Models
GitHub
Discord
Turbo
Sign in
Download
Models
Download
GitHub
Discord
Sign in
dengcao
/
ERNIE-4.5-21B-A3B-PT
:latest
155
Downloads
Updated
1 month ago
This model was converted to GGUF format from baidu/ERNIE-4.5-21B-A3B-PT using llama.cpp via the ggml.ai's GGUF-my-repo space.
This model was converted to GGUF format from baidu/ERNIE-4.5-21B-A3B-PT using llama.cpp via the ggml.ai's GGUF-my-repo space.
Cancel
ERNIE-4.5-21B-A3B-PT:latest
...
/
model
150e2463d764 · 13GB
Metadata
general.architecture
ernie4_5-moe
ernie4_5-moe
general.file_type
Q4_K_M
Q4_K_M
ernie4_5-moe.attention.head_count
20
20
ernie4_5-moe.attention.head_count_kv
4
4
ernie4_5-moe.attention.layer_norm_rms_epsilon
1e-05
1e-05
ernie4_5-moe.block_count
28
28
ernie4_5-moe.context_length
131072
131072
ernie4_5-moe.embedding_length
2560
2560
ernie4_5-moe.expert_count
64
64
ernie4_5-moe.expert_feed_forward_length
1536
1536
ernie4_5-moe.expert_shared_count
2
2
ernie4_5-moe.expert_shared_feed_forward_length
3072
3072
ernie4_5-moe.expert_used_count
6
6
ernie4_5-moe.feed_forward_length
12288
12288
ernie4_5-moe.interleave_moe_layer_step
1
1
ernie4_5-moe.leading_dense_block_count
1
1
ernie4_5-moe.rope.freq_base
500000
500000
tokenizer.ggml.bos_token_id
1
1
tokenizer.ggml.eos_token_id
2
2
tokenizer.ggml.model
llama
llama
tokenizer.ggml.padding_token_id
0
0
tokenizer.ggml.pre
default
default
tokenizer.ggml.scores
[-1000, -1000, -1000, 0, 0, ...]
[-1000, -1000, -1000, 0, 0, ...]
tokenizer.ggml.token_type
[3, 3, 3, 1, 1, ...]
[3, 3, 3, 1, 1, ...]
tokenizer.ggml.tokens
[<unk>, <s>, </s>, 0, 1, ...]
[<unk>, <s>, </s>, 0, 1, ...]
quantize.imatrix.chunks_count
622
622
quantize.imatrix.dataset
unsloth_calibration_ERNIE-4.5-21B-A3B-PT.txt
unsloth_calibration_ERNIE-4.5-21B-A3B-PT.txt
quantize.imatrix.entries_count
304
304
quantize.imatrix.file
ERNIE-4.5-21B-A3B-PT-GGUF/imatrix_unsloth.dat
ERNIE-4.5-21B-A3B-PT-GGUF/imatrix_unsloth.dat
Tensor
Name
Type
Shape
token_embd.weight
Q6_K
Q6_K
[2560, 103424]
blk.0
blk.0.attn_k.weight
Q4_K
Q4_K
[2560, 512]
blk.0.attn_norm.weight
F32
F32
[2560]
blk.0.attn_output.weight
Q4_K
Q4_K
[2560, 2560]
blk.0.attn_q.weight
Q4_K
Q4_K
[2560, 2560]
blk.0.attn_v.weight
Q6_K
Q6_K
[2560, 512]
blk.0.ffn_down.weight
Q6_K
Q6_K
[12288, 2560]
blk.0.ffn_gate.weight
Q4_K
Q4_K
[2560, 12288]
blk.0.ffn_norm.weight
F32
F32
[2560]
blk.0.ffn_up.weight
Q4_K
Q4_K
[2560, 12288]
blk.1
blk.1.attn_k.weight
Q4_K
Q4_K
[2560, 512]
blk.1.attn_norm.weight
F32
F32
[2560]
blk.1.attn_output.weight
Q4_K
Q4_K
[2560, 2560]
blk.1.attn_q.weight
