Models
Docs
Pricing
Sign in
Download
Models
Download
Docs
Pricing
Sign in
SetneufPT
/
Gemma4-12B-IT-QAT_Q4_64K_16GB-GPU
:latest
877
Downloads
Updated
1 month ago
Custom model for coding with agents to use locally with 16gb or 2x8gb GPUs (working fine...)
Custom model for coding with agents to use locally with 16gb or 2x8gb GPUs (working fine...)
Cancel
vision
tools
thinking
Gemma4-12B-IT-QAT_Q4_64K_16GB-GPU:latest
...
/
model
faff1a63667f · 7.0GB
Metadata
general.architecture
gemma4
gemma4
general.file_type
Q4_0
Q4_0
gemma4.attention.head_count
16
16
gemma4.attention.head_count_kv
[8, 8, 8, 8, 8, ...]
[8, 8, 8, 8, 8, ...]
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
0
0
gemma4.attention.sliding_window
1024
1024
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.block_count
48
48
gemma4.context_length
262144
262144
gemma4.embedding_length
3840
3840
gemma4.embedding_length_per_layer_input
0
0
gemma4.feed_forward_length
15360
15360
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
tokenizer.ggml.add_bos_token
true
true
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.suppress_tokens
[258883, 258882]
[258883, 258882]
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
[3840, 262144]
blk.0
blk.0.attn_k.weight
Q4_0
Q4_0
[3840, 2048]
blk.0.attn_k_norm.weight
F32
F32
[256]
blk.0.attn_norm.weight
F32
F32
[3840]
blk.0.attn_output.weight
Q4_0
Q4_0
[4096, 3840]
blk.0.attn_q.weight
Q4_0
Q4_0
[3840, 4096]
blk.0.attn_q_norm.weight
F32
F32
[256]
blk.0.attn_v.weight
Q4_0
Q4_0
[3840, 2048]
blk.0.ffn_down.weight
Q4_0
Q4_0
[15360, 3840]
blk.0.ffn_gate.weight
Q4_0
Q4_0
[3840, 15360]
blk.0.ffn_norm.weight
F32
F32
[3840]
blk.0.ffn_up.weight
Q4_0
Q4_0
[3840, 15360]
blk.0.layer_output_scale.weight
F32
F32
[1]
blk.0.post_attention_norm.weight
F32
F32
[3840]
blk.0.post_ffw_norm.weight
F32
F32
[3840]
blk.1
blk.1.attn_k.weight
Q4_0
Q4_0
[3840, 2048]
blk.1.attn_k_norm.weight
F32
F32
[256]
blk.1.attn_norm.weight
F32
F32
[3840]
blk.1.attn_output.weight
Q4_0
Q4_0
[4096, 3840]
blk.1.attn_q.weight
Q4_0
Q4_0
[3840, 4096]
blk.1.attn_q_norm.weight
F32
F32
[256]
blk.1.attn_v.weight
Q4_0
Q4_0
[3840, 2048]
blk.1.ffn_down.weight
Q4_0
Q4_0
[15360, 3840]
blk.1.ffn_gate.weight
Q4_0
Q4_0
[3840, 15360]
blk.1.ffn_norm.weight
F32
F32
[3840]
blk.1.ffn_up.weight
Q4_0
Q4_0
[3840, 15360]
blk.1.layer_output_scale.weight
F32
F32
[1]
blk.1.post_attention_norm.weight
F32
F32
[3840]
blk.1.post_ffw_norm.weight
F32
F32
[3840]
blk.2
blk.2.attn_k.weight
Q4_0
Q4_0
[3840, 2048]
blk.2.attn_k_norm.weight
F32
F32
[256]
blk.2.attn_norm.weight
F32
F32
[3840]
blk.2.attn_output.weight
Q4_0
Q4_0
[4096, 3840]
blk.2.attn_q.weight
Q4_0
Q4_0
[3840, 4096]
blk.2.attn_q_norm.weight
F32
F32
[256]
blk.2.attn_v.weight
Q4_0
Q4_0
[3840, 2048]
blk.2.ffn_down.weight
Q4_0
Q4_0
[15360, 3840]
blk.2.ffn_gate.weight
Q4_0
Q4_0
[3840, 15360]
blk.2.ffn_norm.weight
F32
F32
[3840]
blk.2.ffn_up.weight
Q4_0
Q4_0
[3840, 15360]
blk.2.layer_output_scale.weight
F32
F32
[1]
blk.2.post_attention_norm.weight
F32
F32
[3840]
blk.2.post_ffw_norm.weight
F32
F32
[3840]
blk.3
blk.3.attn_k.weight
Q4_0
Q4_0
[3840, 2048]
blk.3.attn_k_norm.weight
F32
F32
[256]
blk.3.attn_norm.weight
F32
F32
[3840]
blk.3.attn_output.weight
Q4_0
Q4_0
[4096, 3840]
blk.3.attn_q.weight
Q4_0
Q4_0
[3840, 4096]
blk.3.attn_q_norm.weight
F32
F32
[256]
blk.3.attn_v.weight
Q4_0
Q4_0
[3840, 2048]
blk.3.ffn_down.weight
Q4_0
Q4_0
[15360, 3840]
blk.3.ffn_gate.weight
Q4_0
Q4_0
[3840, 15360]
blk.3.ffn_norm.weight
F32
F32
[3840]
blk.3.ffn_up.weight
Q4_0
Q4_0
[3840, 15360]
blk.3.layer_output_scale.weight
F32
F32
[1]
blk.3.post_attention_norm.weight
F32
F32
[3840]
blk.3.post_ffw_norm.weight
F32
F32
[3840]
blk.4
blk.4.attn_k.weight
Q4_0
Q4_0
[3840, 2048]
blk.4.attn_k_norm.weight
F32
F32
[256]
blk.4.attn_norm.weight
F32
F32
[3840]
blk.4.attn_output.weight
Q4_0
Q4_0
[4096, 3840]
