Models
Docs
Pricing
Sign in
Download
Models
Download
Docs
Pricing
Sign in
gemma4
:latest
63
Downloads
Updated
34 minutes ago
Gemma 4 models are designed to deliver frontier-level performance at each size. They are well-suited for reasoning, agentic workflows, coding, and multimodal understanding.
Gemma 4 models are designed to deliver frontier-level performance at each size. They are well-suited for reasoning, agentic workflows, coding, and multimodal understanding.
Cancel
vision
e2b
e4b
26b
31b
gemma4:latest
...
/
model
4c27e0f5b5ad · 9.6GB
Metadata
general.architecture
gemma4
gemma4
general.file_type
Q4_K_M
Q4_K_M
gemma4.attention.head_count
8
8
gemma4.attention.head_count_kv
2
2
gemma4.attention.key_length
512
512
gemma4.attention.key_length_swa
256
256
gemma4.attention.layer_norm_rms_epsilon
1e-06
1e-06
gemma4.attention.shared_kv_layers
18
18
gemma4.attention.sliding_window
512
512
gemma4.attention.sliding_window_pattern
[true, true, true, true, true, ...]
[true, true, true, true, true, ...]
gemma4.attention.value_length
512
512
gemma4.attention.value_length_swa
256
256
gemma4.audio.attention.head_count
8
8
gemma4.audio.attention.layer_norm_epsilon
1e-06
1e-06
gemma4.audio.block_count
12
12
gemma4.audio.conv_kernel_size
5
5
gemma4.audio.embedding_length
1024
1024
gemma4.audio.feed_forward_length
4096
4096
gemma4.block_count
42
42
gemma4.context_length
131072
131072
gemma4.embedding_length
2560
2560
gemma4.embedding_length_per_layer_input
256
256
gemma4.feed_forward_length
10240
10240
gemma4.final_logit_softcapping
30
30
gemma4.rope.dimension_count
512
512
gemma4.rope.dimension_count_swa
256
256
gemma4.rope.freq_base
1e+06
1e+06
gemma4.rope.freq_base_swa
10000
10000
gemma4.vision.attention.head_count
12
12
gemma4.vision.attention.layer_norm_epsilon
1e-06
1e-06
gemma4.vision.block_count
16
16
gemma4.vision.embedding_length
768
768
gemma4.vision.feed_forward_length
3072
3072
gemma4.vision.num_channels
3
3
gemma4.vision.patch_size
16
16
gemma4.vision.projector.scale_factor
3
3
tokenizer.ggml.add_bos_token
false
false
tokenizer.ggml.add_eos_token
false
false
tokenizer.ggml.add_mask_token
false
false
tokenizer.ggml.add_padding_token
false
false
tokenizer.ggml.add_unknown_token
false
false
tokenizer.ggml.bos_token_id
2
2
tokenizer.ggml.eos_token_id
1
1
tokenizer.ggml.eos_token_ids
[1, 106, 50]
[1, 106, 50]
tokenizer.ggml.mask_token_id
4
4
tokenizer.ggml.merges
[ , ▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁ ▁, , , ▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁ ▁▁, ...]
[ , ▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁ ▁, , , ▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁ ▁▁, ...]
tokenizer.ggml.model
llama
llama
tokenizer.ggml.padding_token_id
0
0
tokenizer.ggml.pre
gemma4
gemma4
tokenizer.ggml.scores
[0, 1, 2, 3, 4, ...]
[0, 1, 2, 3, 4, ...]
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
Q6_K
Q6_K
[2560, 262144]
a.blk.0
a.blk.0.attn_k.input_max
F32
F32
[1]
a.blk.0.attn_k.input_min
F32
F32
[1]
a.blk.0.attn_k.output_max
F32
F32
[1]
a.blk.0.attn_k.output_min
F32
F32
[1]
a.blk.0.attn_k.weight
BF16
BF16
[1024, 1024]
a.blk.0.attn_out.input_max
F32
F32
[1]
a.blk.0.attn_out.input_min
F32
F32
[1]
a.blk.0.attn_out.output_max
F32
F32
[1]
a.blk.0.attn_out.output_min
F32
F32
[1]
a.blk.0.attn_out.weight
BF16
BF16
[1024, 1024]
a.blk.0.attn_q.input_max
F32
F32
[1]
a.blk.0.attn_q.input_min
F32
F32
[1]
a.blk.0.attn_q.output_max
F32
F32
[1]
a.blk.0.attn_q.output_min
F32
F32
[1]
a.blk.0.attn_q.weight
BF16
BF16
[1024, 1024]
a.blk.0.attn_v.input_max
F32
F32
[1]
a.blk.0.attn_v.input_min
F32
F32
[1]
a.blk.0.attn_v.output_max
F32
F32
[1]
a.blk.0.attn_v.output_min
F32
F32
[1]
a.blk.0.attn_v.weight
BF16
BF16
[1024, 1024]
a.blk.0.conv_dw.weight
F32
F32
[5, 1024]
a.blk.0.conv_norm.weight
F32
F32
[1024]
a.blk.0.conv_pw1.input_max
F32
F32
[1]
a.blk.0.conv_pw1.input_min
F32
F32
[1]
a.blk.0.conv_pw1.output_max
F32
F32
[1]
a.blk.0.conv_pw1.output_min
F32
F32
[1]
a.blk.0.conv_pw1.weight
BF16
BF16
[1024, 2048]
a.blk.0.conv_pw2.input_max
F32
F32
[1]
a.blk.0.conv_pw2.input_min
F32
F32
[1]
a.blk.0.conv_pw2.output_max
F32
F32
[1]
a.blk.0.conv_pw2.output_min
F32
F32
[1]
a.blk.0.conv_pw2.weight
BF16
BF16
[1024, 1024]
a.blk.0.ffn_down.input_max
F32
F32
[1]
a.blk.0.ffn_down.input_min
F32
F32
[1]
a.blk.0.ffn_down.output_max
F32
F32
[1]
a.blk.0.ffn_down.output_min
F32
F32
[1]
a.blk.0.ffn_down.weight
BF16
BF16
[4096, 1024]
a.blk.0.ffn_down_1.input_max
F32
F32
[1]
a.blk.0.ffn_down_1.input_min
F32
F32
[1]
a.blk.0.ffn_down_1.output_max
F32
F32
[1]
a.blk.0.ffn_down_1.output_min
F32
F32
[1]
a.blk.0.ffn_down_1.weight
BF16
BF16
[4096, 1024]
a.blk.0.ffn_norm.weight
F32
F32
[1024]
a.blk.0.ffn_norm_1.weight
F32
F32
[1024]
a.blk.0.ffn_post_norm.weight
F32
F32
[1024]
a.blk.0.ffn_post_norm_1.weight
F32
F32
[1024]
a.blk.0.ffn_up.input_max
F32
F32
[1]
a.blk.0.ffn_up.input_min
F32
F32
[1]
a.blk.0.ffn_up.output_max
F32
F32
[1]
a.blk.0.ffn_up.output_min
F32
F32
[1]
a.blk.0.ffn_up.weight
BF16
BF16
[1024, 4096]
a.blk.0.ffn_up_1.input_max
F32
F32
[1]
a.blk.0.ffn_up_1.input_min
F32
F32
[1]
a.blk.0.ffn_up_1.output_max
F32
F32
[1]
a.blk.0.ffn_up_1.output_min
F32
F32
[1]
a.blk.0.ffn_up_1.weight
BF16
BF16
[1024, 4096]
a.blk.0.layer_pre_norm.weight
F32
F32
[1024]
a.blk.0.linear_pos.weight
BF16
BF16
[1024, 1024]
a.blk.0.ln1.weight
F32
F32
[1024]
a.blk.0.ln2.weight
F32
F32
[1024]
a.blk.0.norm_conv.weight
F32
F32
[1024]
a.blk.0.per_dim_scale.weight
F32
F32
[128]
a.blk.1
a.blk.1.attn_k.input_max
F32
F32
[1]
a.blk.1.attn_k.input_min
F32
F32
[1]
a.blk.1.attn_k.output_max
F32
F32
[1]
a.blk.1.attn_k.output_min
F32
F32
[1]
a.blk.1.attn_k.weight
BF16
BF16
[1024, 1024]
a.blk.1.attn_out.input_max
F32
F32
[1]
a.blk.1.attn_out.input_min
F32
F32
[1]
a.blk.1.attn_out.output_max
F32
F32
[1]
a.blk.1.attn_out.output_min
F32
F32
[1]
a.blk.1.attn_out.weight
BF16
BF16
[1024, 1024]
a.blk.1.attn_q.input_max
F32
F32
[1]
a.blk.1.attn_q.input_min
F32
F32
[1]
a.blk.1.attn_q.output_max
F32
F32
[1]
a.blk.1.attn_q.output_min
F32
F32
[1]
a.blk.1.attn_q.weight
BF16
BF16
[1024, 1024]
a.blk.1.attn_v.input_max
F32
F32
[1]
a.blk.1.attn_v.input_min
F32
F32
[1]
a.blk.1.attn_v.output_max
F32
F32
[1]
a.blk.1.attn_v.output_min
F32
F32
[1]
a.blk.1.attn_v.weight
BF16
BF16
[1024, 1024]
a.blk.1.conv_dw.weight
F32
F32
[5, 1024]
a.blk.1.conv_norm.weight
F32
F32
[1024]
a.blk.1.conv_pw1.input_max
F32
F32
[1]
a.blk.1.conv_pw1.input_min
F32
F32
[1]
a.blk.1.conv_pw1.output_max
F32
F32
[1]
a.blk.1.conv_pw1.output_min
F32
F32
[1]
a.blk.1.conv_pw1.weight
BF16
BF16
[1024, 2048]
a.blk.1.conv_pw2.input_max
F32
F32
[1]
a.blk.1.conv_pw2.input_min
F32
F32
[1]
a.blk.1.conv_pw2.output_max
F32
F32
[1]
a.blk.1.conv_pw2.output_min
F32
F32
[1]
a.blk.1.conv_pw2.weight
BF16
BF16
[1024, 1024]
a.blk.1.ffn_down.input_max
F32
F32
[1]
a.blk.1.ffn_down.input_min
F32
F32
[1]
a.blk.1.ffn_down.output_max
F32
F32
[1]
a.blk.1.ffn_down.output_min
F32
F32
[1]
a.blk.1.ffn_down.weight
BF16
BF16
[4096, 1024]
a.blk.1.ffn_down_1.input_max
F32
F32
[1]
a.blk.1.ffn_down_1.input_min
F32
F32
[1]
a.blk.1.ffn_down_1.output_max
F32
F32
[1]
a.blk.1.ffn_down_1.output_min
F32
F32
[1]
a.blk.1.ffn_down_1.weight
BF16
BF16
[4096, 1024]
a.blk.1.ffn_norm.weight
F32
F32
[1024]
a.blk.1.ffn_norm_1.weight
F32
F32
[1024]
a.blk.1.ffn_post_norm.weight
F32
F32
[1024]
a.blk.1.ffn_post_norm_1.weight
F32
F32
[1024]
a.blk.1.ffn_up.input_max
F32
F32
[1]
a.blk.1.ffn_up.input_min
F32
F32
[1]
a.blk.1.ffn_up.output_max
F32
F32
[1]
a.blk.1.ffn_up.output_min
F32
F32
[1]
a.blk.1.ffn_up.weight
BF16
BF16
[1024, 4096]
a.blk.1.ffn_up_1.input_max
F32
F32
[1]
a.blk.1.ffn_up_1.input_min
F32
F32
[1]
a.blk.1.ffn_up_1.output_max
F32
F32
[1]
a.blk.1.ffn_up_1.output_min
F32
F32
[1]
a.blk.1.ffn_up_1.weight
BF16
BF16
[1024, 4096]
a.blk.1.layer_pre_norm.weight
F32
F32
[1024]
a.blk.1.linear_pos.weight
BF16
BF16
[1024, 1024]
a.blk.1.ln1.weight
F32
F32
[1024]
a.blk.1.ln2.weight
F32
F32
[1024]
a.blk.1.norm_conv.weight
F32
F32
[1024]
a.blk.1.per_dim_scale.weight
F32
F32
[128]
a.blk.2
a.blk.2.attn_k.input_max
F32
F32
[1]
a.blk.2.attn_k.input_min
F32
F32
[1]
a.blk.2.attn_k.output_max
F32
F32
[1]
a.blk.2.attn_k.output_min
F32
F32
[1]
a.blk.2.attn_k.weight
BF16
BF16
[1024, 1024]
a.blk.2.attn_out.input_max
F32
F32
[1]
a.blk.2.attn_out.input_min
F32
F32
[1]
a.blk.2.attn_out.output_max
F32
F32
[1]
a.blk.2.attn_out.output_min
F32
F32
[1]
a.blk.2.attn_out.weight
BF16
BF16
[1024, 1024]
a.blk.2.attn_q.input_max
F32
F32
[1]
a.blk.2.attn_q.input_min
F32
F32
[1]
a.blk.2.attn_q.output_max
F32
F32
[1]
a.blk.2.attn_q.output_min
F32
F32
[1]
a.blk.2.attn_q.weight
BF16
BF16
[1024, 1024]
a.blk.2.attn_v.input_max
F32
F32
[1]
a.blk.2.attn_v.input_min
F32
F32
[1]
a.blk.2.attn_v.output_max
F32
F32
[1]
a.blk.2.attn_v.output_min
F32
F32
[1]
a.blk.2.attn_v.weight
BF16
BF16
[1024, 1024]
a.blk.2.conv_dw.weight
F32
F32
[5, 1024]
a.blk.2.conv_norm.weight
F32
F32
[1024]
a.blk.2.conv_pw1.input_max
F32
F32
[1]
a.blk.2.conv_pw1.input_min
F32
F32
[1]
a.blk.2.conv_pw1.output_max
F32
F32
[1]
a.blk.2.conv_pw1.output_min
F32
F32
[1]
a.blk.2.conv_pw1.weight
BF16
BF16
[1024, 2048]
a.blk.2.conv_pw2.input_max
F32
F32
[1]
a.blk.2.conv_pw2.input_min
F32
F32
[1]
a.blk.2.conv_pw2.output_max
F32
F32
[1]
a.blk.2.conv_pw2.output_min
F32
F32
[1]
a.blk.2.conv_pw2.weight
BF16
BF16
[1024, 1024]
a.blk.2.ffn_down.input_max
F32
F32
[1]
a.blk.2.ffn_down.input_min
F32
F32
[1]
a.blk.2.ffn_down.output_max
F32
F32
[1]
a.blk.2.ffn_down.output_min
F32
F32
[1]
a.blk.2.ffn_down.weight
BF16
BF16
[4096, 1024]
a.blk.2.ffn_down_1.input_max
F32
F32
[1]
a.blk.2.ffn_down_1.input_min
F32
F32
[1]
a.blk.2.ffn_down_1.output_max
F32
F32
[1]
a.blk.2.ffn_down_1.output_min
F32
F32
[1]
a.blk.2.ffn_down_1.weight
BF16
BF16
[4096, 1024]
a.blk.2.ffn_norm.weight
F32
F32
[1024]
a.blk.2.ffn_norm_1.weight
F32
F32
[1024]
a.blk.2.ffn_post_norm.weight
F32
F32
[1024]
a.blk.2.ffn_post_norm_1.weight
F32
F32
[1024]
a.blk.2.ffn_up.input_max
F32
F32
[1]
a.blk.2.ffn_up.input_min
F32
F32
[1]
a.blk.2.ffn_up.output_max
F32
F32
[1]
a.blk.2.ffn_up.output_min
F32
F32
[1]
a.blk.2.ffn_up.weight
BF16
BF16
[1024, 4096]
a.blk.2.ffn_up_1.input_max
F32
F32
[1]
a.blk.2.ffn_up_1.input_min
F32
F32
[1]
a.blk.2.ffn_up_1.output_max
F32
F32
[1]
a.blk.2.ffn_up_1.output_min
F32
F32
[1]
a.blk.2.ffn_up_1.weight
BF16
BF16
[1024, 4096]
a.blk.2.layer_pre_norm.weight
F32
F32
[1024]
a.blk.2.linear_pos.weight
BF16
BF16
[1024, 1024]
a.blk.2.ln1.weight
F32
F32
[1024]
a.blk.2.ln2.weight
F32
F32
[1024]
a.blk.2.norm_conv.weight
F32
F32
[1024]
a.blk.2.per_dim_scale.weight
F32
F32
[128]
a.blk.3
a.blk.3.attn_k.input_max
F32
F32
[1]
a.blk.3.attn_k.input_min
F32
F32
[1]
a.blk.3.attn_k.output_max
F32
F32
[1]
a.blk.3.attn_k.output_min
F32
F32
[1]
a.blk.3.attn_k.weight
BF16
BF16
[1024, 1024]
a.blk.3.attn_out.input_max
F32
F32
[1]
a.blk.3.attn_out.input_min
F32
F32
[1]
a.blk.3.attn_out.output_max
F32
F32
[1]
a.blk.3.attn_out.output_min
F32
F32
[1]
a.blk.3.attn_out.weight
BF16
BF16
[1024, 1024]
a.blk.3.attn_q.input_max
F32
F32
[1]
a.blk.3.attn_q.input_min
F32
F32
[1]
a.blk.3.attn_q.output_max
F32
F32
[1]
a.blk.3.attn_q.output_min
F32
F32
[1]
a.blk.3.attn_q.weight
BF16
BF16
[1024, 1024]
a.blk.3.attn_v.input_max
F32
F32
[1]
a.blk.3.attn_v.input_min
F32
F32
[1]
a.blk.3.attn_v.output_max
F32
F32
[1]
a.blk.3.attn_v.output_min
F32
F32
[1]
a.blk.3.attn_v.weight
BF16
BF16
[1024, 1024]
a.blk.3.conv_dw.weight
F32
F32
[5, 1024]
a.blk.3.conv_norm.weight
F32
F32
[1024]
a.blk.3.conv_pw1.input_max