Q4_K
Q4_K
[2560, 2560]
blk.1.attn_v.weight
Q6_K
Q6_K
[2560, 512]
blk.1.exp_probs_b.bias
F32
F32
[64]
blk.1.ffn_down_exps.weight
Q6_K
Q6_K
[1536, 2560, 64]
blk.1.ffn_down_shexp.weight
Q6_K
Q6_K
[3072, 2560]
blk.1.ffn_gate_exps.weight
Q4_K
Q4_K
[2560, 1536, 64]
blk.1.ffn_gate_inp.weight
F32
F32
[2560, 64]
blk.1.ffn_gate_shexp.weight
Q4_K
Q4_K
[2560, 3072]
blk.1.ffn_norm.weight
F32
F32
[2560]
blk.1.ffn_up_exps.weight
Q4_K
Q4_K
[2560, 1536, 64]
blk.1.ffn_up_shexp.weight
Q4_K
Q4_K
[2560, 3072]
blk.2
blk.2.attn_k.weight
Q4_K
Q4_K
[2560, 512]
blk.2.attn_norm.weight
F32
F32
[2560]
blk.2.attn_output.weight
Q4_K
Q4_K
[2560, 2560]
blk.2.attn_q.weight
Q4_K
Q4_K
[2560, 2560]
blk.2.attn_v.weight
Q6_K
Q6_K
[2560, 512]
blk.2.exp_probs_b.bias
F32
F32
[64]
blk.2.ffn_down_exps.weight
Q6_K
Q6_K
[1536, 2560, 64]
blk.2.ffn_down_shexp.weight
Q6_K
Q6_K
[3072, 2560]
blk.2.ffn_gate_exps.weight
Q4_K
Q4_K
[2560, 1536, 64]
blk.2.ffn_gate_inp.weight
F32
F32
[2560, 64]
blk.2.ffn_gate_shexp.weight
Q4_K
Q4_K
[2560, 3072]
blk.2.ffn_norm.weight
F32
F32
[2560]
blk.2.ffn_up_exps.weight
Q4_K
Q4_K
[2560, 1536, 64]
blk.2.ffn_up_shexp.weight
Q4_K
Q4_K
[2560, 3072]
blk.3
blk.3.attn_k.weight
Q4_K
Q4_K
[2560, 512]
blk.3.attn_norm.weight
F32
F32
[2560]
blk.3.attn_output.weight
Q4_K
Q4_K
[2560, 2560]
blk.3.attn_q.weight
Q4_K
Q4_K
[2560, 2560]
blk.3.attn_v.weight
Q4_K
Q4_K
[2560, 512]
blk.3.exp_probs_b.bias
F32
F32
[64]
blk.3.ffn_down_exps.weight
Q4_K
Q4_K
[1536, 2560, 64]
blk.3.ffn_down_shexp.weight
Q4_K
Q4_K
[3072, 2560]
blk.3.ffn_gate_exps.weight
Q4_K
Q4_K
[2560, 1536, 64]
blk.3.ffn_gate_inp.weight
F32
F32
[2560, 64]
blk.3.ffn_gate_shexp.weight
Q4_K
Q4_K
[2560, 3072]
blk.3.ffn_norm.weight
F32
F32
[2560]
blk.3.ffn_up_exps.weight
Q4_K
Q4_K
[2560, 1536, 64]
blk.3.ffn_up_shexp.weight
Q4_K
Q4_K
[2560, 3072]
blk.4
blk.4.attn_k.weight
Q4_K
Q4_K
[2560, 512]
blk.4.attn_norm.weight
F32
F32
[2560]
blk.4.attn_output.weight
Q4_K
Q4_K
[2560, 2560]
blk.4.attn_q.weight
Q4_K
Q4_K
[2560, 2560]
blk.4.attn_v.weight
Q4_K
Q4_K
[2560, 512]
blk.4.exp_probs_b.bias
F32
F32
[64]
blk.4.ffn_down_exps.weight
Q4_K
Q4_K
[1536, 2560, 64]
blk.4.ffn_down_shexp.weight
Q4_K
Q4_K
[3072, 2560]
blk.4.ffn_gate_exps.weight
Q4_K
Q4_K
[2560, 1536, 64]
blk.4.ffn_gate_inp.weight
F32
F32
[2560, 64]
blk.4.ffn_gate_shexp.weight
Q4_K
Q4_K
[2560, 3072]
blk.4.ffn_norm.weight
F32
F32
[2560]
blk.4.ffn_up_exps.weight
Q4_K
Q4_K
[2560, 1536, 64]
blk.4.ffn_up_shexp.weight
Q4_K
Q4_K
[2560, 3072]
blk.5
blk.5.attn_k.weight
Q4_K
Q4_K
[2560, 512]
blk.5.attn_norm.weight
F32
F32