blk.4.attn_q.weight
Q4_0
Q4_0
[3840, 4096]
blk.4.attn_q_norm.weight
F32
F32
[256]
blk.4.attn_v.weight
Q4_0
Q4_0
[3840, 2048]
blk.4.ffn_down.weight
Q4_0
Q4_0
[15360, 3840]
blk.4.ffn_gate.weight
Q4_0
Q4_0
[3840, 15360]
blk.4.ffn_norm.weight
F32
F32
[3840]
blk.4.ffn_up.weight
Q4_0
Q4_0
[3840, 15360]
blk.4.layer_output_scale.weight
F32
F32
[1]
blk.4.post_attention_norm.weight
F32
F32
[3840]
blk.4.post_ffw_norm.weight
F32
F32
[3840]
blk.5
blk.5.attn_k.weight
Q4_0
Q4_0
[3840, 512]
blk.5.attn_k_norm.weight
F32
F32
[512]
blk.5.attn_norm.weight
F32
F32
[3840]
blk.5.attn_output.weight
Q4_0
Q4_0
[8192, 3840]
blk.5.attn_q.weight
Q4_0
Q4_0
[3840, 8192]
blk.5.attn_q_norm.weight
F32
F32
[512]
blk.5.ffn_down.weight
Q4_0
Q4_0
[15360, 3840]
blk.5.ffn_gate.weight
Q4_0
Q4_0
[3840, 15360]
blk.5.ffn_norm.weight
F32
F32
[3840]
blk.5.ffn_up.weight
Q4_0
Q4_0
[3840, 15360]
blk.5.layer_output_scale.weight
F32
F32
[1]
blk.5.post_attention_norm.weight
F32
F32
[3840]
blk.5.post_ffw_norm.weight
F32
F32
[3840]
blk.6
blk.6.attn_k.weight
Q4_0
Q4_0
[3840, 2048]
blk.6.attn_k_norm.weight
F32
F32
[256]
blk.6.attn_norm.weight
F32
F32
[3840]
blk.6.attn_output.weight
Q4_0
Q4_0
[4096, 3840]
blk.6.attn_q.weight
Q4_0
Q4_0
[3840, 4096]
blk.6.attn_q_norm.weight
F32
F32
[256]
blk.6.attn_v.weight
Q4_0
Q4_0
[3840, 2048]
blk.6.ffn_down.weight
Q4_0
Q4_0
[15360, 3840]
blk.6.ffn_gate.weight
Q4_0
Q4_0
[3840, 15360]
blk.6.ffn_norm.weight
F32
F32
[3840]
blk.6.ffn_up.weight
Q4_0
Q4_0
[3840, 15360]
blk.6.layer_output_scale.weight
F32
F32
[1]
blk.6.post_attention_norm.weight
F32
F32
[3840]
blk.6.post_ffw_norm.weight
F32
F32
[3840]
blk.7
blk.7.attn_k.weight
Q4_0
Q4_0
[3840, 2048]
blk.7.attn_k_norm.weight
F32
F32
[256]
blk.7.attn_norm.weight
F32
F32
[3840]
blk.7.attn_output.weight
Q4_0
Q4_0
[4096, 3840]
blk.7.attn_q.weight
Q4_0
Q4_0
[3840, 4096]
blk.7.attn_q_norm.weight
F32
F32
[256]
blk.7.attn_v.weight
Q4_0
Q4_0
[3840, 2048]
blk.7.ffn_down.weight
Q4_0
Q4_0
[15360, 3840]
blk.7.ffn_gate.weight
Q4_0
Q4_0
[3840, 15360]
blk.7.ffn_norm.weight
F32
F32
[3840]
blk.7.ffn_up.weight
Q4_0
Q4_0
[3840, 15360]
blk.7.layer_output_scale.weight
F32
F32
[1]
blk.7.post_attention_norm.weight
F32
F32
[3840]
blk.7.post_ffw_norm.weight
F32
F32
[3840]
blk.8
blk.8.attn_k.weight
Q4_0
Q4_0
[3840, 2048]
blk.8.attn_k_norm.weight
F32
F32
[256]
blk.8.attn_norm.weight
F32
F32
[3840]
blk.8.attn_output.weight
Q4_0
Q4_0
[4096, 3840]
blk.8.attn_q.weight
Q4_0
Q4_0
[3840, 4096]
blk.8.attn_q_norm.weight
F32
F32
[256]
blk.8.attn_v.weight
Q4_0
Q4_0
[3840, 2048]
blk.8.ffn_down.weight
Q4_0
Q4_0
[15360, 3840]
blk.8.ffn_gate.weight
Q4_0
Q4_0
[3840, 15360]
blk.8.ffn_norm.weight
F32
F32
[3840]
blk.8.ffn_up.weight
Q4_0
Q4_0
[3840, 15360]
blk.8.layer_output_scale.weight
F32
F32
[1]
blk.8.post_attention_norm.weight
F32
F32
[3840]
blk.8.post_ffw_norm.weight
F32
F32
[3840]
blk.9
blk.9.attn_k.weight
Q4_0
Q4_0
[3840, 2048]
blk.9.attn_k_norm.weight
F32
F32
[256]
blk.9.attn_norm.weight
F32
F32
[3840]
blk.9.attn_output.weight
Q4_0
Q4_0
[4096, 3840]
blk.9.attn_q.weight
Q4_0
Q4_0
[3840, 4096]
blk.9.attn_q_norm.weight
F32
F32
[256]
blk.9.attn_v.weight
Q4_0
Q4_0
[3840, 2048]
blk.9.ffn_down.weight
Q4_0
Q4_0
[15360, 3840]
blk.9.ffn_gate.weight
Q4_0
Q4_0
[3840, 15360]
blk.9.ffn_norm.weight
F32
F32
[3840]
blk.9.ffn_up.weight
Q4_0
Q4_0
[3840, 15360]
blk.9.layer_output_scale.weight
F32
F32
[1]
blk.9.post_attention_norm.weight
F32
F32
[3840]
blk.9.post_ffw_norm.weight
F32
F32
[3840]
blk.10
blk.10.attn_k.weight
Q4_0
Q4_0
[3840, 2048]
blk.10.attn_k_norm.weight
F32
F32
[256]
blk.10.attn_norm.weight
F32
F32
[3840]
blk.10.attn_output.weight
Q4_0
Q4_0
[4096, 3840]
blk.10.attn_q.weight
Q4_0
Q4_0
[3840, 4096]
blk.10.attn_q_norm.weight
F32
F32
[256]
blk.10.attn_v.weight
Q4_0
Q4_0
[3840, 2048]
blk.10.ffn_down.weight
Q4_0
Q4_0
[15360, 3840]
blk.10.ffn_gate.weight
Q4_0
Q4_0
[3840, 15360]
