deepseek-v2.5

An upgraded version of DeekSeek-V2 that integrates the general and coding abilities of both DeepSeek-V2-Chat and DeepSeek-Coder-V2-Instruct.

236b

9,196 Pulls Updated 2 months ago

7 Tags

deepseek-v2.5:236b-q8_0 ... /

model

31c817e177eb · 251GB

Metadata

deepseek2.attention.head_count

128
deepseek2.attention.head_count_kv

128
deepseek2.attention.key_length

192
deepseek2.attention.kv_lora_rank

512
deepseek2.attention.layer_norm_rms_epsilon

1e-06
deepseek2.attention.q_lora_rank

1536
deepseek2.attention.value_length

128
deepseek2.block_count

60
deepseek2.context_length

163840
deepseek2.embedding_length

5120
deepseek2.expert_count

160
deepseek2.expert_feed_forward_length

1536
deepseek2.expert_shared_count

2
deepseek2.expert_used_count

6
deepseek2.expert_weights_scale

16
deepseek2.feed_forward_length

12288
deepseek2.leading_dense_block_count

1
deepseek2.rope.dimension_count

64
deepseek2.rope.freq_base

10000
deepseek2.rope.scaling.factor

40
deepseek2.rope.scaling.original_context_length

4096
deepseek2.rope.scaling.type

yarn
deepseek2.rope.scaling.yarn_log_multiplier

0.1
deepseek2.vocab_size

102400
general.architecture

deepseek2
general.basename

DeepSeek
general.file_type

7
general.license

other
general.license.link

https://github.com/deepseek-ai/DeepSeek-V2/blob/main/LICENSE-MODEL
general.license.name

deepseek
general.name

DeepSeek V2.5
general.quantization_version

2
general.size_label

160x14B
general.type

model
general.version

V2.5
tokenizer.ggml.add_bos_token

true
tokenizer.ggml.add_eos_token

false
tokenizer.ggml.bos_token_id

100000
tokenizer.ggml.eos_token_id

100001
tokenizer.ggml.merges

[Ġ Ġ, Ġ t, Ġ a, i n, h e, ...]
tokenizer.ggml.model

gpt2
tokenizer.ggml.padding_token_id

100001
tokenizer.ggml.pre

deepseek-llm
tokenizer.ggml.token_type

[1, 1, 1, 1, 1, ...]
tokenizer.ggml.tokens

[!, ", #, $, %, ...]

Tensor

Name

Type

Shape
token_embd.weight

Q8_0

[5120, 102400]

blk.0

blk.0.attn_kv_a_mqa.weight

Q8_0

[5120, 576]
blk.0.attn_kv_a_norm.weight

F32

[512]
blk.0.attn_kv_b.weight

Q8_0

[512, 32768]
blk.0.attn_norm.weight

F32

[5120]
blk.0.attn_output.weight

Q8_0

[16384, 5120]
blk.0.attn_q_a.weight

Q8_0

[5120, 1536]
blk.0.attn_q_a_norm.weight

F32

[1536]
blk.0.attn_q_b.weight

Q8_0

[1536, 24576]
blk.0.ffn_down.weight

Q8_0

[12288, 5120]
blk.0.ffn_gate.weight

Q8_0

[5120, 12288]
blk.0.ffn_norm.weight

F32

[5120]
blk.0.ffn_up.weight

Q8_0

[5120, 12288]

blk.1

blk.1.attn_kv_a_mqa.weight

Q8_0

[5120, 576]
blk.1.attn_kv_a_norm.weight

F32

[512]
blk.1.attn_kv_b.weight

Q8_0

[512, 32768]
blk.1.attn_norm.weight

F32

[5120]
blk.1.attn_output.weight

Q8_0

[16384, 5120]
blk.1.attn_q_a.weight

Q8_0

[5120, 1536]
blk.1.attn_q_a_norm.weight

F32

[1536]
blk.1.attn_q_b.weight

Q8_0

[1536, 24576]
blk.1.ffn_down_exps.weight

Q8_0

[1536, 5120, 160]
blk.1.ffn_down_shexp.weight

Q8_0

[3072, 5120]
blk.1.ffn_gate_exps.weight

Q8_0

[5120, 1536, 160]
blk.1.ffn_gate_inp.weight

F32

[5120, 160]
blk.1.ffn_gate_shexp.weight

Q8_0

[5120, 3072]
blk.1.ffn_norm.weight

F32

[5120]
blk.1.ffn_up_exps.weight

Q8_0

[5120, 1536, 160]
blk.1.ffn_up_shexp.weight

Q8_0

[5120, 3072]