[2560]
blk.5.attn_output.weight
Q4_K
Q4_K
[2560, 2560]
blk.5.attn_q.weight
Q4_K
Q4_K
[2560, 2560]
blk.5.attn_v.weight
Q6_K
Q6_K
[2560, 512]
blk.5.exp_probs_b.bias
F32
F32
[64]
blk.5.ffn_down_exps.weight
Q6_K
Q6_K
[1536, 2560, 64]
blk.5.ffn_down_shexp.weight
Q6_K
Q6_K
[3072, 2560]
blk.5.ffn_gate_exps.weight
Q4_K
Q4_K
[2560, 1536, 64]
blk.5.ffn_gate_inp.weight
F32
F32
[2560, 64]
blk.5.ffn_gate_shexp.weight
Q4_K
Q4_K
[2560, 3072]
blk.5.ffn_norm.weight
F32
F32
[2560]
blk.5.ffn_up_exps.weight
Q4_K
Q4_K
[2560, 1536, 64]
blk.5.ffn_up_shexp.weight
Q4_K
Q4_K
[2560, 3072]
blk.6
blk.6.attn_k.weight
Q4_K
Q4_K
[2560, 512]
blk.6.attn_norm.weight
F32
F32
[2560]
blk.6.attn_output.weight
Q4_K
Q4_K
[2560, 2560]
blk.6.attn_q.weight
Q4_K
Q4_K
[2560, 2560]
blk.6.attn_v.weight
Q4_K
Q4_K
[2560, 512]
blk.6.exp_probs_b.bias
F32
F32
[64]
blk.6.ffn_down_exps.weight
Q4_K
Q4_K
[1536, 2560, 64]
blk.6.ffn_down_shexp.weight
Q4_K
Q4_K
[3072, 2560]
blk.6.ffn_gate_exps.weight
Q4_K
Q4_K
[2560, 1536, 64]
blk.6.ffn_gate_inp.weight
F32
F32
[2560, 64]
blk.6.ffn_gate_shexp.weight
Q4_K
Q4_K
[2560, 3072]
blk.6.ffn_norm.weight
F32
F32
[2560]
blk.6.ffn_up_exps.weight
Q4_K
Q4_K
[2560, 1536, 64]
blk.6.ffn_up_shexp.weight
Q4_K
Q4_K
[2560, 3072]
blk.7
blk.7.attn_k.weight
Q4_K
Q4_K
[2560, 512]
blk.7.attn_norm.weight
F32
F32
[2560]
blk.7.attn_output.weight
Q4_K
Q4_K
[2560, 2560]
blk.7.attn_q.weight
Q4_K
Q4_K
[2560, 2560]
blk.7.attn_v.weight
Q4_K
Q4_K
[2560, 512]
blk.7.exp_probs_b.bias
F32
F32
[64]
blk.7.ffn_down_exps.weight
Q4_K
Q4_K
[1536, 2560, 64]
blk.7.ffn_down_shexp.weight
Q4_K
Q4_K
[3072, 2560]
blk.7.ffn_gate_exps.weight
Q4_K
Q4_K
[2560, 1536, 64]
blk.7.ffn_gate_inp.weight
F32
F32
[2560, 64]
blk.7.ffn_gate_shexp.weight
Q4_K
Q4_K
[2560, 3072]
blk.7.ffn_norm.weight
F32
F32
[2560]
blk.7.ffn_up_exps.weight
Q4_K
Q4_K
[2560, 1536, 64]
blk.7.ffn_up_shexp.weight
Q4_K
Q4_K
[2560, 3072]
blk.8
blk.8.attn_k.weight
Q4_K
Q4_K
[2560, 512]
blk.8.attn_norm.weight
F32
F32
[2560]
blk.8.attn_output.weight
Q4_K
Q4_K
[2560, 2560]
blk.8.attn_q.weight
Q4_K
Q4_K
[2560, 2560]
blk.8.attn_v.weight
Q6_K
Q6_K
[2560, 512]
blk.8.exp_probs_b.bias
F32
F32
[64]
blk.8.ffn_down_exps.weight
Q6_K
Q6_K
[1536, 2560, 64]
blk.8.ffn_down_shexp.weight
Q6_K
Q6_K
[3072, 2560]
blk.8.ffn_gate_exps.weight
Q4_K
Q4_K
[2560, 1536, 64]
blk.8.ffn_gate_inp.weight
F32
F32
[2560, 64]
blk.8.ffn_gate_shexp.weight
Q4_K
Q4_K
[2560, 3072]
blk.8.ffn_norm.weight
F32
F32
[2560]
blk.8.ffn_up_exps.weight
Q4_K
Q4_K
[2560, 1536, 64]
blk.8.ffn_up_shexp.weight
Q4_K
Q4_K
[2560, 3072]
blk.9
