GitHub
182.4K
Models
Docs
Pricing
Sign in
Models
Docs
Pricing
Sign in
johnwaynelucas
/
embeddinggemma-2
:gguf
3
Downloads
Updated
5 minutes ago
Cancel
vision
embedding
audio
embeddinggemma-2:gguf
...
/
model
f315cbbb30dd · 558MB
Metadata
general.architecture
gemma-embedding2
gemma-embedding2
general.file_type
BF16
BF16
gemma-embedding2.attention.causal
false
false
gemma-embedding2.attention.head_count
4
4
gemma-embedding2.attention.head_count_kv
[2, 2, 2, 2, 2, ...]
[2, 2, 2, 2, 2, ...]
gemma-embedding2.attention.key_length
512
512
gemma-embedding2.attention.key_length_swa
256
256
gemma-embedding2.attention.layer_norm_rms_epsilon
1e-06
1e-06
gemma-embedding2.attention.shared_kv_layers
0
0
gemma-embedding2.attention.sliding_window
1024
1024
gemma-embedding2.attention.sliding_window_pattern
[true, true, true, true, true, ...]
[true, true, true, true, true, ...]
gemma-embedding2.attention.value_length
512
512
gemma-embedding2.attention.value_length_swa
256
256
gemma-embedding2.block_count
24
24
gemma-embedding2.context_length
262144
262144
gemma-embedding2.embedding_length
512
512
gemma-embedding2.embedding_length_out
768
768
gemma-embedding2.embedding_length_per_layer_input
512
512
gemma-embedding2.feed_forward_length
2048
2048
gemma-embedding2.pooling_type
Mean
Mean
gemma-embedding2.rope.dimension_count
512
512
gemma-embedding2.rope.dimension_count_swa
256
256
gemma-embedding2.rope.freq_base
1e+06
1e+06
gemma-embedding2.rope.freq_base_swa
10000
10000
tokenizer.ggml.add_bos_token
true
true
tokenizer.ggml.add_eos_token
true
true
tokenizer.ggml.add_sep_token
false
false
tokenizer.ggml.add_space_prefix
false
false
tokenizer.ggml.bos_token_id
2
2
tokenizer.ggml.eos_token_id
1
1
tokenizer.ggml.mask_token_id
4
4
tokenizer.ggml.merges
[ , ▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁ ▁, , , ▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁ ▁▁, ...]
[ , ▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁ ▁, , , ▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁ ▁▁, ...]
tokenizer.ggml.model
gemma4
gemma4
tokenizer.ggml.padding_token_id
0
0
tokenizer.ggml.scores
[-1000, -1000, -1000, -1000, -1000, ...]
[-1000, -1000, -1000, -1000, -1000, ...]
tokenizer.ggml.token_type
[3, 3, 3, 3, 3, ...]
[3, 3, 3, 3, 3, ...]
tokenizer.ggml.tokens
[<pad>, <eos>, <bos>, <unk>, <mask>, ...]
[<pad>, <eos>, <bos>, <unk>, <mask>, ...]