blk.10.ffn_norm.weight
F32
F32
[3840]
blk.10.ffn_up.weight
Q4_0
Q4_0
[3840, 15360]
blk.10.layer_output_scale.weight
F32
F32
[1]
blk.10.post_attention_norm.weight
F32
F32
[3840]
blk.10.post_ffw_norm.weight
F32
F32
[3840]
blk.11
blk.11.attn_k.weight
Q4_0
Q4_0
[3840, 512]
blk.11.attn_k_norm.weight
F32
F32
[512]
blk.11.attn_norm.weight
F32
F32
[3840]
blk.11.attn_output.weight
Q4_0
Q4_0
[8192, 3840]
blk.11.attn_q.weight
Q4_0
Q4_0
[3840, 8192]
blk.11.attn_q_norm.weight
F32
F32
[512]
blk.11.ffn_down.weight
Q4_0
Q4_0
[15360, 3840]
blk.11.ffn_gate.weight
Q4_0
Q4_0
[3840, 15360]
blk.11.ffn_norm.weight
F32
F32
[3840]
blk.11.ffn_up.weight
Q4_0
Q4_0
[3840, 15360]
blk.11.layer_output_scale.weight
F32
F32
[1]
blk.11.post_attention_norm.weight
F32
F32
[3840]
blk.11.post_ffw_norm.weight
F32
F32
[3840]
blk.12
blk.12.attn_k.weight
Q4_0
Q4_0
[3840, 2048]
blk.12.attn_k_norm.weight
F32
F32
[256]
blk.12.attn_norm.weight
F32
F32
[3840]
blk.12.attn_output.weight
Q4_0
Q4_0
[4096, 3840]
blk.12.attn_q.weight
Q4_0
Q4_0
[3840, 4096]
blk.12.attn_q_norm.weight
F32
F32
[256]
blk.12.attn_v.weight
Q4_0
Q4_0
[3840, 2048]
blk.12.ffn_down.weight
Q4_0
Q4_0
[15360, 3840]
blk.12.ffn_gate.weight
Q4_0
Q4_0
[3840, 15360]
blk.12.ffn_norm.weight
F32
F32
[3840]
blk.12.ffn_up.weight
Q4_0
Q4_0
[3840, 15360]
blk.12.layer_output_scale.weight
F32
F32
[1]
blk.12.post_attention_norm.weight
F32
F32
[3840]
blk.12.post_ffw_norm.weight
F32
F32
[3840]
blk.13
blk.13.attn_k.weight
Q4_0
Q4_0
[3840, 2048]
blk.13.attn_k_norm.weight
F32
F32
[256]
blk.13.attn_norm.weight
F32
F32
[3840]
blk.13.attn_output.weight
Q4_0
Q4_0
[4096, 3840]
blk.13.attn_q.weight
Q4_0
Q4_0
[3840, 4096]
blk.13.attn_q_norm.weight
F32
F32
[256]
blk.13.attn_v.weight
Q4_0
Q4_0
[3840, 2048]
blk.13.ffn_down.weight
Q4_0
Q4_0
[15360, 3840]
blk.13.ffn_gate.weight
Q4_0
Q4_0
[3840, 15360]
blk.13.ffn_norm.weight
F32
F32
[3840]
blk.13.ffn_up.weight
Q4_0
Q4_0
[3840, 15360]
blk.13.layer_output_scale.weight
F32
F32
[1]
blk.13.post_attention_norm.weight
F32
F32
[3840]
blk.13.post_ffw_norm.weight
F32
F32
[3840]
blk.14
blk.14.attn_k.weight
Q4_0
Q4_0
[3840, 2048]
blk.14.attn_k_norm.weight
F32
F32
[256]
blk.14.attn_norm.weight
F32
F32
[3840]
blk.14.attn_output.weight
Q4_0
Q4_0
[4096, 3840]
blk.14.attn_q.weight
Q4_0
Q4_0
[3840, 4096]
blk.14.attn_q_norm.weight
F32
F32
[256]
blk.14.attn_v.weight
Q4_0
Q4_0
[3840, 2048]
blk.14.ffn_down.weight
Q4_0
Q4_0
[15360, 3840]
blk.14.ffn_gate.weight
Q4_0
Q4_0
[3840, 15360]
blk.14.ffn_norm.weight
F32
F32
[3840]
blk.14.ffn_up.weight
Q4_0
Q4_0
[3840, 15360]
blk.14.layer_output_scale.weight
F32
F32
[1]
blk.14.post_attention_norm.weight
F32
F32
[3840]
blk.14.post_ffw_norm.weight
F32
F32
[3840]
blk.15
blk.15.attn_k.weight
Q4_0
Q4_0
[3840, 2048]
blk.15.attn_k_norm.weight
F32
F32
[256]
blk.15.attn_norm.weight
F32
F32
[3840]
blk.15.attn_output.weight
Q4_0
Q4_0
[4096, 3840]
blk.15.attn_q.weight
Q4_0
Q4_0
[3840, 4096]
blk.15.attn_q_norm.weight
F32
F32
[256]
blk.15.attn_v.weight
Q4_0
Q4_0
[3840, 2048]
blk.15.ffn_down.weight
Q4_0
Q4_0
[15360, 3840]
blk.15.ffn_gate.weight
Q4_0
Q4_0
[3840, 15360]
blk.15.ffn_norm.weight
F32
F32
[3840]
blk.15.ffn_up.weight
Q4_0
Q4_0
[3840, 15360]
blk.15.layer_output_scale.weight
F32
F32
[1]
blk.15.post_attention_norm.weight
F32
F32
[3840]
blk.15.post_ffw_norm.weight
F32
F32
[3840]
blk.16
blk.16.attn_k.weight
Q4_0
Q4_0
[3840, 2048]
blk.16.attn_k_norm.weight
F32
F32
[256]
blk.16.attn_norm.weight
F32
F32
[3840]
blk.16.attn_output.weight
Q4_0
Q4_0
[4096, 3840]
blk.16.attn_q.weight
Q4_0
Q4_0
[3840, 4096]
blk.16.attn_q_norm.weight
F32
F32
[256]
blk.16.attn_v.weight
Q4_0
Q4_0
[3840, 2048]
blk.16.ffn_down.weight
Q4_0
Q4_0
[15360, 3840]
blk.16.ffn_gate.weight
Q4_0
Q4_0
[3840, 15360]
blk.16.ffn_norm.weight
F32
F32
[3840]
blk.16.ffn_up.weight
Q4_0
Q4_0
[3840, 15360]
blk.16.layer_output_scale.weight
F32
F32
[1]
blk.16.post_attention_norm.weight