blk.9.attn_k.weight
Q4_K
Q4_K
[2560, 512]
blk.9.attn_norm.weight
F32
F32
[2560]
blk.9.attn_output.weight
Q4_K
Q4_K
[2560, 2560]
blk.9.attn_q.weight
Q4_K
Q4_K
[2560, 2560]
blk.9.attn_v.weight
Q4_K
Q4_K
[2560, 512]
blk.9.exp_probs_b.bias
F32
F32
[64]
blk.9.ffn_down_exps.weight
Q4_K
Q4_K
[1536, 2560, 64]
blk.9.ffn_down_shexp.weight
Q4_K
Q4_K
[3072, 2560]
blk.9.ffn_gate_exps.weight
Q4_K
Q4_K
[2560, 1536, 64]
blk.9.ffn_gate_inp.weight
F32
F32
[2560, 64]
blk.9.ffn_gate_shexp.weight
Q4_K
Q4_K
[2560, 3072]
blk.9.ffn_norm.weight
F32
F32
[2560]
blk.9.ffn_up_exps.weight
Q4_K
Q4_K
[2560, 1536, 64]
blk.9.ffn_up_shexp.weight
Q4_K
Q4_K
[2560, 3072]
blk.10
blk.10.attn_k.weight
Q4_K
Q4_K
[2560, 512]
blk.10.attn_norm.weight
F32
F32
[2560]
blk.10.attn_output.weight
Q4_K
Q4_K
[2560, 2560]
blk.10.attn_q.weight
Q4_K
Q4_K
[2560, 2560]
blk.10.attn_v.weight
Q4_K
Q4_K
[2560, 512]
blk.10.exp_probs_b.bias
F32
F32
[64]
blk.10.ffn_down_exps.weight
Q4_K
Q4_K
[1536, 2560, 64]
blk.10.ffn_down_shexp.weight
Q4_K
Q4_K
[3072, 2560]
blk.10.ffn_gate_exps.weight
Q4_K
Q4_K
[2560, 1536, 64]
blk.10.ffn_gate_inp.weight
F32
F32
[2560, 64]
blk.10.ffn_gate_shexp.weight
Q4_K
Q4_K
[2560, 3072]
blk.10.ffn_norm.weight
F32
F32
[2560]
blk.10.ffn_up_exps.weight
Q4_K
Q4_K
[2560, 1536, 64]
blk.10.ffn_up_shexp.weight
Q4_K
Q4_K
[2560, 3072]
blk.11
blk.11.attn_k.weight
Q4_K
Q4_K
[2560, 512]
blk.11.attn_norm.weight
F32
F32
[2560]
blk.11.attn_output.weight
Q4_K
Q4_K
[2560, 2560]
blk.11.attn_q.weight
Q4_K
Q4_K
[2560, 2560]
blk.11.attn_v.weight
Q6_K
Q6_K
[2560, 512]
blk.11.exp_probs_b.bias
F32
F32
[64]
blk.11.ffn_down_exps.weight
Q6_K
Q6_K
[1536, 2560, 64]
blk.11.ffn_down_shexp.weight
Q6_K
Q6_K
[3072, 2560]
blk.11.ffn_gate_exps.weight
Q4_K
Q4_K
[2560, 1536, 64]
blk.11.ffn_gate_inp.weight
F32
F32
[2560, 64]
blk.11.ffn_gate_shexp.weight
Q4_K
Q4_K
[2560, 3072]
blk.11.ffn_norm.weight
F32
F32
[2560]
blk.11.ffn_up_exps.weight
Q4_K
Q4_K
[2560, 1536, 64]
blk.11.ffn_up_shexp.weight
Q4_K
Q4_K
[2560, 3072]
blk.12
blk.12.attn_k.weight
Q4_K
Q4_K
[2560, 512]
blk.12.attn_norm.weight
F32
F32
[2560]
blk.12.attn_output.weight
Q4_K
Q4_K
[2560, 2560]
blk.12.attn_q.weight
Q4_K
Q4_K
[2560, 2560]
blk.12.attn_v.weight
Q4_K
Q4_K
[2560, 512]
blk.12.exp_probs_b.bias
F32
F32
[64]
blk.12.ffn_down_exps.weight
Q4_K
Q4_K
[1536, 2560, 64]
blk.12.ffn_down_shexp.weight
Q4_K
Q4_K
[3072, 2560]
blk.12.ffn_gate_exps.weight
Q4_K
Q4_K
[2560, 1536, 64]
blk.12.ffn_gate_inp.weight
F32
F32
[2560, 64]
blk.12.ffn_gate_shexp.weight
Q4_K
Q4_K
[2560, 3072]
blk.12.ffn_norm.weight
F32
F32