F32
F32
[1]
a.blk.3.conv_pw1.input_min
F32
F32
[1]
a.blk.3.conv_pw1.output_max
F32
F32
[1]
a.blk.3.conv_pw1.output_min
F32
F32
[1]
a.blk.3.conv_pw1.weight
BF16
BF16
[1024, 2048]
a.blk.3.conv_pw2.input_max
F32
F32
[1]
a.blk.3.conv_pw2.input_min
F32
F32
[1]
a.blk.3.conv_pw2.output_max
F32
F32
[1]
a.blk.3.conv_pw2.output_min
F32
F32
[1]
a.blk.3.conv_pw2.weight
BF16
BF16
[1024, 1024]
a.blk.3.ffn_down.input_max
F32
F32
[1]
a.blk.3.ffn_down.input_min
F32
F32
[1]
a.blk.3.ffn_down.output_max
F32
F32
[1]
a.blk.3.ffn_down.output_min
F32
F32
[1]
a.blk.3.ffn_down.weight
BF16
BF16
[4096, 1024]
a.blk.3.ffn_down_1.input_max
F32
F32
[1]
a.blk.3.ffn_down_1.input_min
F32
F32
[1]
a.blk.3.ffn_down_1.output_max
F32
F32
[1]
a.blk.3.ffn_down_1.output_min
F32
F32
[1]
a.blk.3.ffn_down_1.weight
BF16
BF16
[4096, 1024]
a.blk.3.ffn_norm.weight
F32
F32
[1024]
a.blk.3.ffn_norm_1.weight
F32
F32
[1024]
a.blk.3.ffn_post_norm.weight
F32
F32
[1024]
a.blk.3.ffn_post_norm_1.weight
F32
F32
[1024]
a.blk.3.ffn_up.input_max
F32
F32
[1]
a.blk.3.ffn_up.input_min
F32
F32
[1]
a.blk.3.ffn_up.output_max
F32
F32
[1]
a.blk.3.ffn_up.output_min
F32
F32
[1]
a.blk.3.ffn_up.weight
BF16
BF16
[1024, 4096]
a.blk.3.ffn_up_1.input_max
F32
F32
[1]
a.blk.3.ffn_up_1.input_min
F32
F32
[1]
a.blk.3.ffn_up_1.output_max
F32
F32
[1]
a.blk.3.ffn_up_1.output_min
F32
F32
[1]
a.blk.3.ffn_up_1.weight
BF16
BF16
[1024, 4096]
a.blk.3.layer_pre_norm.weight
F32
F32
[1024]
a.blk.3.linear_pos.weight
BF16
BF16
[1024, 1024]
a.blk.3.ln1.weight
F32
F32
[1024]
a.blk.3.ln2.weight
F32
F32
[1024]
a.blk.3.norm_conv.weight
F32
F32
[1024]
a.blk.3.per_dim_scale.weight
F32
F32
[128]
a.blk.4
a.blk.4.attn_k.input_max
F32
F32
[1]
a.blk.4.attn_k.input_min
F32
F32
[1]
a.blk.4.attn_k.output_max
F32
F32
[1]
a.blk.4.attn_k.output_min
F32
F32
[1]
a.blk.4.attn_k.weight
BF16
BF16
[1024, 1024]
a.blk.4.attn_out.input_max
F32
F32
[1]
a.blk.4.attn_out.input_min
F32
F32
[1]
a.blk.4.attn_out.output_max
F32
F32
[1]
a.blk.4.attn_out.output_min
F32
F32
[1]
a.blk.4.attn_out.weight
BF16
BF16
[1024, 1024]
a.blk.4.attn_q.input_max
F32
F32
[1]
a.blk.4.attn_q.input_min
F32
F32
[1]
a.blk.4.attn_q.output_max
F32
F32
[1]
a.blk.4.attn_q.output_min
F32
F32
[1]
a.blk.4.attn_q.weight
BF16
BF16
[1024, 1024]
a.blk.4.attn_v.input_max
F32
F32
[1]
a.blk.4.attn_v.input_min
F32
F32
[1]
a.blk.4.attn_v.output_max
F32
F32
[1]
a.blk.4.attn_v.output_min
F32
F32
[1]
a.blk.4.attn_v.weight
BF16
BF16
[1024, 1024]
a.blk.4.conv_dw.weight
F32
F32
[5, 1024]
a.blk.4.conv_norm.weight
F32
F32
[1024]
a.blk.4.conv_pw1.input_max
F32
F32
[1]
a.blk.4.conv_pw1.input_min
F32
F32
[1]
a.blk.4.conv_pw1.output_max
F32
F32
[1]
a.blk.4.conv_pw1.output_min
F32
F32
[1]
a.blk.4.conv_pw1.weight
BF16
BF16
[1024, 2048]
a.blk.4.conv_pw2.input_max
F32
F32
[1]
a.blk.4.conv_pw2.input_min
F32
F32
[1]
a.blk.4.conv_pw2.output_max
F32
F32
[1]
a.blk.4.conv_pw2.output_min
F32
F32
[1]
a.blk.4.conv_pw2.weight
BF16
BF16
[1024, 1024]
a.blk.4.ffn_down.input_max
F32
F32
[1]
a.blk.4.ffn_down.input_min
F32
F32
[1]
a.blk.4.ffn_down.output_max
F32
F32
[1]
a.blk.4.ffn_down.output_min
F32
F32
[1]
a.blk.4.ffn_down.weight
BF16
BF16
[4096, 1024]
a.blk.4.ffn_down_1.input_max
F32
F32
[1]
a.blk.4.ffn_down_1.input_min
F32
F32
[1]
a.blk.4.ffn_down_1.output_max
F32
F32
[1]
a.blk.4.ffn_down_1.output_min
F32
F32
[1]
a.blk.4.ffn_down_1.weight
BF16
BF16
[4096, 1024]
a.blk.4.ffn_norm.weight
F32
F32
[1024]
a.blk.4.ffn_norm_1.weight
F32
F32
[1024]
a.blk.4.ffn_post_norm.weight
F32
F32
[1024]
a.blk.4.ffn_post_norm_1.weight
F32
F32
[1024]
a.blk.4.ffn_up.input_max
F32
F32
[1]
a.blk.4.ffn_up.input_min
F32
F32
[1]
a.blk.4.ffn_up.output_max
F32
F32
[1]
a.blk.4.ffn_up.output_min
F32
F32
[1]
a.blk.4.ffn_up.weight
BF16
BF16
[1024, 4096]
a.blk.4.ffn_up_1.input_max
F32
F32
[1]
a.blk.4.ffn_up_1.input_min
F32
F32
[1]
a.blk.4.ffn_up_1.output_max
F32
F32
[1]
a.blk.4.ffn_up_1.output_min
F32
F32
[1]
a.blk.4.ffn_up_1.weight
BF16
BF16
[1024, 4096]
a.blk.4.layer_pre_norm.weight
F32
F32
[1024]
a.blk.4.linear_pos.weight
BF16
BF16
[1024, 1024]
a.blk.4.ln1.weight
F32
F32
[1024]
a.blk.4.ln2.weight
F32
F32
[1024]
a.blk.4.norm_conv.weight
F32
F32
[1024]
a.blk.4.per_dim_scale.weight
F32
F32
[128]
a.blk.5
a.blk.5.attn_k.input_max
F32
F32
[1]
a.blk.5.attn_k.input_min
F32
F32
[1]
a.blk.5.attn_k.output_max
F32
F32
[1]
a.blk.5.attn_k.output_min
F32
F32
[1]
a.blk.5.attn_k.weight
BF16
BF16
[1024, 1024]
a.blk.5.attn_out.input_max
F32
F32
[1]
a.blk.5.attn_out.input_min
F32
F32
[1]
a.blk.5.attn_out.output_max
F32
F32
[1]
a.blk.5.attn_out.output_min
F32
F32
[1]
a.blk.5.attn_out.weight
BF16
BF16
[1024, 1024]
a.blk.5.attn_q.input_max
F32
F32
[1]
a.blk.5.attn_q.input_min
F32
F32
[1]
a.blk.5.attn_q.output_max
F32
F32
[1]
a.blk.5.attn_q.output_min
F32
F32
[1]
a.blk.5.attn_q.weight
BF16
BF16
[1024, 1024]
a.blk.5.attn_v.input_max
F32
F32
[1]
a.blk.5.attn_v.input_min
F32
F32
[1]
a.blk.5.attn_v.output_max
F32
F32
[1]
a.blk.5.attn_v.output_min
F32
F32
[1]
a.blk.5.attn_v.weight
BF16
BF16
[1024, 1024]
a.blk.5.conv_dw.weight
F32
F32
[5, 1024]
a.blk.5.conv_norm.weight
F32
F32
[1024]
a.blk.5.conv_pw1.input_max
F32
F32
[1]
a.blk.5.conv_pw1.input_min
F32
F32
[1]
a.blk.5.conv_pw1.output_max
F32
F32
[1]
a.blk.5.conv_pw1.output_min
F32
F32
[1]
a.blk.5.conv_pw1.weight
BF16
BF16
[1024, 2048]
a.blk.5.conv_pw2.input_max
F32
F32
[1]
a.blk.5.conv_pw2.input_min
F32
F32
[1]
a.blk.5.conv_pw2.output_max
F32
F32
[1]
a.blk.5.conv_pw2.output_min
F32
F32
[1]
a.blk.5.conv_pw2.weight
BF16
BF16
[1024, 1024]
a.blk.5.ffn_down.input_max
F32
F32
[1]
a.blk.5.ffn_down.input_min
F32
F32
[1]
a.blk.5.ffn_down.output_max
F32
F32
[1]
a.blk.5.ffn_down.output_min
F32
F32
[1]
a.blk.5.ffn_down.weight
BF16
BF16
[4096, 1024]
a.blk.5.ffn_down_1.input_max
F32
F32
[1]
a.blk.5.ffn_down_1.input_min
F32
F32
[1]
a.blk.5.ffn_down_1.output_max
F32
F32
[1]
a.blk.5.ffn_down_1.output_min
F32
F32
[1]
a.blk.5.ffn_down_1.weight
BF16
BF16
[4096, 1024]
a.blk.5.ffn_norm.weight
F32
F32
[1024]
a.blk.5.ffn_norm_1.weight
F32
F32
[1024]
a.blk.5.ffn_post_norm.weight
F32
F32
[1024]
a.blk.5.ffn_post_norm_1.weight
F32
F32
[1024]
a.blk.5.ffn_up.input_max
F32
F32
[1]
a.blk.5.ffn_up.input_min
F32
F32
[1]
a.blk.5.ffn_up.output_max
F32
F32
[1]
a.blk.5.ffn_up.output_min
F32
F32
[1]
a.blk.5.ffn_up.weight
BF16
BF16
[1024, 4096]
a.blk.5.ffn_up_1.input_max
F32
F32
[1]
a.blk.5.ffn_up_1.input_min
F32
F32
[1]
a.blk.5.ffn_up_1.output_max
F32
F32
[1]
a.blk.5.ffn_up_1.output_min
F32
F32
[1]
a.blk.5.ffn_up_1.weight
BF16
BF16
[1024, 4096]
a.blk.5.layer_pre_norm.weight
F32
F32
[1024]
a.blk.5.linear_pos.weight
BF16
BF16
[1024, 1024]
a.blk.5.ln1.weight
F32
F32
[1024]
a.blk.5.ln2.weight
F32
F32
[1024]
a.blk.5.norm_conv.weight
F32
F32
[1024]
a.blk.5.per_dim_scale.weight
F32
F32
[128]
a.blk.6
a.blk.6.attn_k.input_max
F32
F32
[1]
a.blk.6.attn_k.input_min
F32
F32
[1]
a.blk.6.attn_k.output_max
F32
F32
[1]
a.blk.6.attn_k.output_min
F32
F32
[1]
a.blk.6.attn_k.weight
BF16
BF16
[1024, 1024]
a.blk.6.attn_out.input_max
F32
F32
[1]
a.blk.6.attn_out.input_min
F32
F32
[1]
a.blk.6.attn_out.output_max
F32
F32
[1]
a.blk.6.attn_out.output_min
F32
F32
[1]
a.blk.6.attn_out.weight
BF16
BF16
[1024, 1024]
a.blk.6.attn_q.input_max
F32
F32
[1]
a.blk.6.attn_q.input_min
F32
F32
[1]
a.blk.6.attn_q.output_max
F32
F32
[1]
a.blk.6.attn_q.output_min
F32
F32
[1]
a.blk.6.attn_q.weight
BF16
BF16
[1024, 1024]
a.blk.6.attn_v.input_max
F32
F32
[1]
a.blk.6.attn_v.input_min
F32
F32
[1]
a.blk.6.attn_v.output_max
F32
F32
[1]
a.blk.6.attn_v.output_min
F32
F32
[1]
a.blk.6.attn_v.weight
BF16
BF16
[1024, 1024]
a.blk.6.conv_dw.weight
F32
F32
[5, 1024]
a.blk.6.conv_norm.weight
F32
F32
[1024]
a.blk.6.conv_pw1.input_max
F32
F32
[1]
a.blk.6.conv_pw1.input_min
F32
F32
[1]
a.blk.6.conv_pw1.output_max
F32
F32
[1]
a.blk.6.conv_pw1.output_min
F32
F32
[1]
a.blk.6.conv_pw1.weight
BF16
BF16
[1024, 2048]
a.blk.6.conv_pw2.input_max
F32
F32
[1]
a.blk.6.conv_pw2.input_min
F32
F32
[1]
a.blk.6.conv_pw2.output_max
F32
F32
[1]
a.blk.6.conv_pw2.output_min
F32
F32
[1]
a.blk.6.conv_pw2.weight
BF16
BF16
[1024, 1024]
a.blk.6.ffn_down.input_max
F32
F32
[1]
a.blk.6.ffn_down.input_min
F32
F32
[1]
a.blk.6.ffn_down.output_max
F32
F32
[1]
a.blk.6.ffn_down.output_min
F32
F32
[1]
a.blk.6.ffn_down.weight
BF16
BF16
[4096, 1024]
a.blk.6.ffn_down_1.input_max
F32
F32
[1]
a.blk.6.ffn_down_1.input_min
F32
F32
[1]
a.blk.6.ffn_down_1.output_max
F32
F32
[1]
a.blk.6.ffn_down_1.output_min
F32
F32
[1]
a.blk.6.ffn_down_1.weight
BF16
BF16
[4096, 1024]
a.blk.6.ffn_norm.weight
F32
F32
[1024]
a.blk.6.ffn_norm_1.weight
F32
F32
[1024]
a.blk.6.ffn_post_norm.weight
F32
F32
[1024]
a.blk.6.ffn_post_norm_1.weight
F32
F32
[1024]
a.blk.6.ffn_up.input_max
F32
F32
[1]
a.blk.6.ffn_up.input_min
F32
F32
[1]
a.blk.6.ffn_up.output_max
F32
F32
[1]
a.blk.6.ffn_up.output_min
F32
F32
[1]
a.blk.6.ffn_up.weight
BF16
BF16
[1024, 4096]
a.blk.6.ffn_up_1.input_max
F32
F32
[1]
a.blk.6.ffn_up_1.input_min
F32
F32
[1]
a.blk.6.ffn_up_1.output_max
F32
F32
[1]
a.blk.6.ffn_up_1.output_min
F32
F32
[1]
a.blk.6.ffn_up_1.weight
BF16
BF16
[1024, 4096]
a.blk.6.layer_pre_norm.weight
F32
F32
[1024]
a.blk.6.linear_pos.weight
BF16
BF16
[1024, 1024]
a.blk.6.ln1.weight
F32
F32
[1024]
a.blk.6.ln2.weight
F32
F32
[1024]
a.blk.6.norm_conv.weight
F32
F32
[1024]
a.blk.6.per_dim_scale.weight
F32
F32
[128]
a.blk.7
a.blk.7.attn_k.input_max
F32
F32
[1]
a.blk.7.attn_k.input_min
F32
F32
[1]
a.blk.7.attn_k.output_max
F32
F32
[1]
a.blk.7.attn_k.output_min
F32
F32
[1]
a.blk.7.attn_k.weight
BF16
BF16
[1024, 1024]
a.blk.7.attn_out.input_max
F32
F32
[1]
a.blk.7.attn_out.input_min
F32
F32
[1]
a.blk.7.attn_out.output_max
F32
F32
[1]
a.blk.7.attn_out.output_min
F32
F32
[1]
a.blk.7.attn_out.weight
BF16
BF16
[1024, 1024]
a.blk.7.attn_q.input_max
F32
F32
[1]
a.blk.7.attn_q.input_min
F32
F32
[1]
a.blk.7.attn_q.output_max
F32
F32
[1]
a.blk.7.attn_q.output_min
F32
F32
[1]
a.blk.7.attn_q.weight
BF16
BF16
[1024, 1024]
a.blk.7.attn_v.input_max
F32
F32
[1]
a.blk.7.attn_v.input_min
F32
F32
[1]
a.blk.7.attn_v.output_max
F32
F32
[1]
a.blk.7.attn_v.output_min
F32
F32
[1]
a.blk.7.attn_v.weight
BF16
BF16
[1024, 1024]
a.blk.7.conv_dw.weight
F32
F32
[5, 1024]
a.blk.7.conv_norm.weight
F32
F32
[1024]
a.blk.7.conv_pw1.input_max
F32
F32
[1]
a.blk.7.conv_pw1.input_min
F32
F32
[1]
a.blk.7.conv_pw1.output_max
F32
F32
[1]
a.blk.7.conv_pw1.output_min
F32
F32
[1]
a.blk.7.conv_pw1.weight
BF16
BF16
[1024, 2048]
a.blk.7.conv_pw2.input_max
F32
F32
[1]
a.blk.7.conv_pw2.input_min
F32
F32
[1]
a.blk.7.conv_pw2.output_max
F32
F32
[1]
a.blk.7.conv_pw2.output_min
F32
F32
[1]
a.blk.7.conv_pw2.weight
BF16
BF16
[1024, 1024]
a.blk.7.ffn_down.input_max
F32
F32
[1]
a.blk.7.ffn_down.input_min
F32
F32
[1]
a.blk.7.ffn_down.output_max
F32
F32
[1]
a.blk.7.ffn_down.output_min
F32
F32
[1]
a.blk.7.ffn_down.weight
BF16
BF16
[4096, 1024]
a.blk.7.ffn_down_1.input_max
F32
F32
[1]
a.blk.7.ffn_down_1.input_min
F32
F32
[1]
a.blk.7.ffn_down_1.output_max
F32
F32
[1]
a.blk.7.ffn_down_1.output_min
F32
F32
[1]
a.blk.7.ffn_down_1.weight
BF16
BF16
[4096, 1024]
a.blk.7.ffn_norm.weight
F32
F32
[1024]
a.blk.7.ffn_norm_1.weight
F32
F32
[1024]
a.blk.7.ffn_post_norm.weight
F32
F32
[1024]
a.blk.7.ffn_post_norm_1.weight
F32
F32
[1024]
a.blk.7.ffn_up.input_max
F32
F32
[1]
a.blk.7.ffn_up.input_min
F32
F32
[1]
a.blk.7.ffn_up.output_max
F32
F32
[1]
a.blk.7.ffn_up.output_min
F32
F32
[1]
a.blk.7.ffn_up.weight
BF16
BF16
[1024, 4096]
a.blk.7.ffn_up_1.input_max
F32
F32
[1]
a.blk.7.ffn_up_1.input_min
F32
F32
[1]
a.blk.7.ffn_up_1.output_max
F32
F32
[1]