tokenizer.ggml.unknown_token_id
3
3
Tensor
Name
Type
Shape
token_embd.weight
BF16
BF16
[512, 262144]
blk.0
blk.0.attn_k.weight
BF16
BF16
[512, 512]
blk.0.attn_k_norm.weight
F32
F32
[256]
blk.0.attn_norm.weight
F32
F32
[512]
blk.0.attn_output.weight
BF16
BF16
[1024, 512]
blk.0.attn_q.weight
BF16
BF16
[512, 1024]
blk.0.attn_q_norm.weight
F32
F32
[256]
blk.0.attn_v.weight
BF16
BF16
[512, 512]
blk.0.ffn_down.weight
BF16
BF16
[2048, 512]
blk.0.ffn_gate.weight
BF16
BF16
[512, 2048]
blk.0.ffn_norm.weight
F32
F32
[512]
blk.0.ffn_up.weight
BF16
BF16
[512, 2048]
blk.0.inp_gate.weight
BF16
BF16
[512, 512]
blk.0.layer_output_scale.weight
F32
F32
[1]
blk.0.post_attention_norm.weight
F32
F32
[512]
blk.0.post_ffw_norm.weight
F32
F32
[512]
blk.0.post_norm.weight
F32
F32
[512]
blk.0.proj.weight
BF16
BF16
[512, 512]
blk.1
blk.1.attn_k.weight
BF16
BF16
[512, 512]
blk.1.attn_k_norm.weight
F32
F32
[256]
blk.1.attn_norm.weight
F32
F32
[512]
blk.1.attn_output.weight
BF16
BF16
[1024, 512]
blk.1.attn_q.weight
BF16
BF16
[512, 1024]
blk.1.attn_q_norm.weight
F32
F32
[256]
blk.1.attn_v.weight
BF16
BF16
[512, 512]
blk.1.ffn_down.weight
BF16
BF16
[2048, 512]
blk.1.ffn_gate.weight
BF16
BF16
[512, 2048]
blk.1.ffn_norm.weight
F32
F32
[512]
blk.1.ffn_up.weight
BF16
BF16
[512, 2048]
blk.1.inp_gate.weight
BF16
BF16
[512, 512]
blk.1.layer_output_scale.weight
F32
F32
[1]
blk.1.post_attention_norm.weight
F32
F32
[512]
blk.1.post_ffw_norm.weight
F32
F32
[512]
blk.1.post_norm.weight
F32
F32
[512]
blk.1.proj.weight
BF16
BF16
[512, 512]
blk.2
blk.2.attn_k.weight
BF16
BF16
[512, 512]
blk.2.attn_k_norm.weight
F32
F32
[256]
blk.2.attn_norm.weight
F32
F32
[512]
blk.2.attn_output.weight
BF16
BF16
[1024, 512]
blk.2.attn_q.weight
BF16
BF16
[512, 1024]
blk.2.attn_q_norm.weight
F32
F32
[256]
blk.2.attn_v.weight
BF16
BF16
[512, 512]
blk.2.ffn_down.weight
BF16
BF16
[2048, 512]
blk.2.ffn_gate.weight
BF16
BF16
[512, 2048]
blk.2.ffn_norm.weight
F32
F32
[512]
blk.2.ffn_up.weight
BF16
BF16
[512, 2048]
blk.2.inp_gate.weight
BF16
BF16
[512, 512]
blk.2.layer_output_scale.weight
F32
F32
[1]
blk.2.post_attention_norm.weight
F32
F32
[512]
blk.2.post_ffw_norm.weight
F32
F32
[512]
blk.2.post_norm.weight
F32
F32
[512]
blk.2.proj.weight
BF16
BF16
[512, 512]
blk.3
blk.3.attn_k.weight
BF16
BF16
[512, 512]
blk.3.attn_k_norm.weight
F32
F32
[256]
blk.3.attn_norm.weight
F32
F32
[512]
blk.3.attn_output.weight
BF16
BF16
[1024, 512]
blk.3.attn_q.weight
BF16
BF16
[512, 1024]
blk.3.attn_q_norm.weight
F32
F32
[256]
blk.3.attn_v.weight
BF16