[2560]
blk.12.ffn_up_exps.weight
Q4_K
Q4_K
[2560, 1536, 64]
blk.12.ffn_up_shexp.weight
Q4_K
Q4_K
[2560, 3072]
blk.13
blk.13.attn_k.weight
Q4_K
Q4_K
[2560, 512]
blk.13.attn_norm.weight
F32
F32
[2560]
blk.13.attn_output.weight
Q4_K
Q4_K
[2560, 2560]
blk.13.attn_q.weight
Q4_K
Q4_K
[2560, 2560]
blk.13.attn_v.weight
Q4_K
Q4_K
[2560, 512]
blk.13.exp_probs_b.bias
F32
F32
[64]
blk.13.ffn_down_exps.weight
Q4_K
Q4_K
[1536, 2560, 64]
blk.13.ffn_down_shexp.weight
Q4_K
Q4_K
[3072, 2560]
blk.13.ffn_gate_exps.weight
Q4_K
Q4_K
[2560, 1536, 64]
blk.13.ffn_gate_inp.weight
F32
F32
[2560, 64]
blk.13.ffn_gate_shexp.weight
Q4_K
Q4_K
[2560, 3072]
blk.13.ffn_norm.weight
F32
F32
[2560]
blk.13.ffn_up_exps.weight
Q4_K
Q4_K
[2560, 1536, 64]
blk.13.ffn_up_shexp.weight
Q4_K
Q4_K
[2560, 3072]
blk.14
blk.14.attn_k.weight
Q4_K
Q4_K
[2560, 512]
blk.14.attn_norm.weight
F32
F32
[2560]
blk.14.attn_output.weight
Q4_K
Q4_K
[2560, 2560]
blk.14.attn_q.weight
Q4_K
Q4_K
[2560, 2560]
blk.14.attn_v.weight
Q6_K
Q6_K
[2560, 512]
blk.14.exp_probs_b.bias
F32
F32
[64]
blk.14.ffn_down_exps.weight
Q6_K
Q6_K
[1536, 2560, 64]
blk.14.ffn_down_shexp.weight
Q6_K
Q6_K
[3072, 2560]
blk.14.ffn_gate_exps.weight
Q4_K
Q4_K
[2560, 1536, 64]
blk.14.ffn_gate_inp.weight
F32
F32
[2560, 64]
blk.14.ffn_gate_shexp.weight
Q4_K
Q4_K
[2560, 3072]
blk.14.ffn_norm.weight
F32
F32
[2560]
blk.14.ffn_up_exps.weight
Q4_K
Q4_K
[2560, 1536, 64]
blk.14.ffn_up_shexp.weight
Q4_K
Q4_K
[2560, 3072]
blk.15
blk.15.attn_k.weight
Q4_K
Q4_K
[2560, 512]
blk.15.attn_norm.weight
F32
F32
[2560]
blk.15.attn_output.weight
Q4_K
Q4_K
[2560, 2560]
blk.15.attn_q.weight
Q4_K
Q4_K
[2560, 2560]
blk.15.attn_v.weight
Q4_K
Q4_K
[2560, 512]
blk.15.exp_probs_b.bias
F32
F32
[64]
blk.15.ffn_down_exps.weight
Q4_K
Q4_K
[1536, 2560, 64]
blk.15.ffn_down_shexp.weight
Q4_K
Q4_K
[3072, 2560]
blk.15.ffn_gate_exps.weight
Q4_K
Q4_K
[2560, 1536, 64]
blk.15.ffn_gate_inp.weight
F32
F32
[2560, 64]
blk.15.ffn_gate_shexp.weight
Q4_K
Q4_K
[2560, 3072]
blk.15.ffn_norm.weight
F32
F32
[2560]
blk.15.ffn_up_exps.weight
Q4_K
Q4_K
[2560, 1536, 64]
blk.15.ffn_up_shexp.weight
Q4_K
Q4_K
[2560, 3072]
blk.16
blk.16.attn_k.weight
Q4_K
Q4_K
[2560, 512]
blk.16.attn_norm.weight
F32
F32
[2560]
blk.16.attn_output.weight
Q4_K
Q4_K
[2560, 2560]
blk.16.attn_q.weight
Q4_K
Q4_K
[2560, 2560]
blk.16.attn_v.weight
Q4_K
Q4_K
[2560, 512]
blk.16.exp_probs_b.bias
F32
F32
[64]
blk.16.ffn_down_exps.weight
Q4_K
Q4_K
[1536, 2560, 64]
blk.16.ffn_down_shexp.weight
Q4_K
Q4_K
[3072, 2560]
blk.16.ffn_gate_exps.weight
Q4_K
Q4_K
[2560, 1536, 64]
