled_charsmap

[0, 180, 2, 0, 0, ...]

[0, 180, 2, 0, 0, ...]
tokenizer.ggml.remove_extra_whitespaces

true

true
tokenizer.ggml.scores

[0, 0, 0, 0, -3.4635427, ...]

[0, 0, 0, 0, -3.4635427, ...]
tokenizer.ggml.seperator_token_id

2

2
tokenizer.ggml.token_type

[3, 3, 3, 2, 1, ...]

[3, 3, 3, 2, 1, ...]
tokenizer.ggml.token_type_count

1

1
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

Q6_K

Q6_K

[768, 250048]

blk.0

blk.0.attn_output.bias

F32

F32

[768]
blk.0.attn_output.weight

Q6_K

Q6_K

[768, 768]
blk.0.attn_output_norm.bias

F32

F32

[768]
blk.0.attn_output_norm.weight

F32

F32

[768]
blk.0.attn_qkv.bias

F32

F32

[2304]
blk.0.attn_qkv.weight

Q6_K

Q6_K

[768, 2304]
blk.0.ffn_down.bias

F32

F32

[768]
blk.0.ffn_down.weight

Q6_K

Q6_K

[3072, 768]
blk.0.ffn_up.bias

F32

F32

[3072]
blk.0.ffn_up.weight

Q6_K

Q6_K

[768, 3072]
blk.0.layer_output_norm.bias

F32

F32

[768]
blk.0.layer_output_norm.weight

F32

F32

[768]

blk.1

blk.1.attn_output.bias

F32

F32

[768]
blk.1.attn_output.weight

Q6_K

Q6_K

[768, 768]
blk.1.attn_output_norm.bias

F32

F32

[768]
blk.1.attn_output_norm.weight

F32

F32

[768]
blk.1.attn_qkv.bias

F32

F32

[2304]
blk.1.attn_qkv.weight

Q6_K

Q6_K

[768, 2304]
blk.1.ffn_down_exps.weight

Q6_K

Q6_K

[3072, 768, 8]
blk.1.ffn_gate_inp.weight

F32

F32

[768, 8]
blk.1.ffn_up_exps.weight

Q6_K

Q6_K

[768, 3072, 8]
blk.1.layer_output_norm.bias

F32

F32

[768]
blk.1.layer_output_norm.weight

F32

F32

[768]

blk.2

blk.2.attn_output.bias

F32

F32

[768]
blk.2.attn_output.weight

Q6_K

Q6_K

[768, 768]
blk.2.attn_output_norm.bias

F32

F32

[768]
blk.2.attn_output_norm.weight

F32

F32

[768]
blk.2.attn_qkv.bias

F32

F32

[2304]
blk.2.attn_qkv.weight

Q6_K

Q6_K

[768, 2304]
blk.2.ffn_down.bias

F32

F32

[768]
blk.2.ffn_down.weight

Q6_K

Q6_K

[3072, 768]
blk.2.ffn_up.bias

F32

F32

[3072]
blk.2.ffn_up.weight

Q6_K

Q6_K

[768, 3072]
blk.2.layer_output_norm.bias

F32

F32

[768]
blk.2.layer_output_norm.weight

F32

F32

[768]

blk.3

blk.3.attn_output.bias

F32

F32

[768]
blk.3.attn_output.weight

Q6_K

Q6_K

[768, 768]
blk.3.attn_output_norm.bias

F32

F32

[768]
blk.3.attn_output_norm.weight

F32

F32

[768]
blk.3.attn_qkv.bias

F32

F32

[2304]
blk.3.attn_qkv.weight

Q6_K

Q6_K

[768, 2304]
blk.3.ffn_down_exps.weight

Q6_K

Q6_K

[3072, 768, 8]
blk.3.ffn_gate_inp.weight

F32

F32

[768, 8]
blk.3.ffn_up_exps.weight

Q6_K

Q6_K

[768, 3072, 8]
blk.3.layer_output_norm.bias

F32

F32

[768]
blk.3.layer_output_norm.weight

F32

F32

[768]

blk.4

blk.4.attn_output.bias

F32

F32

[768]
blk.4.attn_output.weight

Q6_K

Q6_K

[768, 768]
blk.4.attn_output_norm.bias

F32

F32

[768]
blk.4.attn_output_norm.weight

F32

F32

[768]
blk.4.attn_qkv.bias

F32

F32

[2304]
blk.4.attn_qkv.weight

Q6_K

Q6_K

[768, 2304]
blk.4.ffn_down.bias

F32

F32

[768]
blk.4.ffn_down.weight

Q6_K

Q6_K

[3072, 768]
blk.4.ffn_up.bias

F32

F32

[3072]
blk.4.ffn_up.weight

Q6_K

Q6_K

[768, 3072]
blk.4.layer_output_norm.bias

F32

F32

[768]
blk.4.layer_output_norm.weight

F32

F32

[768]

blk.5

blk.5.attn_output.bias

F32

F32

[768]
blk.5.attn_output.weight

Q6_K

Q6_K

[768, 768]
blk.5.attn_output_norm.bias

F32

F32

[768]
blk.5.attn_output_norm.weight

F32

F32

[768]
blk.5.attn_qkv.bias

F32

F32

[2304]
blk.5.attn_qkv.weight

Q6_K

Q6_K

[768, 2304]
blk.5.ffn_down_exps.weight

Q6_K

Q6_K

[3072, 768, 8]
blk.5.ffn_gate_inp.weight

F32

F32

[768, 8]
blk.5.ffn_up_exps.weight

Q6_K

Q6_K

[768, 3072, 8]
blk.5.layer_output_norm.bias

F32

F32

[768]
blk.5.layer_output_norm.weight

F32

F32

[768]