a.blk.7.ffn_up_1.output_min
F32
F32
[1]
a.blk.7.ffn_up_1.weight
BF16
BF16
[1024, 4096]
a.blk.7.layer_pre_norm.weight
F32
F32
[1024]
a.blk.7.linear_pos.weight
BF16
BF16
[1024, 1024]
a.blk.7.ln1.weight
F32
F32
[1024]
a.blk.7.ln2.weight
F32
F32
[1024]
a.blk.7.norm_conv.weight
F32
F32
[1024]
a.blk.7.per_dim_scale.weight
F32
F32
[128]
a.blk.8
a.blk.8.attn_k.input_max
F32
F32
[1]
a.blk.8.attn_k.input_min
F32
F32
[1]
a.blk.8.attn_k.output_max
F32
F32
[1]
a.blk.8.attn_k.output_min
F32
F32
[1]
a.blk.8.attn_k.weight
BF16
BF16
[1024, 1024]
a.blk.8.attn_out.input_max
F32
F32
[1]
a.blk.8.attn_out.input_min
F32
F32
[1]
a.blk.8.attn_out.output_max
F32
F32
[1]
a.blk.8.attn_out.output_min
F32
F32
[1]
a.blk.8.attn_out.weight
BF16
BF16
[1024, 1024]
a.blk.8.attn_q.input_max
F32
F32
[1]
a.blk.8.attn_q.input_min
F32
F32
[1]
a.blk.8.attn_q.output_max
F32
F32
[1]
a.blk.8.attn_q.output_min
F32
F32
[1]
a.blk.8.attn_q.weight
BF16
BF16
[1024, 1024]
a.blk.8.attn_v.input_max
F32
F32
[1]
a.blk.8.attn_v.input_min
F32
F32
[1]
a.blk.8.attn_v.output_max
F32
F32
[1]
a.blk.8.attn_v.output_min
F32
F32
[1]
a.blk.8.attn_v.weight
BF16
BF16
[1024, 1024]
a.blk.8.conv_dw.weight
F32
F32
[5, 1024]
a.blk.8.conv_norm.weight
F32
F32
[1024]
a.blk.8.conv_pw1.input_max
F32
F32
[1]
a.blk.8.conv_pw1.input_min
F32
F32
[1]
a.blk.8.conv_pw1.output_max
F32
F32
[1]
a.blk.8.conv_pw1.output_min
F32
F32
[1]
a.blk.8.conv_pw1.weight
BF16
BF16
[1024, 2048]
a.blk.8.conv_pw2.input_max
F32
F32
[1]
a.blk.8.conv_pw2.input_min
F32
F32
[1]
a.blk.8.conv_pw2.output_max
F32
F32
[1]
a.blk.8.conv_pw2.output_min
F32
F32
[1]
a.blk.8.conv_pw2.weight
BF16
BF16
[1024, 1024]
a.blk.8.ffn_down.input_max
F32
F32
[1]
a.blk.8.ffn_down.input_min
F32
F32
[1]
a.blk.8.ffn_down.output_max
F32
F32
[1]
a.blk.8.ffn_down.output_min
F32
F32
[1]
a.blk.8.ffn_down.weight
BF16
BF16
[4096, 1024]
a.blk.8.ffn_down_1.input_max
F32
F32
[1]
a.blk.8.ffn_down_1.input_min
F32
F32
[1]
a.blk.8.ffn_down_1.output_max
F32
F32
[1]
a.blk.8.ffn_down_1.output_min
F32
F32
[1]
a.blk.8.ffn_down_1.weight
BF16
BF16
[4096, 1024]
a.blk.8.ffn_norm.weight
F32
F32
[1024]
a.blk.8.ffn_norm_1.weight
F32
F32
[1024]
a.blk.8.ffn_post_norm.weight
F32
F32
[1024]
a.blk.8.ffn_post_norm_1.weight
F32
F32
[1024]
a.blk.8.ffn_up.input_max
F32
F32
[1]
a.blk.8.ffn_up.input_min
F32
F32
[1]
a.blk.8.ffn_up.output_max
F32
F32
[1]
a.blk.8.ffn_up.output_min
F32
F32
[1]
a.blk.8.ffn_up.weight
BF16
BF16
[1024, 4096]
a.blk.8.ffn_up_1.input_max
F32
F32
[1]
a.blk.8.ffn_up_1.input_min
F32
F32
[1]
a.blk.8.ffn_up_1.output_max
F32
F32
[1]
a.blk.8.ffn_up_1.output_min
F32
F32
[1]
a.blk.8.ffn_up_1.weight
BF16
BF16
[1024, 4096]
a.blk.8.layer_pre_norm.weight
F32
F32
[1024]
a.blk.8.linear_pos.weight
BF16
BF16
[1024, 1024]
a.blk.8.ln1.weight
F32
F32
[1024]
a.blk.8.ln2.weight
F32
F32
[1024]
a.blk.8.norm_conv.weight
F32
F32
[1024]
a.blk.8.per_dim_scale.weight
F32
F32
[128]
a.blk.9
a.blk.9.attn_k.input_max
F32
F32
[1]
a.blk.9.attn_k.input_min
F32
F32
[1]
a.blk.9.attn_k.output_max
F32
F32
[1]
a.blk.9.attn_k.output_min
F32
F32
[1]
a.blk.9.attn_k.weight
BF16
BF16
[1024, 1024]
a.blk.9.attn_out.input_max
F32
F32
[1]
a.blk.9.attn_out.input_min
F32
F32
[1]
a.blk.9.attn_out.output_max
F32
F32
[1]
a.blk.9.attn_out.output_min
F32
F32
[1]
a.blk.9.attn_out.weight
BF16
BF16
[1024, 1024]
a.blk.9.attn_q.input_max
F32
F32
[1]
a.blk.9.attn_q.input_min
F32
F32
[1]
a.blk.9.attn_q.output_max
F32
F32
[1]
a.blk.9.attn_q.output_min
F32
F32
[1]
a.blk.9.attn_q.weight
BF16
BF16
[1024, 1024]
a.blk.9.attn_v.input_max
F32
F32
[1]
a.blk.9.attn_v.input_min
F32
F32
[1]
a.blk.9.attn_v.output_max
F32
F32
[1]
a.blk.9.attn_v.output_min
F32
F32
[1]
a.blk.9.attn_v.weight
BF16
BF16
[1024, 1024]
a.blk.9.conv_dw.weight
F32
F32
[5, 1024]
a.blk.9.conv_norm.weight
F32
F32
[1024]
a.blk.9.conv_pw1.input_max
F32
F32
[1]
a.blk.9.conv_pw1.input_min
F32
F32
[1]
a.blk.9.conv_pw1.output_max
F32
F32
[1]
a.blk.9.conv_pw1.output_min
F32
F32
[1]
a.blk.9.conv_pw1.weight
BF16
BF16
[1024, 2048]
a.blk.9.conv_pw2.input_max
F32
F32
[1]
a.blk.9.conv_pw2.input_min
F32
F32
[1]
a.blk.9.conv_pw2.output_max
F32
F32
[1]
a.blk.9.conv_pw2.output_min
F32
F32
[1]
a.blk.9.conv_pw2.weight
BF16
BF16
[1024, 1024]
a.blk.9.ffn_down.input_max
F32
F32
[1]
a.blk.9.ffn_down.input_min
F32
F32
[1]
a.blk.9.ffn_down.output_max
F32
F32
[1]
a.blk.9.ffn_down.output_min
F32
F32
[1]
a.blk.9.ffn_down.weight
BF16
BF16
[4096, 1024]
a.blk.9.ffn_down_1.input_max
F32
F32
[1]
a.blk.9.ffn_down_1.input_min
F32
F32
[1]
a.blk.9.ffn_down_1.output_max
F32
F32
[1]
a.blk.9.ffn_down_1.output_min
F32
F32
[1]
a.blk.9.ffn_down_1.weight
BF16
BF16
[4096, 1024]
a.blk.9.ffn_norm.weight
F32
F32
[1024]
a.blk.9.ffn_norm_1.weight
F32
F32
[1024]
a.blk.9.ffn_post_norm.weight
F32
F32
[1024]
a.blk.9.ffn_post_norm_1.weight
F32
F32
[1024]
a.blk.9.ffn_up.input_max
F32
F32
[1]
a.blk.9.ffn_up.input_min
F32
F32
[1]
a.blk.9.ffn_up.output_max
F32
F32
[1]
a.blk.9.ffn_up.output_min
F32
F32
[1]
a.blk.9.ffn_up.weight
BF16
BF16
[1024, 4096]
a.blk.9.ffn_up_1.input_max
F32
F32
[1]
a.blk.9.ffn_up_1.input_min
F32
F32
[1]
a.blk.9.ffn_up_1.output_max
F32
F32
[1]
a.blk.9.ffn_up_1.output_min
F32
F32
[1]
a.blk.9.ffn_up_1.weight
BF16
BF16
[1024, 4096]
a.blk.9.layer_pre_norm.weight
F32
F32
[1024]
a.blk.9.linear_pos.weight
BF16
BF16
[1024, 1024]
a.blk.9.ln1.weight
F32
F32
[1024]
a.blk.9.ln2.weight
F32
F32
[1024]
a.blk.9.norm_conv.weight
F32
F32
[1024]
a.blk.9.per_dim_scale.weight
F32
F32
[128]
a.blk.10
a.blk.10.attn_k.input_max
F32
F32
[1]
a.blk.10.attn_k.input_min
F32
F32
[1]
a.blk.10.attn_k.output_max
F32
F32
[1]
a.blk.10.attn_k.output_min
F32
F32
[1]
a.blk.10.attn_k.weight
BF16
BF16
[1024, 1024]
a.blk.10.attn_out.input_max
F32
F32
[1]
a.blk.10.attn_out.input_min
F32
F32
[1]
a.blk.10.attn_out.output_max
F32
F32
[1]
a.blk.10.attn_out.output_min
F32
F32
[1]
a.blk.10.attn_out.weight
BF16
BF16
[1024, 1024]
a.blk.10.attn_q.input_max
F32
F32
[1]
a.blk.10.attn_q.input_min
F32
F32
[1]
a.blk.10.attn_q.output_max
F32
F32
[1]
a.blk.10.attn_q.output_min
F32
F32
[1]
a.blk.10.attn_q.weight
BF16
BF16
[1024, 1024]
a.blk.10.attn_v.input_max
F32
F32
[1]
a.blk.10.attn_v.input_min
F32
F32
[1]
a.blk.10.attn_v.output_max
F32
F32
[1]
a.blk.10.attn_v.output_min
F32
F32
[1]
a.blk.10.attn_v.weight
BF16
BF16
[1024, 1024]
a.blk.10.conv_dw.weight
F32
F32
[5, 1024]
a.blk.10.conv_norm.weight
F32
F32
[1024]
a.blk.10.conv_pw1.input_max
F32
F32
[1]
a.blk.10.conv_pw1.input_min
F32
F32
[1]
a.blk.10.conv_pw1.output_max
F32
F32
[1]
a.blk.10.conv_pw1.output_min
F32
F32
[1]
a.blk.10.conv_pw1.weight
BF16
BF16
[1024, 2048]
a.blk.10.conv_pw2.input_max
F32
F32
[1]
a.blk.10.conv_pw2.input_min
F32
F32
[1]
a.blk.10.conv_pw2.output_max
F32
F32
[1]
a.blk.10.conv_pw2.output_min
F32
F32
[1]
a.blk.10.conv_pw2.weight
BF16
BF16
[1024, 1024]
a.blk.10.ffn_down.input_max
F32
F32
[1]
a.blk.10.ffn_down.input_min
F32
F32
[1]
a.blk.10.ffn_down.output_max
F32
F32
[1]
a.blk.10.ffn_down.output_min
F32
F32
[1]
a.blk.10.ffn_down.weight
BF16
BF16
[4096, 1024]
a.blk.10.ffn_down_1.input_max
F32
F32
[1]
a.blk.10.ffn_down_1.input_min
F32
F32
[1]
a.blk.10.ffn_down_1.output_max
F32
F32
[1]
a.blk.10.ffn_down_1.output_min
F32
F32
[1]
a.blk.10.ffn_down_1.weight
BF16
BF16
[4096, 1024]
a.blk.10.ffn_norm.weight
F32
F32
[1024]
a.blk.10.ffn_norm_1.weight
F32
F32
[1024]
a.blk.10.ffn_post_norm.weight
F32
F32
[1024]
a.blk.10.ffn_post_norm_1.weight
F32
F32
[1024]
a.blk.10.ffn_up.input_max
F32
F32
[1]
a.blk.10.ffn_up.input_min
F32
F32
[1]
a.blk.10.ffn_up.output_max
F32
F32
[1]
a.blk.10.ffn_up.output_min
F32
F32
[1]
a.blk.10.ffn_up.weight
BF16
BF16
[1024, 4096]
a.blk.10.ffn_up_1.input_max
F32
F32
[1]
a.blk.10.ffn_up_1.input_min
F32
F32
[1]
a.blk.10.ffn_up_1.output_max
F32
F32
[1]
a.blk.10.ffn_up_1.output_min
F32
F32
[1]
a.blk.10.ffn_up_1.weight
BF16
BF16
[1024, 4096]
a.blk.10.layer_pre_norm.weight
F32
F32
[1024]
a.blk.10.linear_pos.weight
BF16
BF16
[1024, 1024]
a.blk.10.ln1.weight
F32
F32
[1024]
a.blk.10.ln2.weight
F32
F32
[1024]
a.blk.10.norm_conv.weight
F32
F32
[1024]
a.blk.10.per_dim_scale.weight
F32
F32
[128]
a.blk.11
a.blk.11.attn_k.input_max
F32
F32
[1]
a.blk.11.attn_k.input_min
F32
F32
[1]
a.blk.11.attn_k.output_max
F32
F32
[1]
a.blk.11.attn_k.output_min
F32
F32
[1]
a.blk.11.attn_k.weight
BF16
BF16
[1024, 1024]
a.blk.11.attn_out.input_max
F32
F32
[1]
a.blk.11.attn_out.input_min
F32
F32
[1]
a.blk.11.attn_out.output_max
F32
F32
[1]
a.blk.11.attn_out.output_min
F32
F32
[1]
a.blk.11.attn_out.weight
BF16
BF16
[1024, 1024]
a.blk.11.attn_q.input_max
F32
F32
[1]
a.blk.11.attn_q.input_min
F32
F32
[1]
a.blk.11.attn_q.output_max
F32
F32
[1]
a.blk.11.attn_q.output_min
F32
F32
[1]
a.blk.11.attn_q.weight
BF16
BF16
[1024, 1024]
a.blk.11.attn_v.input_max
F32
F32
[1]
a.blk.11.attn_v.input_min
F32
F32
[1]
a.blk.11.attn_v.output_max
F32
F32
[1]
a.blk.11.attn_v.output_min
F32
F32
[1]
a.blk.11.attn_v.weight
BF16
BF16
[1024, 1024]
a.blk.11.conv_dw.weight
F32
F32
[5, 1024]
a.blk.11.conv_norm.weight
F32
F32
[1024]
a.blk.11.conv_pw1.input_max
F32
F32
[1]
a.blk.11.conv_pw1.input_min
F32
F32
[1]
a.blk.11.conv_pw1.output_max
F32
F32
[1]
a.blk.11.conv_pw1.output_min
F32
F32
[1]
a.blk.11.conv_pw1.weight
BF16
BF16
[1024, 2048]
a.blk.11.conv_pw2.input_max
F32
F32
[1]
a.blk.11.conv_pw2.input_min
F32
F32
[1]
a.blk.11.conv_pw2.output_max
F32
F32
[1]
a.blk.11.conv_pw2.output_min
F32
F32
[1]
a.blk.11.conv_pw2.weight
BF16
BF16
[1024, 1024]
a.blk.11.ffn_down.input_max
F32
F32
[1]
a.blk.11.ffn_down.input_min
F32
F32
[1]
a.blk.11.ffn_down.output_max
F32
F32
[1]
a.blk.11.ffn_down.output_min
F32
F32
[1]
a.blk.11.ffn_down.weight
BF16
BF16
[4096, 1024]
a.blk.11.ffn_down_1.input_max
F32
F32
[1]
a.blk.11.ffn_down_1.input_min
F32
F32
[1]
a.blk.11.ffn_down_1.output_max
F32
F32
[1]
a.blk.11.ffn_down_1.output_min
F32
F32
[1]
a.blk.11.ffn_down_1.weight
BF16
BF16
[4096, 1024]
a.blk.11.ffn_norm.weight
F32
F32
[1024]
a.blk.11.ffn_norm_1.weight
F32
F32
[1024]
a.blk.11.ffn_post_norm.weight
F32
F32
[1024]
a.blk.11.ffn_post_norm_1.weight
F32
F32
[1024]
a.blk.11.ffn_up.input_max
F32
F32
[1]
a.blk.11.ffn_up.input_min
F32
F32
[1]
a.blk.11.ffn_up.output_max
F32
F32
[1]
a.blk.11.ffn_up.output_min
F32
F32
[1]
a.blk.11.ffn_up.weight
BF16
BF16
[1024, 4096]
a.blk.11.ffn_up_1.input_max
F32
F32
[1]
a.blk.11.ffn_up_1.input_min
F32
F32
[1]
a.blk.11.ffn_up_1.output_max
F32
F32
[1]
a.blk.11.ffn_up_1.output_min
F32
F32
[1]
a.blk.11.ffn_up_1.weight
BF16
BF16
[1024, 4096]
a.blk.11.layer_pre_norm.weight
F32
F32
[1024]
a.blk.11.linear_pos.weight
BF16
BF16
[1024, 1024]
a.blk.11.ln1.weight
F32
F32
[1024]
a.blk.11.ln2.weight
F32
F32
[1024]
a.blk.11.norm_conv.weight
F32
F32
[1024]
a.blk.11.per_dim_scale.weight
F32
F32
[128]
a.conv1d.0.norm.weight
F32
F32
[128]
a.conv1d.0.weight
F32
F32
[3, 3, 1, 128]
a.conv1d.1.norm.weight
F32
F32
[32]
a.conv1d.1.weight
F32
F32
[3, 3, 128, 32]
a.pre_encode.out.weight
BF16
BF16
[1024, 1024]
blk.0
blk.0.attn_k.weight
Q4_K
Q4_K
[2560, 512]
blk.0.attn_k_norm.weight
F32
F32
[256]
blk.0.attn_norm.weight
F32
F32
[2560]
blk.0.attn_output.weight
Q4_K
Q4_K
[2048, 2560]
blk.0.attn_q.weight
Q4_K
Q4_K
[2560, 2048]
blk.0.attn_q_norm.weight
F32
F32
[256]
blk.0.attn_v.weight
Q6_K
Q6_K
[2560, 512]
blk.0.ffn_down.weight
Q6_K
Q6_K
[10240, 2560]
blk.0.ffn_gate.weight
Q4_K
Q4_K
[2560, 10240]
blk.0.ffn_norm.weight
F32
F32
[2560]
blk.0.ffn_up.weight
Q4_K
Q4_K
[2560, 10240]
blk.0.inp_gate.weight
Q4_K
Q4_K
[2560, 256]
blk.0.layer_output_scale.weight
F32
F32
[1]
blk.0.post_attention_norm.weight
F32
F32
[2560]
blk.0.post_ffw_norm.weight
F32