blk.16.ffn_gate_inp.weight
F32
F32
[2560, 64]
blk.16.ffn_gate_shexp.weight
Q4_K
Q4_K
[2560, 3072]
blk.16.ffn_norm.weight
F32
F32
[2560]
blk.16.ffn_up_exps.weight
Q4_K
Q4_K
[2560, 1536, 64]
blk.16.ffn_up_shexp.weight
Q4_K
Q4_K
[2560, 3072]
blk.17
blk.17.attn_k.weight
Q4_K
Q4_K
[2560, 512]
blk.17.attn_norm.weight
F32
F32
[2560]
blk.17.attn_output.weight
Q4_K
Q4_K
[2560, 2560]
blk.17.attn_q.weight
Q4_K
Q4_K
[2560, 2560]
blk.17.attn_v.weight
Q6_K
Q6_K
[2560, 512]
blk.17.exp_probs_b.bias
F32
F32
[64]
blk.17.ffn_down_exps.weight
Q6_K
Q6_K
[1536, 2560, 64]
blk.17.ffn_down_shexp.weight
Q6_K
Q6_K
[3072, 2560]
blk.17.ffn_gate_exps.weight
Q4_K
Q4_K
[2560, 1536, 64]
blk.17.ffn_gate_inp.weight
F32
F32
[2560, 64]
blk.17.ffn_gate_shexp.weight
Q4_K
Q4_K
[2560, 3072]
blk.17.ffn_norm.weight
F32
F32
[2560]
blk.17.ffn_up_exps.weight
Q4_K
Q4_K
[2560, 1536, 64]
blk.17.ffn_up_shexp.weight
Q4_K
Q4_K
[2560, 3072]
blk.18
blk.18.attn_k.weight
Q4_K
Q4_K
[2560, 512]
blk.18.attn_norm.weight
F32
F32
[2560]
blk.18.attn_output.weight
Q4_K
Q4_K
[2560, 2560]
blk.18.attn_q.weight
Q4_K
Q4_K
[2560, 2560]
blk.18.attn_v.weight
Q4_K
Q4_K
[2560, 512]
blk.18.exp_probs_b.bias
F32
F32
[64]
blk.18.ffn_down_exps.weight
Q4_K
Q4_K
[1536, 2560, 64]
blk.18.ffn_down_shexp.weight
Q4_K
Q4_K
[3072, 2560]
blk.18.ffn_gate_exps.weight
Q4_K
Q4_K
[2560, 1536, 64]
blk.18.ffn_gate_inp.weight
F32
F32
[2560, 64]
blk.18.ffn_gate_shexp.weight
Q4_K
Q4_K
[2560, 3072]
blk.18.ffn_norm.weight
F32
F32
[2560]
blk.18.ffn_up_exps.weight
Q4_K
Q4_K
[2560, 1536, 64]
blk.18.ffn_up_shexp.weight
Q4_K
Q4_K
[2560, 3072]
blk.19
blk.19.attn_k.weight
Q4_K
Q4_K
[2560, 512]
blk.19.attn_norm.weight
F32
F32
[2560]
blk.19.attn_output.weight
Q4_K
Q4_K
[2560, 2560]
blk.19.attn_q.weight
Q4_K
Q4_K
[2560, 2560]
blk.19.attn_v.weight
Q4_K
Q4_K
[2560, 512]
blk.19.exp_probs_b.bias
F32
F32
[64]
blk.19.ffn_down_exps.weight
Q4_K
Q4_K
[1536, 2560, 64]
blk.19.ffn_down_shexp.weight
Q4_K
Q4_K
[3072, 2560]
blk.19.ffn_gate_exps.weight
Q4_K
Q4_K
[2560, 1536, 64]
blk.19.ffn_gate_inp.weight
F32
F32
[2560, 64]
blk.19.ffn_gate_shexp.weight
Q4_K
Q4_K
[2560, 3072]
blk.19.ffn_norm.weight
F32
F32
[2560]
blk.19.ffn_up_exps.weight
Q4_K
Q4_K
[2560, 1536, 64]
blk.19.ffn_up_shexp.weight
Q4_K
Q4_K
[2560, 3072]
blk.20
blk.20.attn_k.weight
Q4_K
Q4_K
[2560, 512]
blk.20.attn_norm.weight
F32
F32
[2560]
blk.20.attn_output.weight
Q4_K
Q4_K
[2560, 2560]
blk.20.attn_q.weight
Q4_K
Q4_K
[2560, 2560]
blk.20.attn_v.weight
Q6_K
Q6_K
[2560, 512]
blk.20.exp_probs_b.bias
F32
F32
[64]
blk.20.ffn_down_exps.weight
Q6_K