F32
F32
[3840]
blk.16.post_ffw_norm.weight
F32
F32
[3840]
blk.17
blk.17.attn_k.weight
Q4_0
Q4_0
[3840, 512]
blk.17.attn_k_norm.weight
F32
F32
[512]
blk.17.attn_norm.weight
F32
F32
[3840]
blk.17.attn_output.weight
Q4_0
Q4_0
[8192, 3840]
blk.17.attn_q.weight
Q4_0
Q4_0
[3840, 8192]
blk.17.attn_q_norm.weight
F32
F32
[512]
blk.17.ffn_down.weight
Q4_0
Q4_0
[15360, 3840]
blk.17.ffn_gate.weight
Q4_0
Q4_0
[3840, 15360]
blk.17.ffn_norm.weight
F32
F32
[3840]
blk.17.ffn_up.weight
Q4_0
Q4_0
[3840, 15360]
blk.17.layer_output_scale.weight
F32
F32
[1]
blk.17.post_attention_norm.weight
F32
F32
[3840]
blk.17.post_ffw_norm.weight
F32
F32
[3840]
blk.18
blk.18.attn_k.weight
Q4_0
Q4_0
[3840, 2048]
blk.18.attn_k_norm.weight
F32
F32
[256]
blk.18.attn_norm.weight
F32
F32
[3840]
blk.18.attn_output.weight
Q4_0
Q4_0
[4096, 3840]
blk.18.attn_q.weight
Q4_0
Q4_0
[3840, 4096]
blk.18.attn_q_norm.weight
F32
F32
[256]
blk.18.attn_v.weight
Q4_0
Q4_0
[3840, 2048]
blk.18.ffn_down.weight
Q4_0
Q4_0
[15360, 3840]
blk.18.ffn_gate.weight
Q4_0
Q4_0
[3840, 15360]
blk.18.ffn_norm.weight
F32
F32
[3840]
blk.18.ffn_up.weight
Q4_0
Q4_0
[3840, 15360]
blk.18.layer_output_scale.weight
F32
F32
[1]
blk.18.post_attention_norm.weight
F32
F32
[3840]
blk.18.post_ffw_norm.weight
F32
F32
[3840]
blk.19
blk.19.attn_k.weight
Q4_0
Q4_0
[3840, 2048]
blk.19.attn_k_norm.weight
F32
F32
[256]
blk.19.attn_norm.weight
F32
F32
[3840]
blk.19.attn_output.weight
Q4_0
Q4_0
[4096, 3840]
blk.19.attn_q.weight
Q4_0
Q4_0
[3840, 4096]
blk.19.attn_q_norm.weight
F32
F32
[256]
blk.19.attn_v.weight
Q4_0
Q4_0
[3840, 2048]
blk.19.ffn_down.weight
Q4_0
Q4_0
[15360, 3840]
blk.19.ffn_gate.weight
Q4_0
Q4_0
[3840, 15360]
blk.19.ffn_norm.weight
F32
F32
[3840]
blk.19.ffn_up.weight
Q4_0
Q4_0
[3840, 15360]
blk.19.layer_output_scale.weight
F32
F32
[1]
blk.19.post_attention_norm.weight
F32
F32
[3840]
blk.19.post_ffw_norm.weight
F32
F32
[3840]
blk.20
blk.20.attn_k.weight
Q4_0
Q4_0
[3840, 2048]
blk.20.attn_k_norm.weight
F32
F32
[256]
blk.20.attn_norm.weight
F32
F32
[3840]
blk.20.attn_output.weight
Q4_0
Q4_0
[4096, 3840]
blk.20.attn_q.weight
Q4_0
Q4_0
[3840, 4096]
blk.20.attn_q_norm.weight
F32
F32
[256]
blk.20.attn_v.weight
Q4_0
Q4_0
[3840, 2048]
blk.20.ffn_down.weight
Q4_0
Q4_0
[15360, 3840]
blk.20.ffn_gate.weight
Q4_0
Q4_0
[3840, 15360]
blk.20.ffn_norm.weight
F32
F32
[3840]
blk.20.ffn_up.weight
Q4_0
Q4_0
[3840, 15360]
blk.20.layer_output_scale.weight
F32
F32
[1]
blk.20.post_attention_norm.weight
F32
F32
[3840]
blk.20.post_ffw_norm.weight
F32
F32
[3840]
blk.21
blk.21.attn_k.weight
Q4_0
Q4_0
[3840, 2048]
blk.21.attn_k_norm.weight
F32
F32
[256]
blk.21.attn_norm.weight
F32
F32
[3840]
blk.21.attn_output.weight
Q4_0
Q4_0
[4096, 3840]
blk.21.attn_q.weight
Q4_0
Q4_0
[3840, 4096]
blk.21.attn_q_norm.weight
F32
F32
[256]
blk.21.attn_v.weight
Q4_0
Q4_0
[3840, 2048]
blk.21.ffn_down.weight
Q4_0
Q4_0
[15360, 3840]
blk.21.ffn_gate.weight
Q4_0
Q4_0
[3840, 15360]
blk.21.ffn_norm.weight
F32
F32
[3840]
blk.21.ffn_up.weight
Q4_0
Q4_0
[3840, 15360]
blk.21.layer_output_scale.weight
F32
F32
[1]
blk.21.post_attention_norm.weight
F32
F32
[3840]
blk.21.post_ffw_norm.weight
F32
F32
[3840]
blk.22
blk.22.attn_k.weight
Q4_0
Q4_0
[3840, 2048]
blk.22.attn_k_norm.weight
F32
F32
[256]
blk.22.attn_norm.weight
F32
F32
[3840]
blk.22.attn_output.weight
Q4_0
Q4_0
[4096, 3840]
blk.22.attn_q.weight
Q4_0
Q4_0
[3840, 4096]
blk.22.attn_q_norm.weight
F32
F32
[256]
blk.22.attn_v.weight
Q4_0
Q4_0
[3840, 2048]
blk.22.ffn_down.weight
Q4_0
Q4_0
[15360, 3840]
blk.22.ffn_gate.weight
Q4_0
Q4_0
[3840, 15360]
blk.22.ffn_norm.weight
F32
F32
[3840]
blk.22.ffn_up.weight
Q4_0
Q4_0
[3840, 15360]
blk.22.layer_output_scale.weight
F32
F32
[1]
blk.22.post_attention_norm.weight
F32
F32
[3840]
blk.22.post_ffw_norm.weight
F32
F32
[3840]
blk.23
blk.23.attn_k.weight
Q4_0
Q4_0
[3840, 512]
blk.23.attn_k_norm.weight