F32
[2560]
blk.0.post_norm.weight
F32
F32
[2560]
blk.0.proj.weight
Q4_K
Q4_K
[256, 2560]
blk.1
blk.1.attn_k.weight
Q4_K
Q4_K
[2560, 512]
blk.1.attn_k_norm.weight
F32
F32
[256]
blk.1.attn_norm.weight
F32
F32
[2560]
blk.1.attn_output.weight
Q4_K
Q4_K
[2048, 2560]
blk.1.attn_q.weight
Q4_K
Q4_K
[2560, 2048]
blk.1.attn_q_norm.weight
F32
F32
[256]
blk.1.attn_v.weight
Q6_K
Q6_K
[2560, 512]
blk.1.ffn_down.weight
Q6_K
Q6_K
[10240, 2560]
blk.1.ffn_gate.weight
Q4_K
Q4_K
[2560, 10240]
blk.1.ffn_norm.weight
F32
F32
[2560]
blk.1.ffn_up.weight
Q4_K
Q4_K
[2560, 10240]
blk.1.inp_gate.weight
Q4_K
Q4_K
[2560, 256]
blk.1.layer_output_scale.weight
F32
F32
[1]
blk.1.post_attention_norm.weight
F32
F32
[2560]
blk.1.post_ffw_norm.weight
F32
F32
[2560]
blk.1.post_norm.weight
F32
F32
[2560]
blk.1.proj.weight
Q4_K
Q4_K
[256, 2560]
blk.2
blk.2.attn_k.weight
Q4_K
Q4_K
[2560, 512]
blk.2.attn_k_norm.weight
F32
F32
[256]
blk.2.attn_norm.weight
F32
F32
[2560]
blk.2.attn_output.weight
Q4_K
Q4_K
[2048, 2560]
blk.2.attn_q.weight
Q4_K
Q4_K
[2560, 2048]
blk.2.attn_q_norm.weight
F32
F32
[256]
blk.2.attn_v.weight
Q6_K
Q6_K
[2560, 512]
blk.2.ffn_down.weight
Q6_K
Q6_K
[10240, 2560]
blk.2.ffn_gate.weight
Q4_K
Q4_K
[2560, 10240]
blk.2.ffn_norm.weight
F32
F32
[2560]
blk.2.ffn_up.weight
Q4_K
Q4_K
[2560, 10240]
blk.2.inp_gate.weight
Q4_K
Q4_K
[2560, 256]
blk.2.layer_output_scale.weight
F32
F32
[1]
blk.2.post_attention_norm.weight
F32
F32
[2560]
blk.2.post_ffw_norm.weight
F32
F32
[2560]
blk.2.post_norm.weight
F32
F32
[2560]
blk.2.proj.weight
Q4_K
Q4_K
[256, 2560]
blk.3
blk.3.attn_k.weight
Q4_K
Q4_K
[2560, 512]
blk.3.attn_k_norm.weight
F32
F32
[256]
blk.3.attn_norm.weight
F32
F32
[2560]
blk.3.attn_output.weight
Q4_K
Q4_K
[2048, 2560]
blk.3.attn_q.weight
Q4_K
Q4_K
[2560, 2048]
blk.3.attn_q_norm.weight
F32
F32
[256]
blk.3.attn_v.weight
Q6_K
Q6_K
[2560, 512]
blk.3.ffn_down.weight
Q6_K
Q6_K
[10240, 2560]
blk.3.ffn_gate.weight
Q4_K
Q4_K
[2560, 10240]
blk.3.ffn_norm.weight
F32
F32
[2560]
blk.3.ffn_up.weight
Q4_K
Q4_K
[2560, 10240]
blk.3.inp_gate.weight
Q4_K
Q4_K
[2560, 256]
blk.3.layer_output_scale.weight
F32
F32
[1]
blk.3.post_attention_norm.weight
F32
F32
[2560]
blk.3.post_ffw_norm.weight
F32
F32
[2560]
blk.3.post_norm.weight
F32
F32
[2560]
blk.3.proj.weight
Q4_K
Q4_K
[256, 2560]
blk.4
blk.4.attn_k.weight
Q4_K
Q4_K
[2560, 512]
blk.4.attn_k_norm.weight
F32
F32
[256]
blk.4.attn_norm.weight
F32
F32
[2560]
blk.4.attn_output.weight
Q4_K
Q4_K
[2048, 2560]
blk.4.attn_q.weight
Q4_K
Q4_K
[2560, 2048]
blk.4.attn_q_norm.weight
F32
F32
[256]
blk.4.attn_v.weight
Q6_K
Q6_K
[2560, 512]
blk.4.ffn_down.weight
Q6_K
Q6_K
[10240, 2560]
blk.4.ffn_gate.weight
Q4_K
Q4_K
[2560, 10240]
blk.4.ffn_norm.weight
F32
F32
[2560]
blk.4.ffn_up.weight
Q4_K
Q4_K
[2560, 10240]
blk.4.inp_gate.weight
Q4_K
Q4_K
[2560, 256]
blk.4.layer_output_scale.weight
F32
F32
[1]
blk.4.post_attention_norm.weight
F32
F32
[2560]
blk.4.post_ffw_norm.weight
F32
F32
[2560]
blk.4.post_norm.weight
F32
F32
[2560]
blk.4.proj.weight
Q4_K
Q4_K
[256, 2560]
blk.5
blk.5.attn_k.weight
Q4_K
Q4_K
[2560, 1024]
blk.5.attn_k_norm.weight
F32
F32
[512]
blk.5.attn_norm.weight
F32
F32
[2560]
blk.5.attn_output.weight
Q4_K
Q4_K
[4096, 2560]
blk.5.attn_q.weight
Q4_K
Q4_K
[2560, 4096]
blk.5.attn_q_norm.weight
F32
F32
[512]
blk.5.attn_v.weight
Q6_K
Q6_K
[2560, 1024]
blk.5.ffn_down.weight
Q4_K
Q4_K
[10240, 2560]
blk.5.ffn_gate.weight
Q4_K
Q4_K
[2560, 10240]
blk.5.ffn_norm.weight
F32
F32
[2560]
blk.5.ffn_up.weight
Q4_K
Q4_K
[2560, 10240]
blk.5.inp_gate.weight
Q4_K
Q4_K
[2560, 256]
blk.5.layer_output_scale.weight
F32
F32
[1]
blk.5.post_attention_norm.weight
F32
F32
[2560]
blk.5.post_ffw_norm.weight
F32
F32
[2560]
blk.5.post_norm.weight
F32
F32
[2560]
blk.5.proj.weight
Q4_K
Q4_K
[256, 2560]
blk.6
blk.6.attn_k.weight
Q4_K
Q4_K
[2560, 512]
blk.6.attn_k_norm.weight
F32
F32
[256]
blk.6.attn_norm.weight
F32
F32
[2560]
blk.6.attn_output.weight
Q4_K
Q4_K
[2048, 2560]
blk.6.attn_q.weight
Q4_K
Q4_K
[2560, 2048]
blk.6.attn_q_norm.weight
F32
F32
[256]
blk.6.attn_v.weight
Q6_K
Q6_K
[2560, 512]
blk.6.ffn_down.weight
Q4_K
Q4_K
[10240, 2560]
blk.6.ffn_gate.weight
Q4_K
Q4_K
[2560, 10240]
blk.6.ffn_norm.weight
F32
F32
[2560]
blk.6.ffn_up.weight
Q4_K
Q4_K
[2560, 10240]
blk.6.inp_gate.weight
Q4_K
Q4_K
[2560, 256]
blk.6.layer_output_scale.weight
F32
F32
[1]
blk.6.post_attention_norm.weight
F32
F32
[2560]
blk.6.post_ffw_norm.weight
F32
F32
[2560]
blk.6.post_norm.weight
F32
F32
[2560]
blk.6.proj.weight
Q4_K
Q4_K
[256, 2560]
blk.7
blk.7.attn_k.weight
Q4_K
Q4_K
[2560, 512]
blk.7.attn_k_norm.weight
F32
F32
[256]
blk.7.attn_norm.weight
F32
F32
[2560]
blk.7.attn_output.weight
Q4_K
Q4_K
[2048, 2560]
blk.7.attn_q.weight
Q4_K
Q4_K
[2560, 2048]
blk.7.attn_q_norm.weight
F32
F32
[256]
blk.7.attn_v.weight
Q6_K
Q6_K
[2560, 512]
blk.7.ffn_down.weight
Q6_K
Q6_K
[10240, 2560]
blk.7.ffn_gate.weight
Q4_K
Q4_K
[2560, 10240]
blk.7.ffn_norm.weight
F32
F32
[2560]
blk.7.ffn_up.weight
Q4_K
Q4_K
[2560, 10240]
blk.7.inp_gate.weight
Q4_K
Q4_K
[2560, 256]
blk.7.layer_output_scale.weight
F32
F32
[1]
blk.7.post_attention_norm.weight
F32
F32
[2560]
blk.7.post_ffw_norm.weight
F32
F32
[2560]
blk.7.post_norm.weight
F32
F32
[2560]
blk.7.proj.weight
Q4_K
Q4_K
[256, 2560]
blk.8
blk.8.attn_k.weight
Q4_K
Q4_K
[2560, 512]
blk.8.attn_k_norm.weight
F32
F32
[256]
blk.8.attn_norm.weight
F32
F32
[2560]
blk.8.attn_output.weight
Q4_K
Q4_K
[2048, 2560]
blk.8.attn_q.weight
Q4_K
Q4_K
[2560, 2048]
blk.8.attn_q_norm.weight
F32
F32
[256]
blk.8.attn_v.weight
Q4_K
Q4_K
[2560, 512]
blk.8.ffn_down.weight
Q4_K
Q4_K
[10240, 2560]
blk.8.ffn_gate.weight
Q4_K
Q4_K
[2560, 10240]
blk.8.ffn_norm.weight
F32
F32
[2560]
blk.8.ffn_up.weight
Q4_K
Q4_K
[2560, 10240]
blk.8.inp_gate.weight
Q4_K
Q4_K
[2560, 256]
blk.8.layer_output_scale.weight
F32
F32
[1]
blk.8.post_attention_norm.weight
F32
F32
[2560]
blk.8.post_ffw_norm.weight
F32
F32
[2560]
blk.8.post_norm.weight
F32
F32
[2560]
blk.8.proj.weight
Q4_K
Q4_K
[256, 2560]
blk.9
blk.9.attn_k.weight
Q4_K
Q4_K
[2560, 512]
blk.9.attn_k_norm.weight
F32
F32
[256]
blk.9.attn_norm.weight
F32
F32
[2560]
blk.9.attn_output.weight
Q4_K
Q4_K
[2048, 2560]
blk.9.attn_q.weight
Q4_K
Q4_K
[2560, 2048]
blk.9.attn_q_norm.weight
F32
F32
[256]
blk.9.attn_v.weight
Q4_K
Q4_K
[2560, 512]
blk.9.ffn_down.weight
Q4_K
Q4_K
[10240, 2560]
blk.9.ffn_gate.weight
Q4_K
Q4_K
[2560, 10240]
blk.9.ffn_norm.weight
F32
F32
[2560]
blk.9.ffn_up.weight
Q4_K
Q4_K
[2560, 10240]
blk.9.inp_gate.weight
Q4_K
Q4_K
[2560, 256]
blk.9.layer_output_scale.weight
F32
F32
[1]
blk.9.post_attention_norm.weight
F32
F32
[2560]
blk.9.post_ffw_norm.weight
F32
F32
[2560]
blk.9.post_norm.weight
F32
F32
[2560]
blk.9.proj.weight
Q4_K
Q4_K
[256, 2560]
blk.10
blk.10.attn_k.weight
Q4_K
Q4_K
[2560, 512]
blk.10.attn_k_norm.weight
F32
F32
[256]
blk.10.attn_norm.weight
F32
F32
[2560]
blk.10.attn_output.weight
Q4_K
Q4_K
[2048, 2560]
blk.10.attn_q.weight
Q4_K
Q4_K
[2560, 2048]
blk.10.attn_q_norm.weight
F32
F32
[256]
blk.10.attn_v.weight
Q6_K
Q6_K
[2560, 512]
blk.10.ffn_down.weight
Q6_K
Q6_K
[10240, 2560]
blk.10.ffn_gate.weight
Q4_K
Q4_K
[2560, 10240]
blk.10.ffn_norm.weight
F32
F32
[2560]
blk.10.ffn_up.weight
Q4_K
Q4_K
[2560, 10240]
blk.10.inp_gate.weight
Q4_K
Q4_K
[2560, 256]
blk.10.layer_output_scale.weight
F32
F32
[1]
blk.10.post_attention_norm.weight
F32
F32
[2560]
blk.10.post_ffw_norm.weight
F32
F32
[2560]
blk.10.post_norm.weight
F32
F32
[2560]
blk.10.proj.weight
Q4_K
Q4_K
[256, 2560]
blk.11
blk.11.attn_k.weight
Q4_K
Q4_K
[2560, 1024]
blk.11.attn_k_norm.weight
F32
F32
[512]
blk.11.attn_norm.weight
F32
F32
[2560]
blk.11.attn_output.weight
Q4_K
Q4_K
[4096, 2560]
blk.11.attn_q.weight
Q4_K
Q4_K
[2560, 4096]
blk.11.attn_q_norm.weight
F32
F32
[512]
blk.11.attn_v.weight
Q4_K
Q4_K
[2560, 1024]
blk.11.ffn_down.weight
Q4_K
Q4_K
[10240, 2560]
blk.11.ffn_gate.weight
Q4_K
Q4_K
[2560, 10240]
blk.11.ffn_norm.weight
F32
F32
[2560]
blk.11.ffn_up.weight
Q4_K
Q4_K
[2560, 10240]
blk.11.inp_gate.weight
Q4_K
Q4_K
[2560, 256]
blk.11.layer_output_scale.weight
F32
F32
[1]
blk.11.post_attention_norm.weight
F32
F32
[2560]
blk.11.post_ffw_norm.weight
F32
F32
[2560]
blk.11.post_norm.weight
F32
F32
[2560]
blk.11.proj.weight
Q4_K
Q4_K
[256, 2560]
blk.12
blk.12.attn_k.weight
Q4_K
Q4_K
[2560, 512]
blk.12.attn_k_norm.weight
F32
F32
[256]
blk.12.attn_norm.weight
F32
F32
[2560]
blk.12.attn_output.weight
Q4_K
Q4_K
[2048, 2560]
blk.12.attn_q.weight
Q4_K
Q4_K
[2560, 2048]
blk.12.attn_q_norm.weight
F32
F32
[256]
blk.12.attn_v.weight
Q4_K
Q4_K
[2560, 512]
blk.12.ffn_down.weight
Q4_K
Q4_K
[10240, 2560]
blk.12.ffn_gate.weight
Q4_K
Q4_K
[2560, 10240]
blk.12.ffn_norm.weight
F32
F32
[2560]
blk.12.ffn_up.weight
Q4_K
Q4_K
[2560, 10240]
blk.12.inp_gate.weight
Q4_K
Q4_K
[2560, 256]
blk.12.layer_output_scale.weight
F32
F32
[1]
blk.12.post_attention_norm.weight
F32
F32
[2560]
blk.12.post_ffw_norm.weight
F32
F32
[2560]
blk.12.post_norm.weight
F32
F32
[2560]
blk.12.proj.weight
Q4_K
Q4_K
[256, 2560]
blk.13
blk.13.attn_k.weight
Q4_K
Q4_K
[2560, 512]
blk.13.attn_k_norm.weight
F32
F32
[256]
blk.13.attn_norm.weight
F32
F32
[2560]
blk.13.attn_output.weight
Q4_K
Q4_K
[2048, 2560]
blk.13.attn_q.weight
Q4_K
Q4_K
[2560, 2048]
blk.13.attn_q_norm.weight
F32
F32
[256]
blk.13.attn_v.weight
Q6_K
Q6_K
[2560, 512]
blk.13.ffn_down.weight
Q6_K
Q6_K
[10240, 2560]
blk.13.ffn_gate.weight
Q4_K
Q4_K
[2560, 10240]
blk.13.ffn_norm.weight
F32
F32
[2560]
blk.13.ffn_up.weight
Q4_K
Q4_K
[2560, 10240]
blk.13.inp_gate.weight
Q4_K
Q4_K
[2560, 256]
blk.13.layer_output_scale.weight
F32
F32
[1]
blk.13.post_attention_norm.weight
F32
F32
[2560]
blk.13.post_ffw_norm.weight
F32
F32
[2560]
blk.13.post_norm.weight
F32
F32
[2560]
blk.13.proj.weight
Q4_K
Q4_K
[256, 2560]
blk.14
blk.14.attn_k.weight
Q4_K
Q4_K
[2560, 512]
blk.14.attn_k_norm.weight
F32
F32
[256]
blk.14.attn_norm.weight
F32
F32
[2560]
blk.14.attn_output.weight
Q4_K
Q4_K
[2048, 2560]
blk.14.attn_q.weight
Q4_K
Q4_K
[2560, 2048]
blk.14.attn_q_norm.weight
F32
F32
[256]
blk.14.attn_v.weight
Q4_K
Q4_K
[2560, 512]
blk.14.ffn_down.weight
Q4_K
Q4_K
[10240, 2560]
blk.14.ffn_gate.weight
Q4_K
Q4_K
[2560, 10240]
blk.14.ffn_norm.weight
F32
F32
[2560]
blk.14.ffn_up.weight
Q4_K
Q4_K
[2560, 10240]
blk.14.inp_gate.weight
Q4_K
Q4_K
[2560, 256]
blk.14.layer_output_scale.weight
F32
F32
[1]
blk.14.post_attention_norm.weight
F32
F32
[2560]
blk.14.post_ffw_norm.weight
F32
F32
[2560]
blk.14.post_norm.weight
F32
F32
[2560]
blk.14.proj.weight
Q4_K
Q4_K
[256, 2560]
blk.15
blk.15.attn_k.weight
Q4_K
Q4_K
[2560, 512]
blk.15.attn_k_norm.weight
F32
F32
[256]
blk.15.attn_norm.weight
F32
F32
[2560]
blk.15.attn_output.weight
Q4_K
Q4_K
[2048, 2560]
blk.15.attn_q.weight
Q4_K
Q4_K
[2560, 2048]
blk.15.attn_q_norm.weight
F32
F32
[256]
blk.15.attn_v.weight
Q4_K
Q4_K
[2560, 512]
blk.15.ffn_down.weight
Q4_K
Q4_K
[10240, 2560]
blk.15.ffn_gate.weight
Q4_K
Q4_K
[2560, 10240]
blk.15.ffn_norm.weight
F32
F32
[2560]
blk.15.ffn_up.weight
Q4_K
Q4_K
[2560, 10240]
blk.15.inp_gate.weight
Q4_K
Q4_K
[2560, 256]
blk.15.layer_output_scale.weight
F32
F32
[1]
blk.15.post_attention_norm.weight
F32
F32
[2560]
blk.15.post_ffw_norm.weight
F32
F32
[2560]
blk.15.post_norm.weight
F32
F32
[2560]
blk.15.proj.weight
Q4_K
Q4_K
[256, 2560]
blk.16
blk.16.attn_k.weight
Q4_K
Q4_K
[2560, 512]
blk.16.attn_k_norm.weight
F32
F32
[256]
blk.16.attn_norm.weight
F32
F32
[2560]
blk.16.attn_output.weight
Q4_K
Q4_K
[2048, 2560]
blk.16.attn_q.weight
Q4_K
Q4_K
[2560, 2048]
blk.16.attn_q_norm.weight
F32
F32
[256]
blk.16.attn_v.weight
Q6_K
Q6_K