Q6_K
[1536, 2560, 64]
blk.20.ffn_down_shexp.weight
Q6_K
Q6_K
[3072, 2560]
blk.20.ffn_gate_exps.weight
Q4_K
Q4_K
[2560, 1536, 64]
blk.20.ffn_gate_inp.weight
F32
F32
[2560, 64]
blk.20.ffn_gate_shexp.weight
Q4_K
Q4_K
[2560, 3072]
blk.20.ffn_norm.weight
F32
F32
[2560]
blk.20.ffn_up_exps.weight
Q4_K
Q4_K
[2560, 1536, 64]
blk.20.ffn_up_shexp.weight
Q4_K
Q4_K
[2560, 3072]
blk.21
blk.21.attn_k.weight
Q4_K
Q4_K
[2560, 512]
blk.21.attn_norm.weight
F32
F32
[2560]
blk.21.attn_output.weight
Q4_K
Q4_K
[2560, 2560]
blk.21.attn_q.weight
Q4_K
Q4_K
[2560, 2560]
blk.21.attn_v.weight
Q4_K
Q4_K
[2560, 512]
blk.21.exp_probs_b.bias
F32
F32
[64]
blk.21.ffn_down_exps.weight
Q4_K
Q4_K
[1536, 2560, 64]
blk.21.ffn_down_shexp.weight
Q4_K
Q4_K
[3072, 2560]
blk.21.ffn_gate_exps.weight
Q4_K
Q4_K
[2560, 1536, 64]
blk.21.ffn_gate_inp.weight
F32
F32
[2560, 64]
blk.21.ffn_gate_shexp.weight
Q4_K
Q4_K
[2560, 3072]
blk.21.ffn_norm.weight
F32
F32
[2560]
blk.21.ffn_up_exps.weight
Q4_K
Q4_K
[2560, 1536, 64]
blk.21.ffn_up_shexp.weight
Q4_K
Q4_K
[2560, 3072]
blk.22
blk.22.attn_k.weight
Q4_K
Q4_K
[2560, 512]
blk.22.attn_norm.weight
F32
F32
[2560]
blk.22.attn_output.weight
Q4_K
Q4_K
[2560, 2560]
blk.22.attn_q.weight
Q4_K
Q4_K
[2560, 2560]
blk.22.attn_v.weight
Q4_K
Q4_K
[2560, 512]
blk.22.exp_probs_b.bias
F32
F32
[64]
blk.22.ffn_down_exps.weight
Q4_K
Q4_K
[1536, 2560, 64]
blk.22.ffn_down_shexp.weight
Q4_K
Q4_K
[3072, 2560]
blk.22.ffn_gate_exps.weight
Q4_K
Q4_K
[2560, 1536, 64]
blk.22.ffn_gate_inp.weight
F32
F32
[2560, 64]
blk.22.ffn_gate_shexp.weight
Q4_K
Q4_K
[2560, 3072]
blk.22.ffn_norm.weight
F32
F32
[2560]
blk.22.ffn_up_exps.weight
Q4_K
Q4_K
[2560, 1536, 64]
blk.22.ffn_up_shexp.weight
Q4_K
Q4_K
[2560, 3072]
blk.23
blk.23.attn_k.weight
Q4_K
Q4_K
[2560, 512]
blk.23.attn_norm.weight
F32
F32
[2560]
blk.23.attn_output.weight
Q4_K
Q4_K
[2560, 2560]
blk.23.attn_q.weight
Q4_K
Q4_K
[2560, 2560]
blk.23.attn_v.weight
Q6_K
Q6_K
[2560, 512]
blk.23.exp_probs_b.bias
F32
F32
[64]
blk.23.ffn_down_exps.weight
Q6_K
Q6_K
[1536, 2560, 64]
blk.23.ffn_down_shexp.weight
Q6_K
Q6_K
[3072, 2560]
blk.23.ffn_gate_exps.weight
Q4_K
Q4_K
[2560, 1536, 64]
blk.23.ffn_gate_inp.weight
F32
F32
[2560, 64]
blk.23.ffn_gate_shexp.weight
Q4_K
Q4_K
[2560, 3072]
blk.23.ffn_norm.weight
F32
F32
[2560]
blk.23.ffn_up_exps.weight
Q4_K
Q4_K
[2560, 1536, 64]
blk.23.ffn_up_shexp.weight
Q4_K
Q4_K
[2560, 3072]
blk.24
blk.24.attn_k.weight
Q4_K
Q4_K
[2560, 512]
blk.24.attn_norm.weight
F32
F32
[2560]
blk.24.attn_output.weight
Q4_K
Q4_K
[2560, 2560]
blk.24.attn_q.weight
Q4_K