F32
F32
[512]
blk.23.attn_norm.weight
F32
F32
[3840]
blk.23.attn_output.weight
Q4_0
Q4_0
[8192, 3840]
blk.23.attn_q.weight
Q4_0
Q4_0
[3840, 8192]
blk.23.attn_q_norm.weight
F32
F32
[512]
blk.23.ffn_down.weight
Q4_0
Q4_0
[15360, 3840]
blk.23.ffn_gate.weight
Q4_0
Q4_0
[3840, 15360]
blk.23.ffn_norm.weight
F32
F32
[3840]
blk.23.ffn_up.weight
Q4_0
Q4_0
[3840, 15360]
blk.23.layer_output_scale.weight
F32
F32
[1]
blk.23.post_attention_norm.weight
F32
F32
[3840]
blk.23.post_ffw_norm.weight
F32
F32
[3840]
blk.24
blk.24.attn_k.weight
Q4_0
Q4_0
[3840, 2048]
blk.24.attn_k_norm.weight
F32
F32
[256]
blk.24.attn_norm.weight
F32
F32
[3840]
blk.24.attn_output.weight
Q4_0
Q4_0
[4096, 3840]
blk.24.attn_q.weight
Q4_0
Q4_0
[3840, 4096]
blk.24.attn_q_norm.weight
F32
F32
[256]
blk.24.attn_v.weight
Q4_0
Q4_0
[3840, 2048]
blk.24.ffn_down.weight
Q4_0
Q4_0
[15360, 3840]
blk.24.ffn_gate.weight
Q4_0
Q4_0
[3840, 15360]
blk.24.ffn_norm.weight
F32
F32
[3840]
blk.24.ffn_up.weight
Q4_0
Q4_0
[3840, 15360]
blk.24.layer_output_scale.weight
F32
F32
[1]
blk.24.post_attention_norm.weight
F32
F32
[3840]
blk.24.post_ffw_norm.weight
F32
F32
[3840]
blk.25
blk.25.attn_k.weight
Q4_0
Q4_0
[3840, 2048]
blk.25.attn_k_norm.weight
F32
F32
[256]
blk.25.attn_norm.weight
F32
F32
[3840]
blk.25.attn_output.weight
Q4_0
Q4_0
[4096, 3840]
blk.25.attn_q.weight
Q4_0
Q4_0
[3840, 4096]
blk.25.attn_q_norm.weight
F32
F32
[256]
blk.25.attn_v.weight
Q4_0
Q4_0
[3840, 2048]
blk.25.ffn_down.weight
Q4_0
Q4_0
[15360, 3840]
blk.25.ffn_gate.weight
Q4_0
Q4_0
[3840, 15360]
blk.25.ffn_norm.weight
F32
F32
[3840]
blk.25.ffn_up.weight
Q4_0
Q4_0
[3840, 15360]
blk.25.layer_output_scale.weight
F32
F32
[1]
blk.25.post_attention_norm.weight
F32
F32
[3840]
blk.25.post_ffw_norm.weight
F32
F32
[3840]
blk.26
blk.26.attn_k.weight
Q4_0
Q4_0
[3840, 2048]
blk.26.attn_k_norm.weight
F32
F32
[256]
blk.26.attn_norm.weight
F32
F32
[3840]
blk.26.attn_output.weight
Q4_0
Q4_0
[4096, 3840]
blk.26.attn_q.weight
Q4_0
Q4_0
[3840, 4096]
blk.26.attn_q_norm.weight
F32
F32
[256]
blk.26.attn_v.weight
Q4_0
Q4_0
[3840, 2048]
blk.26.ffn_down.weight
Q4_0
Q4_0
[15360, 3840]
blk.26.ffn_gate.weight
Q4_0
Q4_0
[3840, 15360]
blk.26.ffn_norm.weight
F32
F32
[3840]
blk.26.ffn_up.weight
Q4_0
Q4_0
[3840, 15360]
blk.26.layer_output_scale.weight
F32
F32
[1]
blk.26.post_attention_norm.weight
F32
F32
[3840]
blk.26.post_ffw_norm.weight
F32
F32
[3840]
blk.27
blk.27.attn_k.weight
Q4_0
Q4_0
[3840, 2048]
blk.27.attn_k_norm.weight
F32
F32
[256]
blk.27.attn_norm.weight
F32
F32
[3840]
blk.27.attn_output.weight
Q4_0
Q4_0
[4096, 3840]
blk.27.attn_q.weight
Q4_0
Q4_0
[3840, 4096]
blk.27.attn_q_norm.weight
F32
F32
[256]
blk.27.attn_v.weight
Q4_0
Q4_0
[3840, 2048]
blk.27.ffn_down.weight
Q4_0
Q4_0
[15360, 3840]
blk.27.ffn_gate.weight
Q4_0
Q4_0
[3840, 15360]
blk.27.ffn_norm.weight
F32
F32
[3840]
blk.27.ffn_up.weight
Q4_0
Q4_0
[3840, 15360]
blk.27.layer_output_scale.weight
F32
F32
[1]
blk.27.post_attention_norm.weight
F32
F32
[3840]
blk.27.post_ffw_norm.weight
F32
F32
[3840]
blk.28
blk.28.attn_k.weight
Q4_0
Q4_0
[3840, 2048]
blk.28.attn_k_norm.weight
F32
F32
[256]
blk.28.attn_norm.weight
F32
F32
[3840]
blk.28.attn_output.weight
Q4_0
Q4_0
[4096, 3840]
blk.28.attn_q.weight
Q4_0
Q4_0
[3840, 4096]
blk.28.attn_q_norm.weight
F32
F32
[256]
blk.28.attn_v.weight
Q4_0
Q4_0
[3840, 2048]
blk.28.ffn_down.weight
Q4_0
Q4_0
[15360, 3840]
blk.28.ffn_gate.weight
Q4_0
Q4_0
[3840, 15360]
blk.28.ffn_norm.weight
F32
F32
[3840]
blk.28.ffn_up.weight
Q4_0
Q4_0
[3840, 15360]
blk.28.layer_output_scale.weight
F32
F32
[1]
blk.28.post_attention_norm.weight
F32
F32
[3840]
blk.28.post_ffw_norm.weight
F32
F32
[3840]
blk.29
blk.29.attn_k.weight
Q4_0
Q4_0
[3840, 512]
blk.29.attn_k_norm.weight
F32
F32
[512]
blk.29.attn_norm.weight
F32
F32
[3840]
blk.29.attn_output.weight
Q4_0
Q4_0
[8192, 3840]
blk.29.attn_q.weight
Q4_0
Q4_0