BF16
[512, 512]
blk.3.ffn_down.weight
BF16
BF16
[2048, 512]
blk.3.ffn_gate.weight
BF16
BF16
[512, 2048]
blk.3.ffn_norm.weight
F32
F32
[512]
blk.3.ffn_up.weight
BF16
BF16
[512, 2048]
blk.3.inp_gate.weight
BF16
BF16
[512, 512]
blk.3.layer_output_scale.weight
F32
F32
[1]
blk.3.post_attention_norm.weight
F32
F32
[512]
blk.3.post_ffw_norm.weight
F32
F32
[512]
blk.3.post_norm.weight
F32
F32
[512]
blk.3.proj.weight
BF16
BF16
[512, 512]
blk.4
blk.4.attn_k.weight
BF16
BF16
[512, 512]
blk.4.attn_k_norm.weight
F32
F32
[256]
blk.4.attn_norm.weight
F32
F32
[512]
blk.4.attn_output.weight
BF16
BF16
[1024, 512]
blk.4.attn_q.weight
BF16
BF16
[512, 1024]
blk.4.attn_q_norm.weight
F32
F32
[256]
blk.4.attn_v.weight
BF16
BF16
[512, 512]
blk.4.ffn_down.weight
BF16
BF16
[2048, 512]
blk.4.ffn_gate.weight
BF16
BF16
[512, 2048]
blk.4.ffn_norm.weight
F32
F32
[512]
blk.4.ffn_up.weight
BF16
BF16
[512, 2048]
blk.4.inp_gate.weight
BF16
BF16
[512, 512]
blk.4.layer_output_scale.weight
F32
F32
[1]
blk.4.post_attention_norm.weight
F32
F32
[512]
blk.4.post_ffw_norm.weight
F32
F32
[512]
blk.4.post_norm.weight
F32
F32
[512]
blk.4.proj.weight
BF16
BF16
[512, 512]
blk.5
blk.5.attn_k.weight
BF16
BF16
[512, 512]
blk.5.attn_k_norm.weight
F32
F32
[512]
blk.5.attn_norm.weight
F32
F32
[512]
blk.5.attn_output.weight
BF16
BF16
[2048, 512]
blk.5.attn_q.weight
BF16
BF16
[512, 2048]
blk.5.attn_q_norm.weight
F32
F32
[512]
blk.5.attn_v.weight
BF16
BF16
[512, 512]
blk.5.ffn_down.weight
BF16
BF16
[2048, 512]
blk.5.ffn_gate.weight
BF16
BF16
[512, 2048]
blk.5.ffn_norm.weight
F32
F32
[512]
blk.5.ffn_up.weight
BF16
BF16
[512, 2048]
blk.5.inp_gate.weight
BF16
BF16
[512, 512]
blk.5.layer_output_scale.weight
F32
F32
[1]
blk.5.post_attention_norm.weight
F32
F32
[512]
blk.5.post_ffw_norm.weight
F32
F32
[512]
blk.5.post_norm.weight
F32
F32
[512]
blk.5.proj.weight
BF16
BF16
[512, 512]
blk.6
blk.6.attn_k.weight
BF16
BF16
[512, 512]
blk.6.attn_k_norm.weight
F32
F32
[256]
blk.6.attn_norm.weight
F32
F32
[512]
blk.6.attn_output.weight
BF16
BF16
[1024, 512]
blk.6.attn_q.weight
BF16
BF16
[512, 1024]
blk.6.attn_q_norm.weight
F32
F32
[256]
blk.6.attn_v.weight
BF16
BF16
[512, 512]
blk.6.ffn_down.weight
BF16
BF16
[2048, 512]
blk.6.ffn_gate.weight
BF16
BF16
[512, 2048]
blk.6.ffn_norm.weight
F32
F32
[512]
blk.6.ffn_up.weight
BF16
BF16
[512, 2048]
blk.6.inp_gate.weight
BF16
BF16
[512, 512]
blk.6.layer_output_scale.weight
F32
F32
[1]
blk.6.post_attention_norm.weight
F32
F32
[512]
blk.6.post_ffw_norm.weight
F32
F32
[512]
blk.6.post_norm.weight