blk.6

blk.6.attn_output.bias

F32

F32

[768]
blk.6.attn_output.weight

Q6_K

Q6_K

[768, 768]
blk.6.attn_output_norm.bias

F32

F32

[768]
blk.6.attn_output_norm.weight

F32

F32

[768]
blk.6.attn_qkv.bias

F32

F32

[2304]
blk.6.attn_qkv.weight

Q6_K

Q6_K

[768, 2304]
blk.6.ffn_down.bias

F32

F32

[768]
blk.6.ffn_down.weight

Q6_K

Q6_K

[3072, 768]
blk.6.ffn_up.bias

F32

F32

[3072]
blk.6.ffn_up.weight

Q6_K

Q6_K

[768, 3072]
blk.6.layer_output_norm.bias

F32

F32

[768]
blk.6.layer_output_norm.weight

F32

F32

[768]

blk.7

blk.7.attn_output.bias

F32

F32

[768]
blk.7.attn_output.weight

Q6_K

Q6_K

[768, 768]
blk.7.attn_output_norm.bias

F32

F32

[768]
blk.7.attn_output_norm.weight

F32

F32

[768]
blk.7.attn_qkv.bias

F32

F32

[2304]
blk.7.attn_qkv.weight

Q6_K

Q6_K

[768, 2304]
blk.7.ffn_down_exps.weight

Q6_K

Q6_K

[3072, 768, 8]
blk.7.ffn_gate_inp.weight

F32

F32

[768, 8]
blk.7.ffn_up_exps.weight

Q6_K

Q6_K

[768, 3072, 8]
blk.7.layer_output_norm.bias

F32

F32

[768]
blk.7.layer_output_norm.weight

F32

F32

[768]

blk.8

blk.8.attn_output.bias

F32

F32

[768]
blk.8.attn_output.weight

Q6_K

Q6_K

[768, 768]
blk.8.attn_output_norm.bias

F32

F32

[768]
blk.8.attn_output_norm.weight

F32

F32

[768]
blk.8.attn_qkv.bias

F32

F32

[2304]
blk.8.attn_qkv.weight

Q6_K

Q6_K

[768, 2304]
blk.8.ffn_down.bias

F32

F32

[768]
blk.8.ffn_down.weight

Q6_K

Q6_K

[3072, 768]
blk.8.ffn_up.bias

F32

F32

[3072]
blk.8.ffn_up.weight

Q6_K

Q6_K

[768, 3072]
blk.8.layer_output_norm.bias

F32

F32

[768]
blk.8.layer_output_norm.weight

F32

F32

[768]

blk.9

blk.9.attn_output.bias

F32

F32

[768]
blk.9.attn_output.weight

Q6_K

Q6_K

[768, 768]
blk.9.attn_output_norm.bias

F32

F32

[768]
blk.9.attn_output_norm.weight

F32

F32

[768]
blk.9.attn_qkv.bias

F32

F32

[2304]
blk.9.attn_qkv.weight

Q6_K

Q6_K

[768, 2304]
blk.9.ffn_down_exps.weight

Q6_K

Q6_K

[3072, 768, 8]
blk.9.ffn_gate_inp.weight

F32

F32

[768, 8]
blk.9.ffn_up_exps.weight

Q6_K

Q6_K

[768, 3072, 8]
blk.9.layer_output_norm.bias

F32

F32

[768]
blk.9.layer_output_norm.weight

F32

F32

[768]

blk.10

blk.10.attn_output.bias

F32

F32

[768]
blk.10.attn_output.weight

Q6_K

Q6_K

[768, 768]
blk.10.attn_output_norm.bias

F32

F32

[768]
blk.10.attn_output_norm.weight

F32

F32

[768]
blk.10.attn_qkv.bias

F32

F32

[2304]
blk.10.attn_qkv.weight

Q6_K

Q6_K

[768, 2304]
blk.10.ffn_down.bias

F32

F32

[768]
blk.10.ffn_down.weight

Q6_K

Q6_K

[3072, 768]
blk.10.ffn_up.bias

F32

F32

[3072]
blk.10.ffn_up.weight

Q6_K

Q6_K

[768, 3072]
blk.10.layer_output_norm.bias

F32

F32

[768]
blk.10.layer_output_norm.weight

F32

F32

[768]

blk.11

blk.11.attn_output.bias

F32

F32

[768]
blk.11.attn_output.weight

Q6_K

Q6_K

[768, 768]
blk.11.attn_output_norm.bias

F32

F32

[768]
blk.11.attn_output_norm.weight

F32

F32

[768]
blk.11.attn_qkv.bias

F32

F32

[2304]
blk.11.attn_qkv.weight

Q6_K

Q6_K

[768, 2304]
blk.11.ffn_down_exps.weight

Q6_K

Q6_K

[3072, 768, 8]
blk.11.ffn_gate_inp.weight

F32

F32

[768, 8]
blk.11.ffn_up_exps.weight

Q6_K

Q6_K

[768, 3072, 8]
blk.11.layer_output_norm.bias

F32

F32

[768]
blk.11.layer_output_norm.weight

F32

F32

[768]

token_embd_norm.bias

F32

F32

[768]
token_embd_norm.weight

F32

F32

[768]
token_types.weight

F32

F32

[768]