[3840, 8192]
blk.29.attn_q_norm.weight
F32
F32
[512]
blk.29.ffn_down.weight
Q4_0
Q4_0
[15360, 3840]
blk.29.ffn_gate.weight
Q4_0
Q4_0
[3840, 15360]
blk.29.ffn_norm.weight
F32
F32
[3840]
blk.29.ffn_up.weight
Q4_0
Q4_0
[3840, 15360]
blk.29.layer_output_scale.weight
F32
F32
[1]
blk.29.post_attention_norm.weight
F32
F32
[3840]
blk.29.post_ffw_norm.weight
F32
F32
[3840]
blk.30
blk.30.attn_k.weight
Q4_0
Q4_0
[3840, 2048]
blk.30.attn_k_norm.weight
F32
F32
[256]
blk.30.attn_norm.weight
F32
F32
[3840]
blk.30.attn_output.weight
Q4_0
Q4_0
[4096, 3840]
blk.30.attn_q.weight
Q4_0
Q4_0
[3840, 4096]
blk.30.attn_q_norm.weight
F32
F32
[256]
blk.30.attn_v.weight
Q4_0
Q4_0
[3840, 2048]
blk.30.ffn_down.weight
Q4_0
Q4_0
[15360, 3840]
blk.30.ffn_gate.weight
Q4_0
Q4_0
[3840, 15360]
blk.30.ffn_norm.weight
F32
F32
[3840]
blk.30.ffn_up.weight
Q4_0
Q4_0
[3840, 15360]
blk.30.layer_output_scale.weight
F32
F32
[1]
blk.30.post_attention_norm.weight
F32
F32
[3840]
blk.30.post_ffw_norm.weight
F32
F32
[3840]
blk.31
blk.31.attn_k.weight
Q4_0
Q4_0
[3840, 2048]
blk.31.attn_k_norm.weight
F32
F32
[256]
blk.31.attn_norm.weight
F32
F32
[3840]
blk.31.attn_output.weight
Q4_0
Q4_0
[4096, 3840]
blk.31.attn_q.weight
Q4_0
Q4_0
[3840, 4096]
blk.31.attn_q_norm.weight
F32
F32
[256]
blk.31.attn_v.weight
Q4_0
Q4_0
[3840, 2048]
blk.31.ffn_down.weight
Q4_0
Q4_0
[15360, 3840]
blk.31.ffn_gate.weight
Q4_0
Q4_0
[3840, 15360]
blk.31.ffn_norm.weight
F32
F32
[3840]
blk.31.ffn_up.weight
Q4_0
Q4_0
[3840, 15360]
blk.31.layer_output_scale.weight
F32
F32
[1]
blk.31.post_attention_norm.weight
F32
F32
[3840]
blk.31.post_ffw_norm.weight
F32
F32
[3840]
blk.32
blk.32.attn_k.weight
Q4_0
Q4_0
[3840, 2048]
blk.32.attn_k_norm.weight
F32
F32
[256]
blk.32.attn_norm.weight
F32
F32
[3840]
blk.32.attn_output.weight
Q4_0
Q4_0
[4096, 3840]
blk.32.attn_q.weight
Q4_0
Q4_0
[3840, 4096]
blk.32.attn_q_norm.weight
F32
F32
[256]
blk.32.attn_v.weight
Q4_0
Q4_0
[3840, 2048]
blk.32.ffn_down.weight
Q4_0
Q4_0
[15360, 3840]
blk.32.ffn_gate.weight
Q4_0
Q4_0
[3840, 15360]
blk.32.ffn_norm.weight
F32
F32
[3840]
blk.32.ffn_up.weight
Q4_0
Q4_0
[3840, 15360]
blk.32.layer_output_scale.weight
F32
F32
[1]
blk.32.post_attention_norm.weight
F32
F32
[3840]
blk.32.post_ffw_norm.weight
F32
F32
[3840]
blk.33
blk.33.attn_k.weight
Q4_0
Q4_0
[3840, 2048]
blk.33.attn_k_norm.weight
F32
F32
[256]
blk.33.attn_norm.weight
F32
F32
[3840]
blk.33.attn_output.weight
Q4_0
Q4_0
[4096, 3840]
blk.33.attn_q.weight
Q4_0
Q4_0
[3840, 4096]
blk.33.attn_q_norm.weight
F32
F32
[256]
blk.33.attn_v.weight
Q4_0
Q4_0
[3840, 2048]
blk.33.ffn_down.weight
Q4_0
Q4_0
[15360, 3840]
blk.33.ffn_gate.weight
Q4_0
Q4_0
[3840, 15360]
blk.33.ffn_norm.weight
F32
F32
[3840]
blk.33.ffn_up.weight
Q4_0
Q4_0
[3840, 15360]
blk.33.layer_output_scale.weight
F32
F32
[1]
blk.33.post_attention_norm.weight
F32
F32
[3840]
blk.33.post_ffw_norm.weight
F32
F32
[3840]
blk.34
blk.34.attn_k.weight
Q4_0
Q4_0
[3840, 2048]
blk.34.attn_k_norm.weight
F32
F32
[256]
blk.34.attn_norm.weight
F32
F32
[3840]
blk.34.attn_output.weight
Q4_0
Q4_0
[4096, 3840]
blk.34.attn_q.weight
Q4_0
Q4_0
[3840, 4096]
blk.34.attn_q_norm.weight
F32
F32
[256]
blk.34.attn_v.weight
Q4_0
Q4_0
[3840, 2048]
blk.34.ffn_down.weight
Q4_0
Q4_0
[15360, 3840]
blk.34.ffn_gate.weight
Q4_0
Q4_0
[3840, 15360]
blk.34.ffn_norm.weight
F32
F32
[3840]
blk.34.ffn_up.weight
Q4_0
Q4_0
[3840, 15360]
blk.34.layer_output_scale.weight
F32
F32
[1]
blk.34.post_attention_norm.weight
F32
F32
[3840]
blk.34.post_ffw_norm.weight
F32
F32
[3840]
blk.35
blk.35.attn_k.weight
Q4_0
Q4_0
[3840, 512]
blk.35.attn_k_norm.weight
F32
F32
[512]
blk.35.attn_norm.weight
F32
F32
[3840]
blk.35.attn_output.weight
Q4_0
Q4_0
[8192, 3840]
blk.35.attn_q.weight
Q4_0
Q4_0
[3840, 8192]
blk.35.attn_q_norm.weight
F32
F32
[512]
blk.35.ffn_down.weight
Q4_0
Q4_0
[15360, 3840]
blk.35.ffn_gate.weight
Q4_0
Q4_0
[3840, 15360]