[2560, 512]
blk.16.ffn_down.weight
Q6_K
Q6_K
[10240, 2560]
blk.16.ffn_gate.weight
Q4_K
Q4_K
[2560, 10240]
blk.16.ffn_norm.weight
F32
F32
[2560]
blk.16.ffn_up.weight
Q4_K
Q4_K
[2560, 10240]
blk.16.inp_gate.weight
Q4_K
Q4_K
[2560, 256]
blk.16.layer_output_scale.weight
F32
F32
[1]
blk.16.post_attention_norm.weight
F32
F32
[2560]
blk.16.post_ffw_norm.weight
F32
F32
[2560]
blk.16.post_norm.weight
F32
F32
[2560]
blk.16.proj.weight
Q4_K
Q4_K
[256, 2560]
blk.17
blk.17.attn_k.weight
Q4_K
Q4_K
[2560, 1024]
blk.17.attn_k_norm.weight
F32
F32
[512]
blk.17.attn_norm.weight
F32
F32
[2560]
blk.17.attn_output.weight
Q4_K
Q4_K
[4096, 2560]
blk.17.attn_q.weight
Q4_K
Q4_K
[2560, 4096]
blk.17.attn_q_norm.weight
F32
F32
[512]
blk.17.attn_v.weight
Q4_K
Q4_K
[2560, 1024]
blk.17.ffn_down.weight
Q4_K
Q4_K
[10240, 2560]
blk.17.ffn_gate.weight
Q4_K
Q4_K
[2560, 10240]
blk.17.ffn_norm.weight
F32
F32
[2560]
blk.17.ffn_up.weight
Q4_K
Q4_K
[2560, 10240]
blk.17.inp_gate.weight
Q4_K
Q4_K
[2560, 256]
blk.17.layer_output_scale.weight
F32
F32
[1]
blk.17.post_attention_norm.weight
F32
F32
[2560]
blk.17.post_ffw_norm.weight
F32
F32
[2560]
blk.17.post_norm.weight
F32
F32
[2560]
blk.17.proj.weight
Q4_K
Q4_K
[256, 2560]
blk.18
blk.18.attn_k.weight
Q4_K
Q4_K
[2560, 512]
blk.18.attn_k_norm.weight
F32
F32
[256]
blk.18.attn_norm.weight
F32
F32
[2560]
blk.18.attn_output.weight
Q4_K
Q4_K
[2048, 2560]
blk.18.attn_q.weight
Q4_K
Q4_K
[2560, 2048]
blk.18.attn_q_norm.weight
F32
F32
[256]
blk.18.attn_v.weight
Q4_K
Q4_K
[2560, 512]
blk.18.ffn_down.weight
Q4_K
Q4_K
[10240, 2560]
blk.18.ffn_gate.weight
Q4_K
Q4_K
[2560, 10240]
blk.18.ffn_norm.weight
F32
F32
[2560]
blk.18.ffn_up.weight
Q4_K
Q4_K
[2560, 10240]
blk.18.inp_gate.weight
Q4_K
Q4_K
[2560, 256]
blk.18.layer_output_scale.weight
F32
F32
[1]
blk.18.post_attention_norm.weight
F32
F32
[2560]
blk.18.post_ffw_norm.weight
F32
F32
[2560]
blk.18.post_norm.weight
F32
F32
[2560]
blk.18.proj.weight
Q4_K
Q4_K
[256, 2560]
blk.19
blk.19.attn_k.weight
Q4_K
Q4_K
[2560, 512]
blk.19.attn_k_norm.weight
F32
F32
[256]
blk.19.attn_norm.weight
F32
F32
[2560]
blk.19.attn_output.weight
Q4_K
Q4_K
[2048, 2560]
blk.19.attn_q.weight
Q4_K
Q4_K
[2560, 2048]
blk.19.attn_q_norm.weight
F32
F32
[256]
blk.19.attn_v.weight
Q6_K
Q6_K
[2560, 512]
blk.19.ffn_down.weight
Q6_K
Q6_K
[10240, 2560]
blk.19.ffn_gate.weight
Q4_K
Q4_K
[2560, 10240]
blk.19.ffn_norm.weight
F32
F32
[2560]
blk.19.ffn_up.weight
Q4_K
Q4_K
[2560, 10240]
blk.19.inp_gate.weight
Q4_K
Q4_K
[2560, 256]
blk.19.layer_output_scale.weight
F32
F32
[1]
blk.19.post_attention_norm.weight
F32
F32
[2560]
blk.19.post_ffw_norm.weight
F32
F32
[2560]
blk.19.post_norm.weight
F32
F32
[2560]
blk.19.proj.weight
Q4_K
Q4_K
[256, 2560]
blk.20
blk.20.attn_k.weight
Q4_K
Q4_K
[2560, 512]
blk.20.attn_k_norm.weight
F32
F32
[256]
blk.20.attn_norm.weight
F32
F32
[2560]
blk.20.attn_output.weight
Q4_K
Q4_K
[2048, 2560]
blk.20.attn_q.weight
Q4_K
Q4_K
[2560, 2048]
blk.20.attn_q_norm.weight
F32
F32
[256]
blk.20.attn_v.weight
Q4_K
Q4_K
[2560, 512]
blk.20.ffn_down.weight
Q4_K
Q4_K
[10240, 2560]
blk.20.ffn_gate.weight
Q4_K
Q4_K
[2560, 10240]
blk.20.ffn_norm.weight
F32
F32
[2560]
blk.20.ffn_up.weight
Q4_K
Q4_K
[2560, 10240]
blk.20.inp_gate.weight
Q4_K
Q4_K
[2560, 256]
blk.20.layer_output_scale.weight
F32
F32
[1]
blk.20.post_attention_norm.weight
F32
F32
[2560]
blk.20.post_ffw_norm.weight
F32
F32
[2560]
blk.20.post_norm.weight
F32
F32
[2560]
blk.20.proj.weight
Q4_K
Q4_K
[256, 2560]
blk.21
blk.21.attn_k.weight
Q4_K
Q4_K
[2560, 512]
blk.21.attn_k_norm.weight
F32
F32
[256]
blk.21.attn_norm.weight
F32
F32
[2560]
blk.21.attn_output.weight
Q4_K
Q4_K
[2048, 2560]
blk.21.attn_q.weight
Q4_K
Q4_K
[2560, 2048]
blk.21.attn_q_norm.weight
F32
F32
[256]
blk.21.attn_v.weight
Q4_K
Q4_K
[2560, 512]
blk.21.ffn_down.weight
Q4_K
Q4_K
[10240, 2560]
blk.21.ffn_gate.weight
Q4_K
Q4_K
[2560, 10240]
blk.21.ffn_norm.weight
F32
F32
[2560]
blk.21.ffn_up.weight
Q4_K
Q4_K
[2560, 10240]
blk.21.inp_gate.weight
Q4_K
Q4_K
[2560, 256]
blk.21.layer_output_scale.weight
F32
F32
[1]
blk.21.post_attention_norm.weight
F32
F32
[2560]
blk.21.post_ffw_norm.weight
F32
F32
[2560]
blk.21.post_norm.weight
F32
F32
[2560]
blk.21.proj.weight
Q4_K
Q4_K
[256, 2560]
blk.22
blk.22.attn_k.weight
Q4_K
Q4_K
[2560, 512]
blk.22.attn_k_norm.weight
F32
F32
[256]
blk.22.attn_norm.weight
F32
F32
[2560]
blk.22.attn_output.weight
Q4_K
Q4_K
[2048, 2560]
blk.22.attn_q.weight
Q4_K
Q4_K
[2560, 2048]
blk.22.attn_q_norm.weight
F32
F32
[256]
blk.22.attn_v.weight
Q6_K
Q6_K
[2560, 512]
blk.22.ffn_down.weight
Q6_K
Q6_K
[10240, 2560]
blk.22.ffn_gate.weight
Q4_K
Q4_K
[2560, 10240]
blk.22.ffn_norm.weight
F32
F32
[2560]
blk.22.ffn_up.weight
Q4_K
Q4_K
[2560, 10240]
blk.22.inp_gate.weight
Q4_K
Q4_K
[2560, 256]
blk.22.layer_output_scale.weight
F32
F32
[1]
blk.22.post_attention_norm.weight
F32
F32
[2560]
blk.22.post_ffw_norm.weight
F32
F32
[2560]
blk.22.post_norm.weight
F32
F32
[2560]
blk.22.proj.weight
Q4_K
Q4_K
[256, 2560]
blk.23
blk.23.attn_k.weight
Q4_K
Q4_K
[2560, 1024]
blk.23.attn_k_norm.weight
F32
F32
[512]
blk.23.attn_norm.weight
F32
F32
[2560]
blk.23.attn_output.weight
Q4_K
Q4_K
[4096, 2560]
blk.23.attn_q.weight
Q4_K
Q4_K
[2560, 4096]
blk.23.attn_q_norm.weight
F32
F32
[512]
blk.23.attn_v.weight
Q4_K
Q4_K
[2560, 1024]
blk.23.ffn_down.weight
Q4_K
Q4_K
[10240, 2560]
blk.23.ffn_gate.weight
Q4_K
Q4_K
[2560, 10240]
blk.23.ffn_norm.weight
F32
F32
[2560]
blk.23.ffn_up.weight
Q4_K
Q4_K
[2560, 10240]
blk.23.inp_gate.weight
Q4_K
Q4_K
[2560, 256]
blk.23.layer_output_scale.weight
F32
F32
[1]
blk.23.post_attention_norm.weight
F32
F32
[2560]
blk.23.post_ffw_norm.weight
F32
F32
[2560]
blk.23.post_norm.weight
F32
F32
[2560]
blk.23.proj.weight
Q4_K
Q4_K
[256, 2560]
blk.24
blk.24.attn_k.weight
Q4_K
Q4_K
[2560, 512]
blk.24.attn_k_norm.weight
F32
F32
[256]
blk.24.attn_norm.weight
F32
F32
[2560]
blk.24.attn_output.weight
Q4_K
Q4_K
[2048, 2560]
blk.24.attn_q.weight
Q4_K
Q4_K
[2560, 2048]
blk.24.attn_q_norm.weight
F32
F32
[256]
blk.24.attn_v.weight
Q4_K
Q4_K
[2560, 512]
blk.24.ffn_down.weight
Q4_K
Q4_K
[10240, 2560]
blk.24.ffn_gate.weight
Q4_K
Q4_K
[2560, 10240]
blk.24.ffn_norm.weight
F32
F32
[2560]
blk.24.ffn_up.weight
Q4_K
Q4_K
[2560, 10240]
blk.24.inp_gate.weight
Q4_K
Q4_K
[2560, 256]
blk.24.layer_output_scale.weight
F32
F32
[1]
blk.24.post_attention_norm.weight
F32
F32
[2560]
blk.24.post_ffw_norm.weight
F32
F32
[2560]
blk.24.post_norm.weight
F32
F32
[2560]
blk.24.proj.weight
Q4_K
Q4_K
[256, 2560]
blk.25
blk.25.attn_k.weight
Q4_K
Q4_K
[2560, 512]
blk.25.attn_k_norm.weight
F32
F32
[256]
blk.25.attn_norm.weight
F32
F32
[2560]
blk.25.attn_output.weight
Q4_K
Q4_K
[2048, 2560]
blk.25.attn_q.weight
Q4_K
Q4_K
[2560, 2048]
blk.25.attn_q_norm.weight
F32
F32
[256]
blk.25.attn_v.weight
Q6_K
Q6_K
[2560, 512]
blk.25.ffn_down.weight
Q6_K
Q6_K
[10240, 2560]
blk.25.ffn_gate.weight
Q4_K
Q4_K
[2560, 10240]
blk.25.ffn_norm.weight
F32
F32
[2560]
blk.25.ffn_up.weight
Q4_K
Q4_K
[2560, 10240]
blk.25.inp_gate.weight
Q4_K
Q4_K
[2560, 256]
blk.25.layer_output_scale.weight
F32
F32
[1]
blk.25.post_attention_norm.weight
F32
F32
[2560]
blk.25.post_ffw_norm.weight
F32
F32
[2560]
blk.25.post_norm.weight
F32
F32
[2560]
blk.25.proj.weight
Q4_K
Q4_K
[256, 2560]
blk.26
blk.26.attn_k.weight
Q4_K
Q4_K
[2560, 512]
blk.26.attn_k_norm.weight
F32
F32
[256]
blk.26.attn_norm.weight
F32
F32
[2560]
blk.26.attn_output.weight
Q4_K
Q4_K
[2048, 2560]
blk.26.attn_q.weight
Q4_K
Q4_K
[2560, 2048]
blk.26.attn_q_norm.weight
F32
F32
[256]
blk.26.attn_v.weight
Q4_K
Q4_K
[2560, 512]
blk.26.ffn_down.weight
Q4_K
Q4_K
[10240, 2560]
blk.26.ffn_gate.weight
Q4_K
Q4_K
[2560, 10240]
blk.26.ffn_norm.weight
F32
F32
[2560]
blk.26.ffn_up.weight
Q4_K
Q4_K
[2560, 10240]
blk.26.inp_gate.weight
Q4_K
Q4_K
[2560, 256]
blk.26.layer_output_scale.weight
F32
F32
[1]
blk.26.post_attention_norm.weight
F32
F32
[2560]
blk.26.post_ffw_norm.weight
F32
F32
[2560]
blk.26.post_norm.weight
F32
F32
[2560]
blk.26.proj.weight
Q4_K
Q4_K
[256, 2560]
blk.27
blk.27.attn_k.weight
Q4_K
Q4_K
[2560, 512]
blk.27.attn_k_norm.weight
F32
F32
[256]
blk.27.attn_norm.weight
F32
F32
[2560]
blk.27.attn_output.weight
Q4_K
Q4_K
[2048, 2560]
blk.27.attn_q.weight
Q4_K
Q4_K
[2560, 2048]
blk.27.attn_q_norm.weight
F32
F32
[256]
blk.27.attn_v.weight
Q4_K
Q4_K
[2560, 512]
blk.27.ffn_down.weight
Q4_K
Q4_K
[10240, 2560]
blk.27.ffn_gate.weight
Q4_K
Q4_K
[2560, 10240]
blk.27.ffn_norm.weight
F32
F32
[2560]
blk.27.ffn_up.weight
Q4_K
Q4_K
[2560, 10240]
blk.27.inp_gate.weight
Q4_K
Q4_K
[2560, 256]
blk.27.layer_output_scale.weight
F32
F32
[1]
blk.27.post_attention_norm.weight
F32
F32
[2560]
blk.27.post_ffw_norm.weight
F32
F32
[2560]
blk.27.post_norm.weight
F32
F32
[2560]
blk.27.proj.weight
Q4_K
Q4_K
[256, 2560]
blk.28
blk.28.attn_k.weight
Q4_K
Q4_K
[2560, 512]
blk.28.attn_k_norm.weight
F32
F32
[256]
blk.28.attn_norm.weight
F32
F32
[2560]
blk.28.attn_output.weight
Q4_K
Q4_K
[2048, 2560]
blk.28.attn_q.weight
Q4_K
Q4_K
[2560, 2048]
blk.28.attn_q_norm.weight
F32
F32
[256]
blk.28.attn_v.weight
Q6_K
Q6_K
[2560, 512]
blk.28.ffn_down.weight
Q6_K
Q6_K
[10240, 2560]
blk.28.ffn_gate.weight
Q4_K
Q4_K
[2560, 10240]
blk.28.ffn_norm.weight
F32
F32
[2560]
blk.28.ffn_up.weight
Q4_K
Q4_K
[2560, 10240]
blk.28.inp_gate.weight
Q4_K
Q4_K
[2560, 256]
blk.28.layer_output_scale.weight
F32
F32
[1]
blk.28.post_attention_norm.weight
F32
F32
[2560]
blk.28.post_ffw_norm.weight
F32
F32
[2560]
blk.28.post_norm.weight
F32
F32
[2560]
blk.28.proj.weight
Q4_K
Q4_K
[256, 2560]
blk.29
blk.29.attn_k.weight
Q4_K
Q4_K
[2560, 1024]
blk.29.attn_k_norm.weight
F32
F32
[512]
blk.29.attn_norm.weight
F32
F32
[2560]
blk.29.attn_output.weight
Q4_K
Q4_K
[4096, 2560]
blk.29.attn_q.weight
Q4_K
Q4_K
[2560, 4096]
blk.29.attn_q_norm.weight
F32
F32
[512]
blk.29.attn_v.weight
Q4_K
Q4_K
[2560, 1024]
blk.29.ffn_down.weight
Q4_K
Q4_K
[10240, 2560]
blk.29.ffn_gate.weight
Q4_K
Q4_K
[2560, 10240]
blk.29.ffn_norm.weight
F32
F32
[2560]
blk.29.ffn_up.weight
Q4_K
Q4_K
[2560, 10240]
blk.29.inp_gate.weight
Q4_K
Q4_K
[2560, 256]
blk.29.layer_output_scale.weight
F32
F32
[1]
blk.29.post_attention_norm.weight
F32
F32
[2560]
blk.29.post_ffw_norm.weight
F32
F32
[2560]
blk.29.post_norm.weight
F32
F32
[2560]
blk.29.proj.weight
Q4_K
Q4_K
[256, 2560]
blk.30
blk.30.attn_k.weight
Q4_K
Q4_K
[2560, 512]
blk.30.attn_k_norm.weight
F32
F32
[256]
blk.30.attn_norm.weight
F32
F32
[2560]
blk.30.attn_output.weight
Q4_K
Q4_K
[2048, 2560]
blk.30.attn_q.weight
Q4_K
Q4_K
[2560, 2048]
blk.30.attn_q_norm.weight
F32
F32
[256]
blk.30.attn_v.weight
Q4_K
Q4_K
[2560, 512]
blk.30.ffn_down.weight
Q4_K
Q4_K
[10240, 2560]
blk.30.ffn_gate.weight
Q4_K
Q4_K
[2560, 10240]
blk.30.ffn_norm.weight
F32
F32
[2560]
blk.30.ffn_up.weight
Q4_K
Q4_K
[2560, 10240]
blk.30.inp_gate.weight
Q4_K
Q4_K
[2560, 256]
blk.30.layer_output_scale.weight
F32
F32
[1]
blk.30.post_attention_norm.weight
F32
F32
[2560]
blk.30.post_ffw_norm.weight
F32
F32
[2560]
blk.30.post_norm.weight
F32
F32
[2560]
blk.30.proj.weight
Q4_K
Q4_K
[256, 2560]
blk.31
blk.31.attn_k.weight
Q4_K
Q4_K
[2560, 512]
blk.31.attn_k_norm.weight
F32
F32
[256]
blk.31.attn_norm.weight
F32
F32
[2560]
blk.31.attn_output.weight
Q4_K
Q4_K
[2048, 2560]
blk.31.attn_q.weight
Q4_K
Q4_K
[2560, 2048]
blk.31.attn_q_norm.weight
F32
F32
[256]
blk.31.attn_v.weight
Q6_K
Q6_K
[2560, 512]
blk.31.ffn_down.weight
Q6_K
Q6_K
[10240, 2560]
blk.31.ffn_gate.weight
Q4_K
Q4_K
[2560, 10240]
blk.31.ffn_norm.weight
F32
F32
[2560]
blk.31.ffn_up.weight
Q4_K
Q4_K
[2560, 10240]
blk.31.inp_gate.weight
Q4_K
Q4_K