F32
F32
[512]
blk.6.proj.weight
BF16
BF16
[512, 512]
blk.7
blk.7.attn_k.weight
BF16
BF16
[512, 512]
blk.7.attn_k_norm.weight
F32
F32
[256]
blk.7.attn_norm.weight
F32
F32
[512]
blk.7.attn_output.weight
BF16
BF16
[1024, 512]
blk.7.attn_q.weight
BF16
BF16
[512, 1024]
blk.7.attn_q_norm.weight
F32
F32
[256]
blk.7.attn_v.weight
BF16
BF16
[512, 512]
blk.7.ffn_down.weight
BF16
BF16
[2048, 512]
blk.7.ffn_gate.weight
BF16
BF16
[512, 2048]
blk.7.ffn_norm.weight
F32
F32
[512]
blk.7.ffn_up.weight
BF16
BF16
[512, 2048]
blk.7.inp_gate.weight
BF16
BF16
[512, 512]
blk.7.layer_output_scale.weight
F32
F32
[1]
blk.7.post_attention_norm.weight
F32
F32
[512]
blk.7.post_ffw_norm.weight
F32
F32
[512]
blk.7.post_norm.weight
F32
F32
[512]
blk.7.proj.weight
BF16
BF16
[512, 512]
blk.8
blk.8.attn_k.weight
BF16
BF16
[512, 512]
blk.8.attn_k_norm.weight
F32
F32
[256]
blk.8.attn_norm.weight
F32
F32
[512]
blk.8.attn_output.weight
BF16
BF16
[1024, 512]
blk.8.attn_q.weight
BF16
BF16
[512, 1024]
blk.8.attn_q_norm.weight
F32
F32
[256]
blk.8.attn_v.weight
BF16
BF16
[512, 512]
blk.8.ffn_down.weight
BF16
BF16
[2048, 512]
blk.8.ffn_gate.weight
BF16
BF16
[512, 2048]
blk.8.ffn_norm.weight
F32
F32
[512]
blk.8.ffn_up.weight
BF16
BF16
[512, 2048]
blk.8.inp_gate.weight
BF16
BF16
[512, 512]
blk.8.layer_output_scale.weight
F32
F32
[1]
blk.8.post_attention_norm.weight
F32
F32
[512]
blk.8.post_ffw_norm.weight
F32
F32
[512]
blk.8.post_norm.weight
F32
F32
[512]
blk.8.proj.weight
BF16
BF16
[512, 512]
blk.9
blk.9.attn_k.weight
BF16
BF16
[512, 512]
blk.9.attn_k_norm.weight
F32
F32
[256]
blk.9.attn_norm.weight
F32
F32
[512]
blk.9.attn_output.weight
BF16
BF16
[1024, 512]
blk.9.attn_q.weight
BF16
BF16
[512, 1024]
blk.9.attn_q_norm.weight
F32
F32
[256]
blk.9.attn_v.weight
BF16
BF16
[512, 512]
blk.9.ffn_down.weight
BF16
BF16
[2048, 512]
blk.9.ffn_gate.weight
BF16
BF16
[512, 2048]
blk.9.ffn_norm.weight
F32
F32
[512]
blk.9.ffn_up.weight
BF16
BF16
[512, 2048]
blk.9.inp_gate.weight
BF16
BF16
[512, 512]
blk.9.layer_output_scale.weight
F32
F32
[1]
blk.9.post_attention_norm.weight
F32
F32
[512]
blk.9.post_ffw_norm.weight
F32
F32
[512]
blk.9.post_norm.weight
F32
F32
[512]
blk.9.proj.weight
BF16
BF16
[512, 512]
blk.10
blk.10.attn_k.weight
BF16
BF16
[512, 512]
blk.10.attn_k_norm.weight
F32
F32
[256]
blk.10.attn_norm.weight
F32
F32
[512]
blk.10.attn_output.weight
BF16
BF16
[1024, 512]
blk.10.attn_q.weight
BF16
BF16
[512, 1024]
blk.10.attn_q_norm.weight
F32
F32
[256]
blk.10.attn_v.weight
BF16
BF16
[512, 512]
blk.10.ffn_down.weight