blk.35.ffn_norm.weight
F32
F32
[3840]
blk.35.ffn_up.weight
Q4_0
Q4_0
[3840, 15360]
blk.35.layer_output_scale.weight
F32
F32
[1]
blk.35.post_attention_norm.weight
F32
F32
[3840]
blk.35.post_ffw_norm.weight
F32
F32
[3840]
blk.36
blk.36.attn_k.weight
Q4_0
Q4_0
[3840, 2048]
blk.36.attn_k_norm.weight
F32
F32
[256]
blk.36.attn_norm.weight
F32
F32
[3840]
blk.36.attn_output.weight
Q4_0
Q4_0
[4096, 3840]
blk.36.attn_q.weight
Q4_0
Q4_0
[3840, 4096]
blk.36.attn_q_norm.weight
F32
F32
[256]
blk.36.attn_v.weight
Q4_0
Q4_0
[3840, 2048]
blk.36.ffn_down.weight
Q4_0
Q4_0
[15360, 3840]
blk.36.ffn_gate.weight
Q4_0
Q4_0
[3840, 15360]
blk.36.ffn_norm.weight
F32
F32
[3840]
blk.36.ffn_up.weight
Q4_0
Q4_0
[3840, 15360]
blk.36.layer_output_scale.weight
F32
F32
[1]
blk.36.post_attention_norm.weight
F32
F32
[3840]
blk.36.post_ffw_norm.weight
F32
F32
[3840]
blk.37
blk.37.attn_k.weight
Q4_0
Q4_0
[3840, 2048]
blk.37.attn_k_norm.weight
F32
F32
[256]
blk.37.attn_norm.weight
F32
F32
[3840]
blk.37.attn_output.weight
Q4_0
Q4_0
[4096, 3840]
blk.37.attn_q.weight
Q4_0
Q4_0
[3840, 4096]
blk.37.attn_q_norm.weight
F32
F32
[256]
blk.37.attn_v.weight
Q4_0
Q4_0
[3840, 2048]
blk.37.ffn_down.weight
Q4_0
Q4_0
[15360, 3840]
blk.37.ffn_gate.weight
Q4_0
Q4_0
[3840, 15360]
blk.37.ffn_norm.weight
F32
F32
[3840]
blk.37.ffn_up.weight
Q4_0
Q4_0
[3840, 15360]
blk.37.layer_output_scale.weight
F32
F32
[1]
blk.37.post_attention_norm.weight
F32
F32
[3840]
blk.37.post_ffw_norm.weight
F32
F32
[3840]
blk.38
blk.38.attn_k.weight
Q4_0
Q4_0
[3840, 2048]
blk.38.attn_k_norm.weight
F32
F32
[256]
blk.38.attn_norm.weight
F32
F32
[3840]
blk.38.attn_output.weight
Q4_0
Q4_0
[4096, 3840]
blk.38.attn_q.weight
Q4_0
Q4_0
[3840, 4096]
blk.38.attn_q_norm.weight
F32
F32
[256]
blk.38.attn_v.weight
Q4_0
Q4_0
[3840, 2048]
blk.38.ffn_down.weight
Q4_0
Q4_0
[15360, 3840]
blk.38.ffn_gate.weight
Q4_0
Q4_0
[3840, 15360]
blk.38.ffn_norm.weight
F32
F32
[3840]
blk.38.ffn_up.weight
Q4_0
Q4_0
[3840, 15360]
blk.38.layer_output_scale.weight
F32
F32
[1]
blk.38.post_attention_norm.weight
F32
F32
[3840]
blk.38.post_ffw_norm.weight
F32
F32
[3840]
blk.39
blk.39.attn_k.weight
Q4_0
Q4_0
[3840, 2048]
blk.39.attn_k_norm.weight
F32
F32
[256]
blk.39.attn_norm.weight
F32
F32
[3840]
blk.39.attn_output.weight
Q4_0
Q4_0
[4096, 3840]
blk.39.attn_q.weight
Q4_0
Q4_0
[3840, 4096]
blk.39.attn_q_norm.weight
F32
F32
[256]
blk.39.attn_v.weight
Q4_0
Q4_0
[3840, 2048]
blk.39.ffn_down.weight
Q4_0
Q4_0
[15360, 3840]
blk.39.ffn_gate.weight
Q4_0
Q4_0
[3840, 15360]
blk.39.ffn_norm.weight
F32
F32
[3840]
blk.39.ffn_up.weight
Q4_0
Q4_0
[3840, 15360]
blk.39.layer_output_scale.weight
F32
F32
[1]
blk.39.post_attention_norm.weight
F32
F32
[3840]
blk.39.post_ffw_norm.weight
F32
F32
[3840]
blk.40
blk.40.attn_k.weight
Q4_0
Q4_0
[3840, 2048]
blk.40.attn_k_norm.weight
F32
F32
[256]
blk.40.attn_norm.weight
F32
F32
[3840]
blk.40.attn_output.weight
Q4_0
Q4_0
[4096, 3840]
blk.40.attn_q.weight
Q4_0
Q4_0
[3840, 4096]
blk.40.attn_q_norm.weight
F32
F32
[256]
blk.40.attn_v.weight
Q4_0
Q4_0
[3840, 2048]
blk.40.ffn_down.weight
Q4_0
Q4_0
[15360, 3840]
blk.40.ffn_gate.weight
Q4_0
Q4_0
[3840, 15360]
blk.40.ffn_norm.weight
F32
F32
[3840]
blk.40.ffn_up.weight
Q4_0
Q4_0
[3840, 15360]
blk.40.layer_output_scale.weight
F32
F32
[1]
blk.40.post_attention_norm.weight
F32
F32
[3840]
blk.40.post_ffw_norm.weight
F32
F32
[3840]
blk.41
blk.41.attn_k.weight
Q4_0
Q4_0
[3840, 512]
blk.41.attn_k_norm.weight
F32
F32
[512]
blk.41.attn_norm.weight
F32
F32
[3840]
blk.41.attn_output.weight
Q4_0
Q4_0
[8192, 3840]
blk.41.attn_q.weight
Q4_0
Q4_0
[3840, 8192]
blk.41.attn_q_norm.weight
F32
F32
[512]
blk.41.ffn_down.weight
Q4_0
Q4_0
[15360, 3840]
blk.41.ffn_gate.weight
Q4_0
Q4_0
[3840, 15360]
blk.41.ffn_norm.weight
F32
F32
[3840]
blk.41.ffn_up.weight
Q4_0
Q4_0
[3840, 15360]
blk.41.layer_output_scale.weight
F32
F32
[1]