[2560, 256]
blk.31.layer_output_scale.weight
F32
F32
[1]
blk.31.post_attention_norm.weight
F32
F32
[2560]
blk.31.post_ffw_norm.weight
F32
F32
[2560]
blk.31.post_norm.weight
F32
F32
[2560]
blk.31.proj.weight
Q4_K
Q4_K
[256, 2560]
blk.32
blk.32.attn_k.weight
Q4_K
Q4_K
[2560, 512]
blk.32.attn_k_norm.weight
F32
F32
[256]
blk.32.attn_norm.weight
F32
F32
[2560]
blk.32.attn_output.weight
Q4_K
Q4_K
[2048, 2560]
blk.32.attn_q.weight
Q4_K
Q4_K
[2560, 2048]
blk.32.attn_q_norm.weight
F32
F32
[256]
blk.32.attn_v.weight
Q4_K
Q4_K
[2560, 512]
blk.32.ffn_down.weight
Q4_K
Q4_K
[10240, 2560]
blk.32.ffn_gate.weight
Q4_K
Q4_K
[2560, 10240]
blk.32.ffn_norm.weight
F32
F32
[2560]
blk.32.ffn_up.weight
Q4_K
Q4_K
[2560, 10240]
blk.32.inp_gate.weight
Q4_K
Q4_K
[2560, 256]
blk.32.layer_output_scale.weight
F32
F32
[1]
blk.32.post_attention_norm.weight
F32
F32
[2560]
blk.32.post_ffw_norm.weight
F32
F32
[2560]
blk.32.post_norm.weight
F32
F32
[2560]
blk.32.proj.weight
Q4_K
Q4_K
[256, 2560]
blk.33
blk.33.attn_k.weight
Q4_K
Q4_K
[2560, 512]
blk.33.attn_k_norm.weight
F32
F32
[256]
blk.33.attn_norm.weight
F32
F32
[2560]
blk.33.attn_output.weight
Q4_K
Q4_K
[2048, 2560]
blk.33.attn_q.weight
Q4_K
Q4_K
[2560, 2048]
blk.33.attn_q_norm.weight
F32
F32
[256]
blk.33.attn_v.weight
Q4_K
Q4_K
[2560, 512]
blk.33.ffn_down.weight
Q4_K
Q4_K
[10240, 2560]
blk.33.ffn_gate.weight
Q4_K
Q4_K
[2560, 10240]
blk.33.ffn_norm.weight
F32
F32
[2560]
blk.33.ffn_up.weight
Q4_K
Q4_K
[2560, 10240]
blk.33.inp_gate.weight
Q4_K
Q4_K
[2560, 256]
blk.33.layer_output_scale.weight
F32
F32
[1]
blk.33.post_attention_norm.weight
F32
F32
[2560]
blk.33.post_ffw_norm.weight
F32
F32
[2560]
blk.33.post_norm.weight
F32
F32
[2560]
blk.33.proj.weight
Q4_K
Q4_K
[256, 2560]
blk.34
blk.34.attn_k.weight
Q4_K
Q4_K
[2560, 512]
blk.34.attn_k_norm.weight
F32
F32
[256]
blk.34.attn_norm.weight
F32
F32
[2560]
blk.34.attn_output.weight
Q4_K
Q4_K
[2048, 2560]
blk.34.attn_q.weight
Q4_K
Q4_K
[2560, 2048]
blk.34.attn_q_norm.weight
F32
F32
[256]
blk.34.attn_v.weight
Q6_K
Q6_K
[2560, 512]
blk.34.ffn_down.weight
Q6_K
Q6_K
[10240, 2560]
blk.34.ffn_gate.weight
Q4_K
Q4_K
[2560, 10240]
blk.34.ffn_norm.weight
F32
F32
[2560]
blk.34.ffn_up.weight
Q4_K
Q4_K
[2560, 10240]
blk.34.inp_gate.weight
Q4_K
Q4_K
[2560, 256]
blk.34.layer_output_scale.weight
F32
F32
[1]
blk.34.post_attention_norm.weight
F32
F32
[2560]
blk.34.post_ffw_norm.weight
F32
F32
[2560]
blk.34.post_norm.weight
F32
F32
[2560]
blk.34.proj.weight
Q4_K
Q4_K
[256, 2560]
blk.35
blk.35.attn_k.weight
Q4_K
Q4_K
[2560, 1024]
blk.35.attn_k_norm.weight
F32
F32
[512]
blk.35.attn_norm.weight
F32
F32
[2560]
blk.35.attn_output.weight
Q4_K
Q4_K
[4096, 2560]
blk.35.attn_q.weight
Q4_K
Q4_K
[2560, 4096]
blk.35.attn_q_norm.weight
F32
F32
[512]
blk.35.attn_v.weight
Q4_K
Q4_K
[2560, 1024]
blk.35.ffn_down.weight
Q4_K
Q4_K
[10240, 2560]
blk.35.ffn_gate.weight
Q4_K
Q4_K
[2560, 10240]
blk.35.ffn_norm.weight
F32
F32
[2560]
blk.35.ffn_up.weight
Q4_K
Q4_K
[2560, 10240]
blk.35.inp_gate.weight
Q4_K
Q4_K
[2560, 256]
blk.35.layer_output_scale.weight
F32
F32
[1]
blk.35.post_attention_norm.weight
F32
F32
[2560]
blk.35.post_ffw_norm.weight
F32
F32
[2560]
blk.35.post_norm.weight
F32
F32
[2560]
blk.35.proj.weight
Q4_K
Q4_K
[256, 2560]
blk.36
blk.36.attn_k.weight
Q4_K
Q4_K
[2560, 512]
blk.36.attn_k_norm.weight
F32
F32
[256]
blk.36.attn_norm.weight
F32
F32
[2560]
blk.36.attn_output.weight
Q4_K
Q4_K
[2048, 2560]
blk.36.attn_q.weight
Q4_K
Q4_K
[2560, 2048]
blk.36.attn_q_norm.weight
F32
F32
[256]
blk.36.attn_v.weight
Q4_K
Q4_K
[2560, 512]
blk.36.ffn_down.weight
Q6_K
Q6_K
[10240, 2560]
blk.36.ffn_gate.weight
Q4_K
Q4_K
[2560, 10240]
blk.36.ffn_norm.weight
F32
F32
[2560]
blk.36.ffn_up.weight
Q4_K
Q4_K
[2560, 10240]
blk.36.inp_gate.weight
Q4_K
Q4_K
[2560, 256]
blk.36.layer_output_scale.weight
F32
F32
[1]
blk.36.post_attention_norm.weight
F32
F32
[2560]
blk.36.post_ffw_norm.weight
F32
F32
[2560]
blk.36.post_norm.weight
F32
F32
[2560]
blk.36.proj.weight
Q4_K
Q4_K
[256, 2560]
blk.37
blk.37.attn_k.weight
Q4_K
Q4_K
[2560, 512]
blk.37.attn_k_norm.weight
F32
F32
[256]
blk.37.attn_norm.weight
F32
F32
[2560]
blk.37.attn_output.weight
Q4_K
Q4_K
[2048, 2560]
blk.37.attn_q.weight
Q4_K
Q4_K
[2560, 2048]
blk.37.attn_q_norm.weight
F32
F32
[256]
blk.37.attn_v.weight
Q6_K
Q6_K
[2560, 512]
blk.37.ffn_down.weight
Q6_K
Q6_K
[10240, 2560]
blk.37.ffn_gate.weight
Q4_K
Q4_K
[2560, 10240]
blk.37.ffn_norm.weight
F32
F32
[2560]
blk.37.ffn_up.weight
Q4_K
Q4_K
[2560, 10240]
blk.37.inp_gate.weight
Q4_K
Q4_K
[2560, 256]
blk.37.layer_output_scale.weight
F32
F32
[1]
blk.37.post_attention_norm.weight
F32
F32
[2560]
blk.37.post_ffw_norm.weight
F32
F32
[2560]
blk.37.post_norm.weight
F32
F32
[2560]
blk.37.proj.weight
Q4_K
Q4_K
[256, 2560]
blk.38
blk.38.attn_k.weight
Q4_K
Q4_K
[2560, 512]
blk.38.attn_k_norm.weight
F32
F32
[256]
blk.38.attn_norm.weight
F32
F32
[2560]
blk.38.attn_output.weight
Q4_K
Q4_K
[2048, 2560]
blk.38.attn_q.weight
Q4_K
Q4_K
[2560, 2048]
blk.38.attn_q_norm.weight
F32
F32
[256]
blk.38.attn_v.weight
Q4_K
Q4_K
[2560, 512]
blk.38.ffn_down.weight
Q6_K
Q6_K
[10240, 2560]
blk.38.ffn_gate.weight
Q4_K
Q4_K
[2560, 10240]
blk.38.ffn_norm.weight
F32
F32
[2560]
blk.38.ffn_up.weight
Q4_K
Q4_K
[2560, 10240]
blk.38.inp_gate.weight
Q4_K
Q4_K
[2560, 256]
blk.38.layer_output_scale.weight
F32
F32
[1]
blk.38.post_attention_norm.weight
F32
F32
[2560]
blk.38.post_ffw_norm.weight
F32
F32
[2560]
blk.38.post_norm.weight
F32
F32
[2560]
blk.38.proj.weight
Q4_K
Q4_K
[256, 2560]
blk.39
blk.39.attn_k.weight
Q4_K
Q4_K
[2560, 512]
blk.39.attn_k_norm.weight
F32
F32
[256]
blk.39.attn_norm.weight
F32
F32
[2560]
blk.39.attn_output.weight
Q4_K
Q4_K
[2048, 2560]
blk.39.attn_q.weight
Q4_K
Q4_K
[2560, 2048]
blk.39.attn_q_norm.weight
F32
F32
[256]
blk.39.attn_v.weight
Q4_K
Q4_K
[2560, 512]
blk.39.ffn_down.weight
Q6_K
Q6_K
[10240, 2560]
blk.39.ffn_gate.weight
Q4_K
Q4_K
[2560, 10240]
blk.39.ffn_norm.weight
F32
F32
[2560]
blk.39.ffn_up.weight
Q4_K
Q4_K
[2560, 10240]
blk.39.inp_gate.weight
Q4_K
Q4_K
[2560, 256]
blk.39.layer_output_scale.weight
F32
F32
[1]
blk.39.post_attention_norm.weight
F32
F32
[2560]
blk.39.post_ffw_norm.weight
F32
F32
[2560]
blk.39.post_norm.weight
F32
F32
[2560]
blk.39.proj.weight
Q4_K
Q4_K
[256, 2560]
blk.40
blk.40.attn_k.weight
Q4_K
Q4_K
[2560, 512]
blk.40.attn_k_norm.weight
F32
F32
[256]
blk.40.attn_norm.weight
F32
F32
[2560]
blk.40.attn_output.weight
Q4_K
Q4_K
[2048, 2560]
blk.40.attn_q.weight
Q4_K
Q4_K
[2560, 2048]
blk.40.attn_q_norm.weight
F32
F32
[256]
blk.40.attn_v.weight
Q6_K
Q6_K
[2560, 512]
blk.40.ffn_down.weight
Q6_K
Q6_K
[10240, 2560]
blk.40.ffn_gate.weight
Q4_K
Q4_K
[2560, 10240]
blk.40.ffn_norm.weight
F32
F32
[2560]
blk.40.ffn_up.weight
Q4_K
Q4_K
[2560, 10240]
blk.40.inp_gate.weight
Q4_K
Q4_K
[2560, 256]
blk.40.layer_output_scale.weight
F32
F32
[1]
blk.40.post_attention_norm.weight
F32
F32
[2560]
blk.40.post_ffw_norm.weight
F32
F32
[2560]
blk.40.post_norm.weight
F32
F32
[2560]
blk.40.proj.weight
Q4_K
Q4_K
[256, 2560]
blk.41
blk.41.attn_k.weight
Q4_K
Q4_K
[2560, 1024]
blk.41.attn_k_norm.weight
F32
F32
[512]
blk.41.attn_norm.weight
F32
F32
[2560]
blk.41.attn_output.weight
Q4_K
Q4_K
[4096, 2560]
blk.41.attn_q.weight
Q4_K
Q4_K
[2560, 4096]
blk.41.attn_q_norm.weight
F32
F32
[512]
blk.41.attn_v.weight
Q4_K
Q4_K
[2560, 1024]
blk.41.ffn_down.weight
Q6_K
Q6_K
[10240, 2560]
blk.41.ffn_gate.weight
Q4_K
Q4_K
[2560, 10240]
blk.41.ffn_norm.weight
F32
F32
[2560]
blk.41.ffn_up.weight
Q4_K
Q4_K
[2560, 10240]
blk.41.inp_gate.weight
Q4_K
Q4_K
[2560, 256]
blk.41.layer_output_scale.weight
F32
F32
[1]
blk.41.post_attention_norm.weight
F32
F32
[2560]
blk.41.post_ffw_norm.weight
F32
F32
[2560]
blk.41.post_norm.weight
F32
F32
[2560]
blk.41.proj.weight
Q4_K
Q4_K
[256, 2560]
mm.a.fc.bias
F32
F32
[1536]
mm.a.fc.weight
F16
F16
[1024, 1536]
mm.a.input_projection.weight
F16
F16
[1536, 2560]
mm.input_projection.weight
F16
F16
[768, 2560]
per_layer_model_proj.weight
Q4_K
Q4_K
[2560, 10752]
per_layer_proj_norm.weight
F32
F32
[256]
per_layer_token_embd.weight
BF16
BF16
[10752, 262144]
rope_freqs.weight
F32
F32
[256]
v.blk.0
v.blk.0.attn_k.input_max
F32
F32
[1]
v.blk.0.attn_k.input_min
F32
F32
[1]
v.blk.0.attn_k.output_max
F32
F32
[1]
v.blk.0.attn_k.output_min
F32
F32
[1]
v.blk.0.attn_k.weight
F16
F16
[768, 768]
v.blk.0.attn_k_norm.weight
F32
F32
[64]
v.blk.0.attn_out.input_max
F32
F32
[1]
v.blk.0.attn_out.input_min
F32
F32
[1]
v.blk.0.attn_out.output_max
F32
F32
[1]
v.blk.0.attn_out.output_min
F32
F32
[1]
v.blk.0.attn_out.weight
F16
F16
[768, 768]
v.blk.0.attn_post_norm.weight
F32
F32
[768]
v.blk.0.attn_q.input_max
F32
F32
[1]
v.blk.0.attn_q.input_min
F32
F32
[1]
v.blk.0.attn_q.output_max
F32
F32
[1]
v.blk.0.attn_q.output_min
F32
F32
[1]
v.blk.0.attn_q.weight
F16
F16
[768, 768]
v.blk.0.attn_q_norm.weight
F32
F32
[64]
v.blk.0.attn_v.input_max
F32
F32
[1]
v.blk.0.attn_v.input_min
F32
F32
[1]
v.blk.0.attn_v.output_max
F32
F32
[1]
v.blk.0.attn_v.output_min
F32
F32
[1]
v.blk.0.attn_v.weight
F16
F16
[768, 768]
v.blk.0.ffn_down.input_max
F32
F32
[1]
v.blk.0.ffn_down.input_min
F32
F32
[1]
v.blk.0.ffn_down.output_max
F32
F32
[1]
v.blk.0.ffn_down.output_min
F32
F32
[1]
v.blk.0.ffn_down.weight
F16
F16
[3072, 768]
v.blk.0.ffn_gate.input_max
F32
F32
[1]
v.blk.0.ffn_gate.input_min
F32
F32
[1]
v.blk.0.ffn_gate.output_max
F32
F32
[1]
v.blk.0.ffn_gate.output_min
F32
F32
[1]
v.blk.0.ffn_gate.weight
F16
F16
[768, 3072]
v.blk.0.ffn_post_norm.weight
F32
F32
[768]
v.blk.0.ffn_up.input_max
F32
F32
[1]
v.blk.0.ffn_up.input_min
F32
F32
[1]
v.blk.0.ffn_up.output_max
F32
F32
[1]
v.blk.0.ffn_up.output_min
F32
F32
[1]
v.blk.0.ffn_up.weight
F16
F16
[768, 3072]
v.blk.0.ln1.weight
F32
F32
[768]
v.blk.0.ln2.weight
F32
F32
[768]
v.blk.1
v.blk.1.attn_k.input_max
F32
F32
[1]
v.blk.1.attn_k.input_min
F32
F32
[1]
v.blk.1.attn_k.output_max
F32
F32
[1]
v.blk.1.attn_k.output_min
F32
F32
[1]
v.blk.1.attn_k.weight
F16
F16
[768, 768]
v.blk.1.attn_k_norm.weight
F32
F32
[64]
v.blk.1.attn_out.input_max
F32
F32
[1]
v.blk.1.attn_out.input_min
F32
F32
[1]
v.blk.1.attn_out.output_max
F32
F32
[1]
v.blk.1.attn_out.output_min
F32
F32
[1]
v.blk.1.attn_out.weight
F16
F16
[768, 768]
v.blk.1.attn_post_norm.weight
F32
F32
[768]
v.blk.1.attn_q.input_max
F32
F32
[1]
v.blk.1.attn_q.input_min
F32
F32
[1]
v.blk.1.attn_q.output_max
F32
F32
[1]
v.blk.1.attn_q.output_min
F32
F32
[1]
v.blk.1.attn_q.weight
F16
F16
[768, 768]
v.blk.1.attn_q_norm.weight
F32
F32
[64]
v.blk.1.attn_v.input_max
F32
F32
[1]
v.blk.1.attn_v.input_min
F32
F32
[1]
v.blk.1.attn_v.output_max
F32
F32
[1]
v.blk.1.attn_v.output_min
F32
F32
[1]
v.blk.1.attn_v.weight
F16
F16
[768, 768]
v.blk.1.ffn_down.input_max
F32
F32
[1]
v.blk.1.ffn_down.input_min
F32
F32
[1]
v.blk.1.ffn_down.output_max
F32
F32
[1]
v.blk.1.ffn_down.output_min
F32
F32
[1]
v.blk.1.ffn_down.weight
F16
F16
[3072, 768]
v.blk.1.ffn_gate.input_max
F32
F32
[1]
v.blk.1.ffn_gate.input_min
F32
F32
[1]
v.blk.1.ffn_gate.output_max
F32
F32
[1]
v.blk.1.ffn_gate.output_min
F32
F32
[1]
v.blk.1.ffn_gate.weight
F16
F16
[768, 3072]
v.blk.1.ffn_post_norm.weight
F32
F32
[768]
v.blk.1.ffn_up.input_max
F32
F32
[1]
v.blk.1.ffn_up.input_min
F32
F32
[1]
v.blk.1.ffn_up.output_max
F32
F32
[1]
v.blk.1.ffn_up.output_min
F32
F32
[1]
v.blk.1.ffn_up.weight
F16
F16
[768, 3072]
v.blk.1.ln1.weight
F32
F32
[768]
v.blk.1.ln2.weight
F32
F32
[768]
v.blk.2
v.blk.2.attn_k.input_max
F32
F32
[1]
v.blk.2.attn_k.input_min
F32
F32
[1]
v.blk.2.attn_k.output_max
F32
F32
[1]
v.blk.2.attn_k.output_min
F32
F32