BF16
BF16
[2048, 512]
blk.10.ffn_gate.weight
BF16
BF16
[512, 2048]
blk.10.ffn_norm.weight
F32
F32
[512]
blk.10.ffn_up.weight
BF16
BF16
[512, 2048]
blk.10.inp_gate.weight
BF16
BF16
[512, 512]
blk.10.layer_output_scale.weight
F32
F32
[1]
blk.10.post_attention_norm.weight
F32
F32
[512]
blk.10.post_ffw_norm.weight
F32
F32
[512]
blk.10.post_norm.weight
F32
F32
[512]
blk.10.proj.weight
BF16
BF16
[512, 512]
blk.11
blk.11.attn_k.weight
BF16
BF16
[512, 512]
blk.11.attn_k_norm.weight
F32
F32
[512]
blk.11.attn_norm.weight
F32
F32
[512]
blk.11.attn_output.weight
BF16
BF16
[2048, 512]
blk.11.attn_q.weight
BF16
BF16
[512, 2048]
blk.11.attn_q_norm.weight
F32
F32
[512]
blk.11.attn_v.weight
BF16
BF16
[512, 512]
blk.11.ffn_down.weight
BF16
BF16
[2048, 512]
blk.11.ffn_gate.weight
BF16
BF16
[512, 2048]
blk.11.ffn_norm.weight
F32
F32
[512]
blk.11.ffn_up.weight
BF16
BF16
[512, 2048]
blk.11.inp_gate.weight
BF16
BF16
[512, 512]
blk.11.layer_output_scale.weight
F32
F32
[1]
blk.11.post_attention_norm.weight
F32
F32
[512]
blk.11.post_ffw_norm.weight
F32
F32
[512]
blk.11.post_norm.weight
F32
F32
[512]
blk.11.proj.weight
BF16
BF16
[512, 512]
blk.12
blk.12.attn_k.weight
BF16
BF16
[512, 512]
blk.12.attn_k_norm.weight
F32
F32
[256]
blk.12.attn_norm.weight
F32
F32
[512]
blk.12.attn_output.weight
BF16
BF16
[1024, 512]
blk.12.attn_q.weight
BF16
BF16
[512, 1024]
blk.12.attn_q_norm.weight
F32
F32
[256]
blk.12.attn_v.weight
BF16
BF16
[512, 512]
blk.12.ffn_down.weight
BF16
BF16
[2048, 512]
blk.12.ffn_gate.weight
BF16
BF16
[512, 2048]
blk.12.ffn_norm.weight
F32
F32
[512]
blk.12.ffn_up.weight
BF16
BF16
[512, 2048]
blk.12.inp_gate.weight
BF16
BF16
[512, 512]
blk.12.layer_output_scale.weight
F32
F32
[1]
blk.12.post_attention_norm.weight
F32
F32
[512]
blk.12.post_ffw_norm.weight
F32
F32
[512]
blk.12.post_norm.weight
F32
F32
[512]
blk.12.proj.weight
BF16
BF16
[512, 512]
blk.13
blk.13.attn_k.weight
BF16
BF16
[512, 512]
blk.13.attn_k_norm.weight
F32
F32
[256]
blk.13.attn_norm.weight
F32
F32
[512]
blk.13.attn_output.weight
BF16
BF16
[1024, 512]
blk.13.attn_q.weight
BF16
BF16
[512, 1024]
blk.13.attn_q_norm.weight
F32
F32
[256]
blk.13.attn_v.weight
BF16
BF16
[512, 512]
blk.13.ffn_down.weight
BF16
BF16
[2048, 512]
blk.13.ffn_gate.weight
BF16
BF16
[512, 2048]
blk.13.ffn_norm.weight
F32
F32
[512]
blk.13.ffn_up.weight
BF16
BF16
[512, 2048]
blk.13.inp_gate.weight
BF16
BF16
[512, 512]
blk.13.layer_output_scale.weight
F32
F32
[1]
blk.13.post_attention_norm.weight
F32
F32
[512]
blk.13.post_ffw_norm.weight
F32
F32