blk.41.post_attention_norm.weight
F32
F32
[3840]
blk.41.post_ffw_norm.weight
F32
F32
[3840]
blk.42
blk.42.attn_k.weight
Q4_0
Q4_0
[3840, 2048]
blk.42.attn_k_norm.weight
F32
F32
[256]
blk.42.attn_norm.weight
F32
F32
[3840]
blk.42.attn_output.weight
Q4_0
Q4_0
[4096, 3840]
blk.42.attn_q.weight
Q4_0
Q4_0
[3840, 4096]
blk.42.attn_q_norm.weight
F32
F32
[256]
blk.42.attn_v.weight
Q4_0
Q4_0
[3840, 2048]
blk.42.ffn_down.weight
Q4_0
Q4_0
[15360, 3840]
blk.42.ffn_gate.weight
Q4_0
Q4_0
[3840, 15360]
blk.42.ffn_norm.weight
F32
F32
[3840]
blk.42.ffn_up.weight
Q4_0
Q4_0
[3840, 15360]
blk.42.layer_output_scale.weight
F32
F32
[1]
blk.42.post_attention_norm.weight
F32
F32
[3840]
blk.42.post_ffw_norm.weight
F32
F32
[3840]
blk.43
blk.43.attn_k.weight
Q4_0
Q4_0
[3840, 2048]
blk.43.attn_k_norm.weight
F32
F32
[256]
blk.43.attn_norm.weight
F32
F32
[3840]
blk.43.attn_output.weight
Q4_0
Q4_0
[4096, 3840]
blk.43.attn_q.weight
Q4_0
Q4_0
[3840, 4096]
blk.43.attn_q_norm.weight
F32
F32
[256]
blk.43.attn_v.weight
Q4_0
Q4_0
[3840, 2048]
blk.43.ffn_down.weight
Q4_0
Q4_0
[15360, 3840]
blk.43.ffn_gate.weight
Q4_0
Q4_0
[3840, 15360]
blk.43.ffn_norm.weight
F32
F32
[3840]
blk.43.ffn_up.weight
Q4_0
Q4_0
[3840, 15360]
blk.43.layer_output_scale.weight
F32
F32
[1]
blk.43.post_attention_norm.weight
F32
F32
[3840]
blk.43.post_ffw_norm.weight
F32
F32
[3840]
blk.44
blk.44.attn_k.weight
Q4_0
Q4_0
[3840, 2048]
blk.44.attn_k_norm.weight
F32
F32
[256]
blk.44.attn_norm.weight
F32
F32
[3840]
blk.44.attn_output.weight
Q4_0
Q4_0
[4096, 3840]
blk.44.attn_q.weight
Q4_0
Q4_0
[3840, 4096]
blk.44.attn_q_norm.weight
F32
F32
[256]
blk.44.attn_v.weight
Q4_0
Q4_0
[3840, 2048]
blk.44.ffn_down.weight
Q4_0
Q4_0
[15360, 3840]
blk.44.ffn_gate.weight
Q4_0
Q4_0
[3840, 15360]
blk.44.ffn_norm.weight
F32
F32
[3840]
blk.44.ffn_up.weight
Q4_0
Q4_0
[3840, 15360]
blk.44.layer_output_scale.weight
F32
F32
[1]
blk.44.post_attention_norm.weight
F32
F32
[3840]
blk.44.post_ffw_norm.weight
F32
F32
[3840]
blk.45
blk.45.attn_k.weight
Q4_0
Q4_0
[3840, 2048]
blk.45.attn_k_norm.weight
F32
F32
[256]
blk.45.attn_norm.weight
F32
F32
[3840]
blk.45.attn_output.weight
Q4_0
Q4_0
[4096, 3840]
blk.45.attn_q.weight
Q4_0
Q4_0
[3840, 4096]
blk.45.attn_q_norm.weight
F32
F32
[256]
blk.45.attn_v.weight
Q4_0
Q4_0
[3840, 2048]
blk.45.ffn_down.weight
Q4_0
Q4_0
[15360, 3840]
blk.45.ffn_gate.weight
Q4_0
Q4_0
[3840, 15360]
blk.45.ffn_norm.weight
F32
F32
[3840]
blk.45.ffn_up.weight
Q4_0
Q4_0
[3840, 15360]
blk.45.layer_output_scale.weight
F32
F32
[1]
blk.45.post_attention_norm.weight
F32
F32
[3840]
blk.45.post_ffw_norm.weight
F32
F32
[3840]
blk.46
blk.46.attn_k.weight
Q4_0
Q4_0
[3840, 2048]
blk.46.attn_k_norm.weight
F32
F32
[256]
blk.46.attn_norm.weight
F32
F32
[3840]
blk.46.attn_output.weight
Q4_0
Q4_0
[4096, 3840]
blk.46.attn_q.weight
Q4_0
Q4_0
[3840, 4096]
blk.46.attn_q_norm.weight
F32
F32
[256]
blk.46.attn_v.weight
Q4_0
Q4_0
[3840, 2048]
blk.46.ffn_down.weight
Q4_0
Q4_0
[15360, 3840]
blk.46.ffn_gate.weight
Q4_0
Q4_0
[3840, 15360]
blk.46.ffn_norm.weight
F32
F32
[3840]
blk.46.ffn_up.weight
Q4_0
Q4_0
[3840, 15360]
blk.46.layer_output_scale.weight
F32
F32
[1]
blk.46.post_attention_norm.weight
F32
F32
[3840]
blk.46.post_ffw_norm.weight
F32
F32
[3840]
blk.47
blk.47.attn_k.weight
Q4_0
Q4_0
[3840, 512]
blk.47.attn_k_norm.weight
F32
F32
[512]
blk.47.attn_norm.weight
F32
F32
[3840]
blk.47.attn_output.weight
Q4_0
Q4_0
[8192, 3840]
blk.47.attn_q.weight
Q4_0
Q4_0
[3840, 8192]
blk.47.attn_q_norm.weight
F32
F32
[512]
blk.47.ffn_down.weight
Q4_0
Q4_0
[15360, 3840]
blk.47.ffn_gate.weight
Q4_0
Q4_0
[3840, 15360]
blk.47.ffn_norm.weight
F32
F32
[3840]
blk.47.ffn_up.weight
Q4_0
Q4_0
[3840, 15360]
blk.47.layer_output_scale.weight
F32
F32
[1]
blk.47.post_attention_norm.weight
F32
F32
[3840]
blk.47.post_ffw_norm.weight
F32
F32
[3840]
rope_freqs.weight
F32
F32
[256]
output_norm.weight
F32
F32
[3840]