[1]
v.blk.2.attn_k.weight
F16
F16
[768, 768]
v.blk.2.attn_k_norm.weight
F32
F32
[64]
v.blk.2.attn_out.input_max
F32
F32
[1]
v.blk.2.attn_out.input_min
F32
F32
[1]
v.blk.2.attn_out.output_max
F32
F32
[1]
v.blk.2.attn_out.output_min
F32
F32
[1]
v.blk.2.attn_out.weight
F16
F16
[768, 768]
v.blk.2.attn_post_norm.weight
F32
F32
[768]
v.blk.2.attn_q.input_max
F32
F32
[1]
v.blk.2.attn_q.input_min
F32
F32
[1]
v.blk.2.attn_q.output_max
F32
F32
[1]
v.blk.2.attn_q.output_min
F32
F32
[1]
v.blk.2.attn_q.weight
F16
F16
[768, 768]
v.blk.2.attn_q_norm.weight
F32
F32
[64]
v.blk.2.attn_v.input_max
F32
F32
[1]
v.blk.2.attn_v.input_min
F32
F32
[1]
v.blk.2.attn_v.output_max
F32
F32
[1]
v.blk.2.attn_v.output_min
F32
F32
[1]
v.blk.2.attn_v.weight
F16
F16
[768, 768]
v.blk.2.ffn_down.input_max
F32
F32
[1]
v.blk.2.ffn_down.input_min
F32
F32
[1]
v.blk.2.ffn_down.output_max
F32
F32
[1]
v.blk.2.ffn_down.output_min
F32
F32
[1]
v.blk.2.ffn_down.weight
F16
F16
[3072, 768]
v.blk.2.ffn_gate.input_max
F32
F32
[1]
v.blk.2.ffn_gate.input_min
F32
F32
[1]
v.blk.2.ffn_gate.output_max
F32
F32
[1]
v.blk.2.ffn_gate.output_min
F32
F32
[1]
v.blk.2.ffn_gate.weight
F16
F16
[768, 3072]
v.blk.2.ffn_post_norm.weight
F32
F32
[768]
v.blk.2.ffn_up.input_max
F32
F32
[1]
v.blk.2.ffn_up.input_min
F32
F32
[1]
v.blk.2.ffn_up.output_max
F32
F32
[1]
v.blk.2.ffn_up.output_min
F32
F32
[1]
v.blk.2.ffn_up.weight
F16
F16
[768, 3072]
v.blk.2.ln1.weight
F32
F32
[768]
v.blk.2.ln2.weight
F32
F32
[768]
v.blk.3
v.blk.3.attn_k.input_max
F32
F32
[1]
v.blk.3.attn_k.input_min
F32
F32
[1]
v.blk.3.attn_k.output_max
F32
F32
[1]
v.blk.3.attn_k.output_min
F32
F32
[1]
v.blk.3.attn_k.weight
F16
F16
[768, 768]
v.blk.3.attn_k_norm.weight
F32
F32
[64]
v.blk.3.attn_out.input_max
F32
F32
[1]
v.blk.3.attn_out.input_min
F32
F32
[1]
v.blk.3.attn_out.output_max
F32
F32
[1]
v.blk.3.attn_out.output_min
F32
F32
[1]
v.blk.3.attn_out.weight
F16
F16
[768, 768]
v.blk.3.attn_post_norm.weight
F32
F32
[768]
v.blk.3.attn_q.input_max
F32
F32
[1]
v.blk.3.attn_q.input_min
F32
F32
[1]
v.blk.3.attn_q.output_max
F32
F32
[1]
v.blk.3.attn_q.output_min
F32
F32
[1]
v.blk.3.attn_q.weight
F16
F16
[768, 768]
v.blk.3.attn_q_norm.weight
F32
F32
[64]
v.blk.3.attn_v.input_max
F32
F32
[1]
v.blk.3.attn_v.input_min
F32
F32
[1]
v.blk.3.attn_v.output_max
F32
F32
[1]
v.blk.3.attn_v.output_min
F32
F32
[1]
v.blk.3.attn_v.weight
F16
F16
[768, 768]
v.blk.3.ffn_down.input_max
F32
F32
[1]
v.blk.3.ffn_down.input_min
F32
F32
[1]
v.blk.3.ffn_down.output_max
F32
F32
[1]
v.blk.3.ffn_down.output_min
F32
F32
[1]
v.blk.3.ffn_down.weight
F16
F16
[3072, 768]
v.blk.3.ffn_gate.input_max
F32
F32
[1]
v.blk.3.ffn_gate.input_min
F32
F32
[1]
v.blk.3.ffn_gate.output_max
F32
F32
[1]
v.blk.3.ffn_gate.output_min
F32
F32
[1]
v.blk.3.ffn_gate.weight
F16
F16
[768, 3072]
v.blk.3.ffn_post_norm.weight
F32
F32
[768]
v.blk.3.ffn_up.input_max
F32
F32
[1]
v.blk.3.ffn_up.input_min
F32
F32
[1]
v.blk.3.ffn_up.output_max
F32
F32
[1]
v.blk.3.ffn_up.output_min
F32
F32
[1]
v.blk.3.ffn_up.weight
F16
F16
[768, 3072]
v.blk.3.ln1.weight
F32
F32
[768]
v.blk.3.ln2.weight
F32
F32
[768]
v.blk.4
v.blk.4.attn_k.input_max
F32
F32
[1]
v.blk.4.attn_k.input_min
F32
F32
[1]
v.blk.4.attn_k.output_max
F32
F32
[1]
v.blk.4.attn_k.output_min
F32
F32
[1]
v.blk.4.attn_k.weight
F16
F16
[768, 768]
v.blk.4.attn_k_norm.weight
F32
F32
[64]
v.blk.4.attn_out.input_max
F32
F32
[1]
v.blk.4.attn_out.input_min
F32
F32
[1]
v.blk.4.attn_out.output_max
F32
F32
[1]
v.blk.4.attn_out.output_min
F32
F32
[1]
v.blk.4.attn_out.weight
F16
F16
[768, 768]
v.blk.4.attn_post_norm.weight
F32
F32
[768]
v.blk.4.attn_q.input_max
F32
F32
[1]
v.blk.4.attn_q.input_min
F32
F32
[1]
v.blk.4.attn_q.output_max
F32
F32
[1]
v.blk.4.attn_q.output_min
F32
F32
[1]
v.blk.4.attn_q.weight
F16
F16
[768, 768]
v.blk.4.attn_q_norm.weight
F32
F32
[64]
v.blk.4.attn_v.input_max
F32
F32
[1]
v.blk.4.attn_v.input_min
F32
F32
[1]
v.blk.4.attn_v.output_max
F32
F32
[1]
v.blk.4.attn_v.output_min
F32
F32
[1]
v.blk.4.attn_v.weight
F16
F16
[768, 768]
v.blk.4.ffn_down.input_max
F32
F32
[1]
v.blk.4.ffn_down.input_min
F32
F32
[1]
v.blk.4.ffn_down.output_max
F32
F32
[1]
v.blk.4.ffn_down.output_min
F32
F32
[1]
v.blk.4.ffn_down.weight
F16
F16
[3072, 768]
v.blk.4.ffn_gate.input_max
F32
F32
[1]
v.blk.4.ffn_gate.input_min
F32
F32
[1]
v.blk.4.ffn_gate.output_max
F32
F32
[1]
v.blk.4.ffn_gate.output_min
F32
F32
[1]
v.blk.4.ffn_gate.weight
F16
F16
[768, 3072]
v.blk.4.ffn_post_norm.weight
F32
F32
[768]
v.blk.4.ffn_up.input_max
F32
F32
[1]
v.blk.4.ffn_up.input_min
F32
F32
[1]
v.blk.4.ffn_up.output_max
F32
F32
[1]
v.blk.4.ffn_up.output_min
F32
F32
[1]
v.blk.4.ffn_up.weight
F16
F16
[768, 3072]
v.blk.4.ln1.weight
F32
F32
[768]
v.blk.4.ln2.weight
F32
F32
[768]
v.blk.5
v.blk.5.attn_k.input_max
F32
F32
[1]
v.blk.5.attn_k.input_min
F32
F32
[1]
v.blk.5.attn_k.output_max
F32
F32
[1]
v.blk.5.attn_k.output_min
F32
F32
[1]
v.blk.5.attn_k.weight
F16
F16
[768, 768]
v.blk.5.attn_k_norm.weight
F32
F32
[64]
v.blk.5.attn_out.input_max
F32
F32
[1]
v.blk.5.attn_out.input_min
F32
F32
[1]
v.blk.5.attn_out.output_max
F32
F32
[1]
v.blk.5.attn_out.output_min
F32
F32
[1]
v.blk.5.attn_out.weight
F16
F16
[768, 768]
v.blk.5.attn_post_norm.weight
F32
F32
[768]
v.blk.5.attn_q.input_max
F32
F32
[1]
v.blk.5.attn_q.input_min
F32
F32
[1]
v.blk.5.attn_q.output_max
F32
F32
[1]
v.blk.5.attn_q.output_min
F32
F32
[1]
v.blk.5.attn_q.weight
F16
F16
[768, 768]
v.blk.5.attn_q_norm.weight
F32
F32
[64]
v.blk.5.attn_v.input_max
F32
F32
[1]
v.blk.5.attn_v.input_min
F32
F32
[1]
v.blk.5.attn_v.output_max
F32
F32
[1]
v.blk.5.attn_v.output_min
F32
F32
[1]
v.blk.5.attn_v.weight
F16
F16
[768, 768]
v.blk.5.ffn_down.input_max
F32
F32
[1]
v.blk.5.ffn_down.input_min
F32
F32
[1]
v.blk.5.ffn_down.output_max
F32
F32
[1]
v.blk.5.ffn_down.output_min
F32
F32
[1]
v.blk.5.ffn_down.weight
F16
F16
[3072, 768]
v.blk.5.ffn_gate.input_max
F32
F32
[1]
v.blk.5.ffn_gate.input_min
F32
F32
[1]
v.blk.5.ffn_gate.output_max
F32
F32
[1]
v.blk.5.ffn_gate.output_min
F32
F32
[1]
v.blk.5.ffn_gate.weight
F16
F16
[768, 3072]
v.blk.5.ffn_post_norm.weight
F32
F32
[768]
v.blk.5.ffn_up.input_max
F32
F32
[1]
v.blk.5.ffn_up.input_min
F32
F32
[1]
v.blk.5.ffn_up.output_max
F32
F32
[1]
v.blk.5.ffn_up.output_min
F32
F32
[1]
v.blk.5.ffn_up.weight
F16
F16
[768, 3072]
v.blk.5.ln1.weight
F32
F32
[768]
v.blk.5.ln2.weight
F32
F32
[768]
v.blk.6
v.blk.6.attn_k.input_max
F32
F32
[1]
v.blk.6.attn_k.input_min
F32
F32
[1]
v.blk.6.attn_k.output_max
F32
F32
[1]
v.blk.6.attn_k.output_min
F32
F32
[1]
v.blk.6.attn_k.weight
F16
F16
[768, 768]
v.blk.6.attn_k_norm.weight
F32
F32
[64]
v.blk.6.attn_out.input_max
F32
F32
[1]
v.blk.6.attn_out.input_min
F32
F32
[1]
v.blk.6.attn_out.output_max
F32
F32
[1]
v.blk.6.attn_out.output_min
F32
F32
[1]
v.blk.6.attn_out.weight
F16
F16
[768, 768]
v.blk.6.attn_post_norm.weight
F32
F32
[768]
v.blk.6.attn_q.input_max
F32
F32
[1]
v.blk.6.attn_q.input_min
F32
F32
[1]
v.blk.6.attn_q.output_max
F32
F32
[1]
v.blk.6.attn_q.output_min
F32
F32
[1]
v.blk.6.attn_q.weight
F16
F16
[768, 768]
v.blk.6.attn_q_norm.weight
F32
F32
[64]
v.blk.6.attn_v.input_max
F32
F32
[1]
v.blk.6.attn_v.input_min
F32
F32
[1]
v.blk.6.attn_v.output_max
F32
F32
[1]
v.blk.6.attn_v.output_min
F32
F32
[1]
v.blk.6.attn_v.weight
F16
F16
[768, 768]
v.blk.6.ffn_down.input_max
F32
F32
[1]
v.blk.6.ffn_down.input_min
F32
F32
[1]
v.blk.6.ffn_down.output_max
F32
F32
[1]
v.blk.6.ffn_down.output_min
F32
F32
[1]
v.blk.6.ffn_down.weight
F16
F16
[3072, 768]
v.blk.6.ffn_gate.input_max
F32
F32
[1]
v.blk.6.ffn_gate.input_min
F32
F32
[1]
v.blk.6.ffn_gate.output_max
F32
F32
[1]
v.blk.6.ffn_gate.output_min
F32
F32
[1]
v.blk.6.ffn_gate.weight
F16
F16
[768, 3072]
v.blk.6.ffn_post_norm.weight
F32
F32
[768]
v.blk.6.ffn_up.input_max
F32
F32
[1]
v.blk.6.ffn_up.input_min
F32
F32
[1]
v.blk.6.ffn_up.output_max
F32
F32
[1]
v.blk.6.ffn_up.output_min
F32
F32
[1]
v.blk.6.ffn_up.weight
F16
F16
[768, 3072]
v.blk.6.ln1.weight
F32
F32
[768]
v.blk.6.ln2.weight
F32
F32
[768]
v.blk.7
v.blk.7.attn_k.input_max
F32
F32
[1]
v.blk.7.attn_k.input_min
F32
F32
[1]
v.blk.7.attn_k.output_max
F32
F32
[1]
v.blk.7.attn_k.output_min
F32
F32
[1]
v.blk.7.attn_k.weight
F16
F16
[768, 768]
v.blk.7.attn_k_norm.weight
F32
F32
[64]
v.blk.7.attn_out.input_max
F32
F32
[1]
v.blk.7.attn_out.input_min
F32
F32
[1]
v.blk.7.attn_out.output_max
F32
F32
[1]
v.blk.7.attn_out.output_min
F32
F32
[1]
v.blk.7.attn_out.weight
F16
F16
[768, 768]
v.blk.7.attn_post_norm.weight
F32
F32
[768]
v.blk.7.attn_q.input_max
F32
F32
[1]
v.blk.7.attn_q.input_min
F32
F32
[1]
v.blk.7.attn_q.output_max
F32
F32
[1]
v.blk.7.attn_q.output_min
F32
F32
[1]
v.blk.7.attn_q.weight
F16
F16
[768, 768]
v.blk.7.attn_q_norm.weight
F32
F32
[64]
v.blk.7.attn_v.input_max
F32
F32
[1]
v.blk.7.attn_v.input_min
F32
F32
[1]
v.blk.7.attn_v.output_max
F32
F32
[1]
v.blk.7.attn_v.output_min
F32
F32
[1]
v.blk.7.attn_v.weight
F16
F16
[768, 768]
v.blk.7.ffn_down.input_max
F32
F32
[1]
v.blk.7.ffn_down.input_min
F32
F32
[1]
v.blk.7.ffn_down.output_max
F32
F32
[1]
v.blk.7.ffn_down.output_min
F32
F32
[1]
v.blk.7.ffn_down.weight
F16
F16
[3072, 768]
v.blk.7.ffn_gate.input_max
F32
F32
[1]
v.blk.7.ffn_gate.input_min
F32
F32
[1]
v.blk.7.ffn_gate.output_max
F32
F32
[1]
v.blk.7.ffn_gate.output_min
F32
F32
[1]
v.blk.7.ffn_gate.weight
F16
F16
[768, 3072]
v.blk.7.ffn_post_norm.weight
F32
F32
[768]
v.blk.7.ffn_up.input_max
F32
F32
[1]
v.blk.7.ffn_up.input_min
F32
F32
[1]
v.blk.7.ffn_up.output_max
F32
F32
[1]
v.blk.7.ffn_up.output_min
F32
F32
[1]
v.blk.7.ffn_up.weight
F16
F16
[768, 3072]
v.blk.7.ln1.weight
F32
F32
[768]
v.blk.7.ln2.weight
F32
F32
[768]
v.blk.8
v.blk.8.attn_k.input_max
F32
F32
[1]
v.blk.8.attn_k.input_min
F32
F32
[1]
v.blk.8.attn_k.output_max
F32
F32
[1]
v.blk.8.attn_k.output_min
F32
F32
[1]
v.blk.8.attn_k.weight
F16
F16
[768, 768]
v.blk.8.attn_k_norm.weight
F32
F32
[64]
v.blk.8.attn_out.input_max
F32
F32
[1]
v.blk.8.attn_out.input_min
F32
F32
[1]
v.blk.8.attn_out.output_max
F32
F32
[1]
v.blk.8.attn_out.output_min
F32
F32
[1]
v.blk.8.attn_out.weight
F16
F16
[768, 768]
v.blk.8.attn_post_norm.weight
F32
F32
[768]
v.blk.8.attn_q.input_max
F32
F32
[1]
v.blk.8.attn_q.input_min
F32
F32
[1]
v.blk.8.attn_q.output_max
F32
F32
[1]
v.blk.8.attn_q.output_min
F32
F32
[1]
v.blk.8.attn_q.weight
F16
F16
[768, 768]
v.blk.8.attn_q_norm.weight
F32
F32
[64]
v.blk.8.attn_v.input_max
F32
F32
[1]
v.blk.8.attn_v.input_min
F32
F32
[1]
v.blk.8.attn_v.output_max
F32
F32
[1]
v.blk.8.attn_v.output_min
F32
F32
[1]
v.blk.8.attn_v.weight
F16
F16
[768, 768]
v.blk.8.ffn_down.input_max
F32
F32
[1]
v.blk.8.ffn_down.input_min
F32
F32
[1]
v.blk.8.ffn_down.output_max
F32
F32
[1]
v.blk.8.ffn_down.output_min
F32
F32
[1]
v.blk.8.ffn_down.weight
F16
F16
[3072, 768]
v.blk.8.ffn_gate.input_max
F32
F32
[1]
v.blk.8.ffn_gate.input_min
F32
F32
[1]
v.blk.8.ffn_gate.output_max
F32
F32
[1]
v.blk.8.ffn_gate.output_min
F32
F32
[1]
v.blk.8.ffn_gate.weight
F16
F16
[768, 3072]
v.blk.8.ffn_post_norm.weight
F32
F32
[768]
v.blk.8.ffn_up.input_max
F32
F32
[1]
v.blk.8.ffn_up.input_min
F32
F32
[1]
v.blk.8.ffn_up.output_max
F32
F32
[1]
v.blk.8.ffn_up.output_min
F32
F32
[1]
v.blk.8.ffn_up.weight
F16
F16
[768, 3072]
v.blk.8.ln1.weight
F32
F32
[768]
v.blk.8.ln2.weight
F32
F32
[768]
v.blk.9
v.blk.9.attn_k.input_max
F32
F32
[1]
v.blk.9.attn_k.input_min
F32
F32
[1]
v.blk.9.attn_k.output_max
F32
F32
[1]
v.blk.9.attn_k.output_min
F32
F32
[1]
v.blk.9.attn_k.weight
F16
F16
[768, 768]
v.blk.9.attn_k_norm.weight
F32
F32
[64]
v.blk.9.attn_out.input_max