[512]
blk.13.post_norm.weight
F32
F32
[512]
blk.13.proj.weight
BF16
BF16
[512, 512]
blk.14
blk.14.attn_k.weight
BF16
BF16
[512, 512]
blk.14.attn_k_norm.weight
F32
F32
[256]
blk.14.attn_norm.weight
F32
F32
[512]
blk.14.attn_output.weight
BF16
BF16
[1024, 512]
blk.14.attn_q.weight
BF16
BF16
[512, 1024]
blk.14.attn_q_norm.weight
F32
F32
[256]
blk.14.attn_v.weight
BF16
BF16
[512, 512]
blk.14.ffn_down.weight
BF16
BF16
[2048, 512]
blk.14.ffn_gate.weight
BF16
BF16
[512, 2048]
blk.14.ffn_norm.weight
F32
F32
[512]
blk.14.ffn_up.weight
BF16
BF16
[512, 2048]
blk.14.inp_gate.weight
BF16
BF16
[512, 512]
blk.14.layer_output_scale.weight
F32
F32
[1]
blk.14.post_attention_norm.weight
F32
F32
[512]
blk.14.post_ffw_norm.weight
F32
F32
[512]
blk.14.post_norm.weight
F32
F32
[512]
blk.14.proj.weight
BF16
BF16
[512, 512]
blk.15
blk.15.attn_k.weight
BF16
BF16
[512, 512]
blk.15.attn_k_norm.weight
F32
F32
[256]
blk.15.attn_norm.weight
F32
F32
[512]
blk.15.attn_output.weight
BF16
BF16
[1024, 512]
blk.15.attn_q.weight
BF16
BF16
[512, 1024]
blk.15.attn_q_norm.weight
F32
F32
[256]
blk.15.attn_v.weight
BF16
BF16
[512, 512]
blk.15.ffn_down.weight
BF16
BF16
[2048, 512]
blk.15.ffn_gate.weight
BF16
BF16
[512, 2048]
blk.15.ffn_norm.weight
F32
F32
[512]
blk.15.ffn_up.weight
BF16
BF16
[512, 2048]
blk.15.inp_gate.weight
BF16
BF16
[512, 512]
blk.15.layer_output_scale.weight
F32
F32
[1]
blk.15.post_attention_norm.weight
F32
F32
[512]
blk.15.post_ffw_norm.weight
F32
F32
[512]
blk.15.post_norm.weight
F32
F32
[512]
blk.15.proj.weight
BF16
BF16
[512, 512]
blk.16
blk.16.attn_k.weight
BF16
BF16
[512, 512]
blk.16.attn_k_norm.weight
F32
F32
[256]
blk.16.attn_norm.weight
F32
F32
[512]
blk.16.attn_output.weight
BF16
BF16
[1024, 512]
blk.16.attn_q.weight
BF16
BF16
[512, 1024]
blk.16.attn_q_norm.weight
F32
F32
[256]
blk.16.attn_v.weight
BF16
BF16
[512, 512]
blk.16.ffn_down.weight
BF16
BF16
[2048, 512]
blk.16.ffn_gate.weight
BF16
BF16
[512, 2048]
blk.16.ffn_norm.weight
F32
F32
[512]
blk.16.ffn_up.weight
BF16
BF16
[512, 2048]
blk.16.inp_gate.weight
BF16
BF16
[512, 512]
blk.16.layer_output_scale.weight
F32
F32
[1]
blk.16.post_attention_norm.weight
F32
F32
[512]
blk.16.post_ffw_norm.weight
F32
F32
[512]
blk.16.post_norm.weight
F32
F32
[512]
blk.16.proj.weight
BF16
BF16
[512, 512]
blk.17
blk.17.attn_k.weight
BF16
BF16
[512, 512]
blk.17.attn_k_norm.weight
F32
F32
[512]
blk.17.attn_norm.weight
F32
F32
[512]
blk.17.attn_output.weight
BF16
BF16
[2048, 512]
blk.17.attn_q.weight
BF16
BF16
[512, 2048]
blk.17.attn_q_norm.weight