Q4_K
[2560, 2560]
blk.24.attn_v.weight
Q6_K
Q6_K
[2560, 512]
blk.24.exp_probs_b.bias
F32
F32
[64]
blk.24.ffn_down_exps.weight
Q6_K
Q6_K
[1536, 2560, 64]
blk.24.ffn_down_shexp.weight
Q6_K
Q6_K
[3072, 2560]
blk.24.ffn_gate_exps.weight
Q4_K
Q4_K
[2560, 1536, 64]
blk.24.ffn_gate_inp.weight
F32
F32
[2560, 64]
blk.24.ffn_gate_shexp.weight
Q4_K
Q4_K
[2560, 3072]
blk.24.ffn_norm.weight
F32
F32
[2560]
blk.24.ffn_up_exps.weight
Q4_K
Q4_K
[2560, 1536, 64]
blk.24.ffn_up_shexp.weight
Q4_K
Q4_K
[2560, 3072]
blk.25
blk.25.attn_k.weight
Q4_K
Q4_K
[2560, 512]
blk.25.attn_norm.weight
F32
F32
[2560]
blk.25.attn_output.weight
Q4_K
Q4_K
[2560, 2560]
blk.25.attn_q.weight
Q4_K
Q4_K
[2560, 2560]
blk.25.attn_v.weight
Q6_K
Q6_K
[2560, 512]
blk.25.exp_probs_b.bias
F32
F32
[64]
blk.25.ffn_down_exps.weight
Q6_K
Q6_K
[1536, 2560, 64]
blk.25.ffn_down_shexp.weight
Q6_K
Q6_K
[3072, 2560]
blk.25.ffn_gate_exps.weight
Q4_K
Q4_K
[2560, 1536, 64]
blk.25.ffn_gate_inp.weight
F32
F32
[2560, 64]
blk.25.ffn_gate_shexp.weight
Q4_K
Q4_K
[2560, 3072]
blk.25.ffn_norm.weight
F32
F32
[2560]
blk.25.ffn_up_exps.weight
Q4_K
Q4_K
[2560, 1536, 64]
blk.25.ffn_up_shexp.weight
Q4_K
Q4_K
[2560, 3072]
blk.26
blk.26.attn_k.weight
Q4_K
Q4_K
[2560, 512]
blk.26.attn_norm.weight
F32
F32
[2560]
blk.26.attn_output.weight
Q4_K
Q4_K
[2560, 2560]
blk.26.attn_q.weight
Q4_K
Q4_K
[2560, 2560]
blk.26.attn_v.weight
Q6_K
Q6_K
[2560, 512]
blk.26.exp_probs_b.bias
F32
F32
[64]
blk.26.ffn_down_exps.weight
Q6_K
Q6_K
[1536, 2560, 64]
blk.26.ffn_down_shexp.weight
Q6_K
Q6_K
[3072, 2560]
blk.26.ffn_gate_exps.weight
Q4_K
Q4_K
[2560, 1536, 64]
blk.26.ffn_gate_inp.weight
F32
F32
[2560, 64]
blk.26.ffn_gate_shexp.weight
Q4_K
Q4_K
[2560, 3072]
blk.26.ffn_norm.weight
F32
F32
[2560]
blk.26.ffn_up_exps.weight
Q4_K
Q4_K
[2560, 1536, 64]
blk.26.ffn_up_shexp.weight
Q4_K
Q4_K
[2560, 3072]
blk.27
blk.27.attn_k.weight
Q4_K
Q4_K
[2560, 512]
blk.27.attn_norm.weight
F32
F32
[2560]
blk.27.attn_output.weight
Q4_K
Q4_K
[2560, 2560]
blk.27.attn_q.weight
Q4_K
Q4_K
[2560, 2560]
blk.27.attn_v.weight
Q6_K
Q6_K
[2560, 512]
blk.27.exp_probs_b.bias
F32
F32
[64]
blk.27.ffn_down_exps.weight
Q6_K
Q6_K
[1536, 2560, 64]
blk.27.ffn_down_shexp.weight
Q6_K
Q6_K
[3072, 2560]
blk.27.ffn_gate_exps.weight
Q4_K
Q4_K
[2560, 1536, 64]
blk.27.ffn_gate_inp.weight
F32
F32
[2560, 64]
blk.27.ffn_gate_shexp.weight
Q4_K
Q4_K
[2560, 3072]
blk.27.ffn_norm.weight
F32
F32
[2560]
blk.27.ffn_up_exps.weight
Q4_K
Q4_K
[2560, 1536, 64]
blk.27.ffn_up_shexp.weight
Q4_K
Q4_K
[2560, 3072]
output_norm.weight
F32
F32
[2560]