F32
F32
[1]
v.blk.9.attn_out.input_min
F32
F32
[1]
v.blk.9.attn_out.output_max
F32
F32
[1]
v.blk.9.attn_out.output_min
F32
F32
[1]
v.blk.9.attn_out.weight
F16
F16
[768, 768]
v.blk.9.attn_post_norm.weight
F32
F32
[768]
v.blk.9.attn_q.input_max
F32
F32
[1]
v.blk.9.attn_q.input_min
F32
F32
[1]
v.blk.9.attn_q.output_max
F32
F32
[1]
v.blk.9.attn_q.output_min
F32
F32
[1]
v.blk.9.attn_q.weight
F16
F16
[768, 768]
v.blk.9.attn_q_norm.weight
F32
F32
[64]
v.blk.9.attn_v.input_max
F32
F32
[1]
v.blk.9.attn_v.input_min
F32
F32
[1]
v.blk.9.attn_v.output_max
F32
F32
[1]
v.blk.9.attn_v.output_min
F32
F32
[1]
v.blk.9.attn_v.weight
F16
F16
[768, 768]
v.blk.9.ffn_down.input_max
F32
F32
[1]
v.blk.9.ffn_down.input_min
F32
F32
[1]
v.blk.9.ffn_down.output_max
F32
F32
[1]
v.blk.9.ffn_down.output_min
F32
F32
[1]
v.blk.9.ffn_down.weight
F16
F16
[3072, 768]
v.blk.9.ffn_gate.input_max
F32
F32
[1]
v.blk.9.ffn_gate.input_min
F32
F32
[1]
v.blk.9.ffn_gate.output_max
F32
F32
[1]
v.blk.9.ffn_gate.output_min
F32
F32
[1]
v.blk.9.ffn_gate.weight
F16
F16
[768, 3072]
v.blk.9.ffn_post_norm.weight
F32
F32
[768]
v.blk.9.ffn_up.input_max
F32
F32
[1]
v.blk.9.ffn_up.input_min
F32
F32
[1]
v.blk.9.ffn_up.output_max
F32
F32
[1]
v.blk.9.ffn_up.output_min
F32
F32
[1]
v.blk.9.ffn_up.weight
F16
F16
[768, 3072]
v.blk.9.ln1.weight
F32
F32
[768]
v.blk.9.ln2.weight
F32
F32
[768]
v.blk.10
v.blk.10.attn_k.input_max
F32
F32
[1]
v.blk.10.attn_k.input_min
F32
F32
[1]
v.blk.10.attn_k.output_max
F32
F32
[1]
v.blk.10.attn_k.output_min
F32
F32
[1]
v.blk.10.attn_k.weight
F16
F16
[768, 768]
v.blk.10.attn_k_norm.weight
F32
F32
[64]
v.blk.10.attn_out.input_max
F32
F32
[1]
v.blk.10.attn_out.input_min
F32
F32
[1]
v.blk.10.attn_out.output_max
F32
F32
[1]
v.blk.10.attn_out.output_min
F32
F32
[1]
v.blk.10.attn_out.weight
F16
F16
[768, 768]
v.blk.10.attn_post_norm.weight
F32
F32
[768]
v.blk.10.attn_q.input_max
F32
F32
[1]
v.blk.10.attn_q.input_min
F32
F32
[1]
v.blk.10.attn_q.output_max
F32
F32
[1]
v.blk.10.attn_q.output_min
F32
F32
[1]
v.blk.10.attn_q.weight
F16
F16
[768, 768]
v.blk.10.attn_q_norm.weight
F32
F32
[64]
v.blk.10.attn_v.input_max
F32
F32
[1]
v.blk.10.attn_v.input_min
F32
F32
[1]
v.blk.10.attn_v.output_max
F32
F32
[1]
v.blk.10.attn_v.output_min
F32
F32
[1]
v.blk.10.attn_v.weight
F16
F16
[768, 768]
v.blk.10.ffn_down.input_max
F32
F32
[1]
v.blk.10.ffn_down.input_min
F32
F32
[1]
v.blk.10.ffn_down.output_max
F32
F32
[1]
v.blk.10.ffn_down.output_min
F32
F32
[1]
v.blk.10.ffn_down.weight
F16
F16
[3072, 768]
v.blk.10.ffn_gate.input_max
F32
F32
[1]
v.blk.10.ffn_gate.input_min
F32
F32
[1]
v.blk.10.ffn_gate.output_max
F32
F32
[1]
v.blk.10.ffn_gate.output_min
F32
F32
[1]
v.blk.10.ffn_gate.weight
F16
F16
[768, 3072]
v.blk.10.ffn_post_norm.weight
F32
F32
[768]
v.blk.10.ffn_up.input_max
F32
F32
[1]
v.blk.10.ffn_up.input_min
F32
F32
[1]
v.blk.10.ffn_up.output_max
F32
F32
[1]
v.blk.10.ffn_up.output_min
F32
F32
[1]
v.blk.10.ffn_up.weight
F16
F16
[768, 3072]
v.blk.10.ln1.weight
F32
F32
[768]
v.blk.10.ln2.weight
F32
F32
[768]
v.blk.11
v.blk.11.attn_k.input_max
F32
F32
[1]
v.blk.11.attn_k.input_min
F32
F32
[1]
v.blk.11.attn_k.output_max
F32
F32
[1]
v.blk.11.attn_k.output_min
F32
F32
[1]
v.blk.11.attn_k.weight
F16
F16
[768, 768]
v.blk.11.attn_k_norm.weight
F32
F32
[64]
v.blk.11.attn_out.input_max
F32
F32
[1]
v.blk.11.attn_out.input_min
F32
F32
[1]
v.blk.11.attn_out.output_max
F32
F32
[1]
v.blk.11.attn_out.output_min
F32
F32
[1]
v.blk.11.attn_out.weight
F16
F16
[768, 768]
v.blk.11.attn_post_norm.weight
F32
F32
[768]
v.blk.11.attn_q.input_max
F32
F32
[1]
v.blk.11.attn_q.input_min
F32
F32
[1]
v.blk.11.attn_q.output_max
F32
F32
[1]
v.blk.11.attn_q.output_min
F32
F32
[1]
v.blk.11.attn_q.weight
F16
F16
[768, 768]
v.blk.11.attn_q_norm.weight
F32
F32
[64]
v.blk.11.attn_v.input_max
F32
F32
[1]
v.blk.11.attn_v.input_min
F32
F32
[1]
v.blk.11.attn_v.output_max
F32
F32
[1]
v.blk.11.attn_v.output_min
F32
F32
[1]
v.blk.11.attn_v.weight
F16
F16
[768, 768]
v.blk.11.ffn_down.input_max
F32
F32
[1]
v.blk.11.ffn_down.input_min
F32
F32
[1]
v.blk.11.ffn_down.output_max
F32
F32
[1]
v.blk.11.ffn_down.output_min
F32
F32
[1]
v.blk.11.ffn_down.weight
F16
F16
[3072, 768]
v.blk.11.ffn_gate.input_max
F32
F32
[1]
v.blk.11.ffn_gate.input_min
F32
F32
[1]
v.blk.11.ffn_gate.output_max
F32
F32
[1]
v.blk.11.ffn_gate.output_min
F32
F32
[1]
v.blk.11.ffn_gate.weight
F16
F16
[768, 3072]
v.blk.11.ffn_post_norm.weight
F32
F32
[768]
v.blk.11.ffn_up.input_max
F32
F32
[1]
v.blk.11.ffn_up.input_min
F32
F32
[1]
v.blk.11.ffn_up.output_max
F32
F32
[1]
v.blk.11.ffn_up.output_min
F32
F32
[1]
v.blk.11.ffn_up.weight
F16
F16
[768, 3072]
v.blk.11.ln1.weight
F32
F32
[768]
v.blk.11.ln2.weight
F32
F32
[768]
v.blk.12
v.blk.12.attn_k.input_max
F32
F32
[1]
v.blk.12.attn_k.input_min
F32
F32
[1]
v.blk.12.attn_k.output_max
F32
F32
[1]
v.blk.12.attn_k.output_min
F32
F32
[1]
v.blk.12.attn_k.weight
F16
F16
[768, 768]
v.blk.12.attn_k_norm.weight
F32
F32
[64]
v.blk.12.attn_out.input_max
F32
F32
[1]
v.blk.12.attn_out.input_min
F32
F32
[1]
v.blk.12.attn_out.output_max
F32
F32
[1]
v.blk.12.attn_out.output_min
F32
F32
[1]
v.blk.12.attn_out.weight
F16
F16
[768, 768]
v.blk.12.attn_post_norm.weight
F32
F32
[768]
v.blk.12.attn_q.input_max
F32
F32
[1]
v.blk.12.attn_q.input_min
F32
F32
[1]
v.blk.12.attn_q.output_max
F32
F32
[1]
v.blk.12.attn_q.output_min
F32
F32
[1]
v.blk.12.attn_q.weight
F16
F16
[768, 768]
v.blk.12.attn_q_norm.weight
F32
F32
[64]
v.blk.12.attn_v.input_max
F32
F32
[1]
v.blk.12.attn_v.input_min
F32
F32
[1]
v.blk.12.attn_v.output_max
F32
F32
[1]
v.blk.12.attn_v.output_min
F32
F32
[1]
v.blk.12.attn_v.weight
F16
F16
[768, 768]
v.blk.12.ffn_down.input_max
F32
F32
[1]
v.blk.12.ffn_down.input_min
F32
F32
[1]
v.blk.12.ffn_down.output_max
F32
F32
[1]
v.blk.12.ffn_down.output_min
F32
F32
[1]
v.blk.12.ffn_down.weight
F16
F16
[3072, 768]
v.blk.12.ffn_gate.input_max
F32
F32
[1]
v.blk.12.ffn_gate.input_min
F32
F32
[1]
v.blk.12.ffn_gate.output_max
F32
F32
[1]
v.blk.12.ffn_gate.output_min
F32
F32
[1]
v.blk.12.ffn_gate.weight
F16
F16
[768, 3072]
v.blk.12.ffn_post_norm.weight
F32
F32
[768]
v.blk.12.ffn_up.input_max
F32
F32
[1]
v.blk.12.ffn_up.input_min
F32
F32
[1]
v.blk.12.ffn_up.output_max
F32
F32
[1]
v.blk.12.ffn_up.output_min
F32
F32
[1]
v.blk.12.ffn_up.weight
F16
F16
[768, 3072]
v.blk.12.ln1.weight
F32
F32
[768]
v.blk.12.ln2.weight
F32
F32
[768]
v.blk.13
v.blk.13.attn_k.input_max
F32
F32
[1]
v.blk.13.attn_k.input_min
F32
F32
[1]
v.blk.13.attn_k.output_max
F32
F32
[1]
v.blk.13.attn_k.output_min
F32
F32
[1]
v.blk.13.attn_k.weight
F16
F16
[768, 768]
v.blk.13.attn_k_norm.weight
F32
F32
[64]
v.blk.13.attn_out.input_max
F32
F32
[1]
v.blk.13.attn_out.input_min
F32
F32
[1]
v.blk.13.attn_out.output_max
F32
F32
[1]
v.blk.13.attn_out.output_min
F32
F32
[1]
v.blk.13.attn_out.weight
F16
F16
[768, 768]
v.blk.13.attn_post_norm.weight
F32
F32
[768]
v.blk.13.attn_q.input_max
F32
F32
[1]
v.blk.13.attn_q.input_min
F32
F32
[1]
v.blk.13.attn_q.output_max
F32
F32
[1]
v.blk.13.attn_q.output_min
F32
F32
[1]
v.blk.13.attn_q.weight
F16
F16
[768, 768]
v.blk.13.attn_q_norm.weight
F32
F32
[64]
v.blk.13.attn_v.input_max
F32
F32
[1]
v.blk.13.attn_v.input_min
F32
F32
[1]
v.blk.13.attn_v.output_max
F32
F32
[1]
v.blk.13.attn_v.output_min
F32
F32
[1]
v.blk.13.attn_v.weight
F16
F16
[768, 768]
v.blk.13.ffn_down.input_max
F32
F32
[1]
v.blk.13.ffn_down.input_min
F32
F32
[1]
v.blk.13.ffn_down.output_max
F32
F32
[1]
v.blk.13.ffn_down.output_min
F32
F32
[1]
v.blk.13.ffn_down.weight
F16
F16
[3072, 768]
v.blk.13.ffn_gate.input_max
F32
F32
[1]
v.blk.13.ffn_gate.input_min
F32
F32
[1]
v.blk.13.ffn_gate.output_max
F32
F32
[1]
v.blk.13.ffn_gate.output_min
F32
F32
[1]
v.blk.13.ffn_gate.weight
F16
F16
[768, 3072]
v.blk.13.ffn_post_norm.weight
F32
F32
[768]
v.blk.13.ffn_up.input_max
F32
F32
[1]
v.blk.13.ffn_up.input_min
F32
F32
[1]
v.blk.13.ffn_up.output_max
F32
F32
[1]
v.blk.13.ffn_up.output_min
F32
F32
[1]
v.blk.13.ffn_up.weight
F16
F16
[768, 3072]
v.blk.13.ln1.weight
F32
F32
[768]
v.blk.13.ln2.weight
F32
F32
[768]
v.blk.14
v.blk.14.attn_k.input_max
F32
F32
[1]
v.blk.14.attn_k.input_min
F32
F32
[1]
v.blk.14.attn_k.output_max
F32
F32
[1]
v.blk.14.attn_k.output_min
F32
F32
[1]
v.blk.14.attn_k.weight
F16
F16
[768, 768]
v.blk.14.attn_k_norm.weight
F32
F32
[64]
v.blk.14.attn_out.input_max
F32
F32
[1]
v.blk.14.attn_out.input_min
F32
F32
[1]
v.blk.14.attn_out.output_max
F32
F32
[1]
v.blk.14.attn_out.output_min
F32
F32
[1]
v.blk.14.attn_out.weight
F16
F16
[768, 768]
v.blk.14.attn_post_norm.weight
F32
F32
[768]
v.blk.14.attn_q.input_max
F32
F32
[1]
v.blk.14.attn_q.input_min
F32
F32
[1]
v.blk.14.attn_q.output_max
F32
F32
[1]
v.blk.14.attn_q.output_min
F32
F32
[1]
v.blk.14.attn_q.weight
F16
F16
[768, 768]
v.blk.14.attn_q_norm.weight
F32
F32
[64]
v.blk.14.attn_v.input_max
F32
F32
[1]
v.blk.14.attn_v.input_min
F32
F32
[1]
v.blk.14.attn_v.output_max
F32
F32
[1]
v.blk.14.attn_v.output_min
F32
F32
[1]
v.blk.14.attn_v.weight
F16
F16
[768, 768]
v.blk.14.ffn_down.input_max
F32
F32
[1]
v.blk.14.ffn_down.input_min
F32
F32
[1]
v.blk.14.ffn_down.output_max
F32
F32
[1]
v.blk.14.ffn_down.output_min
F32
F32
[1]
v.blk.14.ffn_down.weight
F16
F16
[3072, 768]
v.blk.14.ffn_gate.input_max
F32
F32
[1]
v.blk.14.ffn_gate.input_min
F32
F32
[1]
v.blk.14.ffn_gate.output_max
F32
F32
[1]
v.blk.14.ffn_gate.output_min
F32
F32
[1]
v.blk.14.ffn_gate.weight
F16
F16
[768, 3072]
v.blk.14.ffn_post_norm.weight
F32
F32
[768]
v.blk.14.ffn_up.input_max
F32
F32
[1]
v.blk.14.ffn_up.input_min
F32
F32
[1]
v.blk.14.ffn_up.output_max
F32
F32
[1]
v.blk.14.ffn_up.output_min
F32
F32
[1]
v.blk.14.ffn_up.weight
F16
F16
[768, 3072]
v.blk.14.ln1.weight
F32
F32
[768]
v.blk.14.ln2.weight
F32
F32
[768]
v.blk.15
v.blk.15.attn_k.input_max
F32
F32
[1]
v.blk.15.attn_k.input_min
F32
F32
[1]
v.blk.15.attn_k.output_max
F32
F32
[1]
v.blk.15.attn_k.output_min
F32
F32
[1]
v.blk.15.attn_k.weight
F16
F16
[768, 768]
v.blk.15.attn_k_norm.weight
F32
F32
[64]
v.blk.15.attn_out.input_max
F32
F32
[1]
v.blk.15.attn_out.input_min
F32
F32
[1]
v.blk.15.attn_out.output_max
F32
F32
[1]
v.blk.15.attn_out.output_min
F32
F32
[1]
v.blk.15.attn_out.weight
F16
F16
[768, 768]
v.blk.15.attn_post_norm.weight
F32
F32
[768]
v.blk.15.attn_q.input_max
F32
F32
[1]
v.blk.15.attn_q.input_min
F32
F32
[1]
v.blk.15.attn_q.output_max
F32
F32
[1]
v.blk.15.attn_q.output_min
F32
F32
[1]
v.blk.15.attn_q.weight
F16
F16
[768, 768]
v.blk.15.attn_q_norm.weight
F32
F32
[64]
v.blk.15.attn_v.input_max
F32
F32
[1]
v.blk.15.attn_v.input_min
F32
F32
[1]
v.blk.15.attn_v.output_max
F32
F32
[1]
v.blk.15.attn_v.output_min
F32
F32
[1]
v.blk.15.attn_v.weight
F16
F16
[768, 768]
v.blk.15.ffn_down.input_max
F32
F32
[1]
v.blk.15.ffn_down.input_min
F32
F32
[1]
v.blk.15.ffn_down.output_max
F32
F32
[1]
v.blk.15.ffn_down.output_min
F32
F32
[1]
v.blk.15.ffn_down.weight
F16
F16
[3072, 768]
v.blk.15.ffn_gate.input_max
F32
F32
[1]
v.blk.15.ffn_gate.input_min
F32
F32
[1]
v.blk.15.ffn_gate.output_max
F32
F32
[1]
v.blk.15.ffn_gate.output_min
F32
F32
[1]
v.blk.15.ffn_gate.weight
F16
F16
[768, 3072]
v.blk.15.ffn_post_norm.weight
F32
F32
[768]
v.blk.15.ffn_up.input_max
F32
F32
[1]
v.blk.15.ffn_up.input_min
F32
F32
[1]
v.blk.15.ffn_up.output_max
F32
F32
[1]
v.blk.15.ffn_up.output_min
F32
F32
[1]
v.blk.15.ffn_up.weight
F16
F16
[768, 3072]
v.blk.15.ln1.weight
F32
F32
[768]
v.blk.15.ln2.weight
F32
F32
[768]
v.patch_embd.weight
F16
F16
[16, 16, 3, 768]
v.position_embd.weight
F32
F32
[768, 10240, 2]
output_norm.weight
F32
F32
[2560]