F32
F32
[512]
blk.17.attn_v.weight
BF16
BF16
[512, 512]
blk.17.ffn_down.weight
BF16
BF16
[2048, 512]
blk.17.ffn_gate.weight
BF16
BF16
[512, 2048]
blk.17.ffn_norm.weight
F32
F32
[512]
blk.17.ffn_up.weight
BF16
BF16
[512, 2048]
blk.17.inp_gate.weight
BF16
BF16
[512, 512]
blk.17.layer_output_scale.weight
F32
F32
[1]
blk.17.post_attention_norm.weight
F32
F32
[512]
blk.17.post_ffw_norm.weight
F32
F32
[512]
blk.17.post_norm.weight
F32
F32
[512]
blk.17.proj.weight
BF16
BF16
[512, 512]
blk.18
blk.18.attn_k.weight
BF16
BF16
[512, 512]
blk.18.attn_k_norm.weight
F32
F32
[256]
blk.18.attn_norm.weight
F32
F32
[512]
blk.18.attn_output.weight
BF16
BF16
[1024, 512]
blk.18.attn_q.weight
BF16
BF16
[512, 1024]
blk.18.attn_q_norm.weight
F32
F32
[256]
blk.18.attn_v.weight
BF16
BF16
[512, 512]
blk.18.ffn_down.weight
BF16
BF16
[2048, 512]
blk.18.ffn_gate.weight
BF16
BF16
[512, 2048]
blk.18.ffn_norm.weight
F32
F32
[512]
blk.18.ffn_up.weight
BF16
BF16
[512, 2048]
blk.18.inp_gate.weight
BF16
BF16
[512, 512]
blk.18.layer_output_scale.weight
F32
F32
[1]
blk.18.post_attention_norm.weight
F32
F32
[512]
blk.18.post_ffw_norm.weight
F32
F32
[512]
blk.18.post_norm.weight
F32
F32
[512]
blk.18.proj.weight
BF16
BF16
[512, 512]
blk.19
blk.19.attn_k.weight
BF16
BF16
[512, 512]
blk.19.attn_k_norm.weight
F32
F32
[256]
blk.19.attn_norm.weight
F32
F32
[512]
blk.19.attn_output.weight
BF16
BF16
[1024, 512]
blk.19.attn_q.weight
BF16
BF16
[512, 1024]
blk.19.attn_q_norm.weight
F32
F32
[256]
blk.19.attn_v.weight
BF16
BF16
[512, 512]
blk.19.ffn_down.weight
BF16
BF16
[2048, 512]
blk.19.ffn_gate.weight
BF16
BF16
[512, 2048]
blk.19.ffn_norm.weight
F32
F32
[512]
blk.19.ffn_up.weight
BF16
BF16
[512, 2048]
blk.19.inp_gate.weight
BF16
BF16
[512, 512]
blk.19.layer_output_scale.weight
F32
F32
[1]
blk.19.post_attention_norm.weight
F32
F32
[512]
blk.19.post_ffw_norm.weight
F32
F32
[512]
blk.19.post_norm.weight
F32
F32
[512]
blk.19.proj.weight
BF16
BF16
[512, 512]
blk.20
blk.20.attn_k.weight
BF16
BF16
[512, 512]
blk.20.attn_k_norm.weight
F32
F32
[256]
blk.20.attn_norm.weight
F32
F32
[512]
blk.20.attn_output.weight
BF16
BF16
[1024, 512]
blk.20.attn_q.weight
BF16
BF16
[512, 1024]
blk.20.attn_q_norm.weight
F32
F32
[256]
blk.20.attn_v.weight
BF16
BF16
[512, 512]
blk.20.ffn_down.weight
BF16
BF16
[2048, 512]
blk.20.ffn_gate.weight
BF16
BF16
[512, 2048]
blk.20.ffn_norm.weight
F32
F32
[512]
blk.20.ffn_up.weight
BF16
BF16
[512, 2048]
blk.20.inp_gate.weight
BF16
BF16
[512, 512]
blk.20.layer_output_scale.weight
F32
F32
[1]
blk.20.post_attention_norm.weight
F32
F32
[512]
blk.20.post_ffw_norm.weight
F32
F32
[512]
blk.20.post_norm.weight
F32
F32
[512]
blk.20.proj.weight
BF16
BF16
[512, 512]
blk.21
blk.21.attn_k.weight
BF16
BF16
[512, 512]
blk.21.attn_k_norm.weight
F32
F32
[256]
blk.21.attn_norm.weight
F32
F32
[512]
blk.21.attn_output.weight
BF16
BF16
[1024, 512]
blk.21.attn_q.weight
BF16
BF16
[512, 1024]
blk.21.attn_q_norm.weight
F32
F32
[256]
blk.21.attn_v.weight
BF16
BF16
[512, 512]
blk.21.ffn_down.weight
BF16
BF16
[2048, 512]
blk.21.ffn_gate.weight
BF16
BF16
[512, 2048]
blk.21.ffn_norm.weight
F32
F32
[512]
blk.21.ffn_up.weight
BF16
BF16
[512, 2048]
blk.21.inp_gate.weight
BF16
BF16
[512, 512]
blk.21.layer_output_scale.weight
F32
F32
[1]
blk.21.post_attention_norm.weight
F32
F32
[512]
blk.21.post_ffw_norm.weight
F32
F32
[512]
blk.21.post_norm.weight
F32
F32
[512]
blk.21.proj.weight
BF16
BF16
[512, 512]
blk.22
blk.22.attn_k.weight
BF16
BF16
[512, 512]
blk.22.attn_k_norm.weight
F32
F32
[256]
blk.22.attn_norm.weight
F32
F32
[512]
blk.22.attn_output.weight
BF16
BF16
[1024, 512]
blk.22.attn_q.weight
BF16
BF16
[512, 1024]
blk.22.attn_q_norm.weight
F32
F32
[256]
blk.22.attn_v.weight
BF16
BF16
[512, 512]
blk.22.ffn_down.weight
BF16
BF16
[2048, 512]
blk.22.ffn_gate.weight
BF16
BF16
[512, 2048]
blk.22.ffn_norm.weight
F32
F32
[512]
blk.22.ffn_up.weight
BF16
BF16
[512, 2048]
blk.22.inp_gate.weight
BF16
BF16
[512, 512]
blk.22.layer_output_scale.weight
F32
F32
[1]
blk.22.post_attention_norm.weight
F32
F32
[512]
blk.22.post_ffw_norm.weight
F32
F32
[512]
blk.22.post_norm.weight
F32
F32
[512]
blk.22.proj.weight
BF16
BF16
[512, 512]
blk.23
blk.23.attn_k.weight
BF16
BF16
[512, 512]
blk.23.attn_k_norm.weight
F32
F32
[512]
blk.23.attn_norm.weight
F32
F32
[512]
blk.23.attn_output.weight
BF16
BF16
[2048, 512]
blk.23.attn_q.weight
BF16
BF16
[512, 2048]
blk.23.attn_q_norm.weight
F32
F32
[512]
blk.23.attn_v.weight
BF16
BF16
[512, 512]
blk.23.ffn_down.weight
BF16
BF16
[2048, 512]
blk.23.ffn_gate.weight
BF16
BF16
[512, 2048]
blk.23.ffn_norm.weight
F32
F32
[512]
blk.23.ffn_up.weight
BF16
BF16
[512, 2048]
blk.23.inp_gate.weight
BF16
BF16
[512, 512]
blk.23.layer_output_scale.weight
F32
F32
[1]
blk.23.post_attention_norm.weight
F32
F32
[512]
blk.23.post_ffw_norm.weight
F32
F32
[512]
blk.23.post_norm.weight
F32
F32
[512]
blk.23.proj.weight
BF16
BF16
[512, 512]
output.weight
BF16
BF16
[512, 768]
per_layer_model_proj.weight
BF16
BF16
[512, 12288]
per_layer_proj_norm.weight
F32
F32
[512]
output_norm.weight
F32
F32
[512]