blk.2

blk.2.attn_kv_a_mqa.weight

Q8_0

[5120, 576]
blk.2.attn_kv_a_norm.weight

F32

[512]
blk.2.attn_kv_b.weight

Q8_0

[512, 32768]
blk.2.attn_norm.weight

F32

[5120]
blk.2.attn_output.weight

Q8_0

[16384, 5120]
blk.2.attn_q_a.weight

Q8_0

[5120, 1536]
blk.2.attn_q_a_norm.weight

F32

[1536]
blk.2.attn_q_b.weight

Q8_0

[1536, 24576]
blk.2.ffn_down_exps.weight

Q8_0

[1536, 5120, 160]
blk.2.ffn_down_shexp.weight

Q8_0

[3072, 5120]
blk.2.ffn_gate_exps.weight

Q8_0

[5120, 1536, 160]
blk.2.ffn_gate_inp.weight

F32

[5120, 160]
blk.2.ffn_gate_shexp.weight

Q8_0

[5120, 3072]
blk.2.ffn_norm.weight

F32

[5120]
blk.2.ffn_up_exps.weight

Q8_0

[5120, 1536, 160]
blk.2.ffn_up_shexp.weight

Q8_0

[5120, 3072]

blk.3

blk.3.attn_kv_a_mqa.weight

Q8_0

[5120, 576]
blk.3.attn_kv_a_norm.weight

F32

[512]
blk.3.attn_kv_b.weight

Q8_0

[512, 32768]
blk.3.attn_norm.weight

F32

[5120]
blk.3.attn_output.weight

Q8_0

[16384, 5120]
blk.3.attn_q_a.weight

Q8_0

[5120, 1536]
blk.3.attn_q_a_norm.weight

F32

[1536]
blk.3.attn_q_b.weight

Q8_0

[1536, 24576]
blk.3.ffn_down_exps.weight

Q8_0

[1536, 5120, 160]
blk.3.ffn_down_shexp.weight

Q8_0

[3072, 5120]
blk.3.ffn_gate_exps.weight

Q8_0

[5120, 1536, 160]
blk.3.ffn_gate_inp.weight

F32

[5120, 160]
blk.3.ffn_gate_shexp.weight

Q8_0

[5120, 3072]
blk.3.ffn_norm.weight

F32

[5120]
blk.3.ffn_up_exps.weight

Q8_0

[5120, 1536, 160]
blk.3.ffn_up_shexp.weight

Q8_0

[5120, 3072]

blk.4

blk.4.attn_kv_a_mqa.weight

Q8_0

[5120, 576]
blk.4.attn_kv_a_norm.weight

F32

[512]
blk.4.attn_kv_b.weight

Q8_0

[512, 32768]
blk.4.attn_norm.weight

F32

[5120]
blk.4.attn_output.weight

Q8_0

[16384, 5120]
blk.4.attn_q_a.weight

Q8_0

[5120, 1536]
blk.4.attn_q_a_norm.weight

F32

[1536]
blk.4.attn_q_b.weight

Q8_0

[1536, 24576]
blk.4.ffn_down_exps.weight

Q8_0

[1536, 5120, 160]
blk.4.ffn_down_shexp.weight

Q8_0

[3072, 5120]
blk.4.ffn_gate_exps.weight

Q8_0

[5120, 1536, 160]
blk.4.ffn_gate_inp.weight

F32

[5120, 160]
blk.4.ffn_gate_shexp.weight

Q8_0

[5120, 3072]
blk.4.ffn_norm.weight

F32

[5120]
blk.4.ffn_up_exps.weight

Q8_0

[5120, 1536, 160]
blk.4.ffn_up_shexp.weight

Q8_0

[5120, 3072]

blk.5

blk.5.attn_kv_a_mqa.weight

Q8_0

[5120, 576]
blk.5.attn_kv_a_norm.weight

F32

[512]
blk.5.attn_kv_b.weight

Q8_0

[512, 32768]
blk.5.attn_norm.weight

F32

[5120]
blk.5.attn_output.weight

Q8_0

[16384, 5120]
blk.5.attn_q_a.weight

Q8_0

[5120, 1536]
blk.5.attn_q_a_norm.weight

F32

[1536]
blk.5.attn_q_b.weight

Q8_0

[1536, 24576]
blk.5.ffn_down_exps.weight

Q8_0

[1536, 5120, 160]
blk.5.ffn_down_shexp.weight

Q8_0

[3072, 5120]
blk.5.ffn_gate_exps.weight

Q8_0

[5120, 1536, 160]
blk.5.ffn_gate_inp.weight

F32

[5120, 160]
blk.5.ffn_gate_shexp.weight

Q8_0

[5120, 3072]
blk.5.ffn_norm.weight

F32

[5120]
blk.5.ffn_up_exps.weight

Q8_0

[5120, 1536, 160]
blk.5.ffn_up_shexp.weight

Q8_0

[5120, 3072]

blk.6

blk.6.attn_kv_a_mqa.weight

Q8_0

[5120, 576]
blk.6.attn_kv_a_norm.weight

F32

[512]
blk.6.attn_kv_b.weight

Q8_0

[512, 32768]
blk.6.attn_norm.weight

F32

[5120]
blk.6.attn_output.weight

Q8_0

[16384, 5120]
blk.6.attn_q_a.weight

Q8_0

[5120, 1536]
blk.6.attn_q_a_norm.weight

F32

[1536]
blk.6.attn_q_b.weight

Q8_0

[1536, 24576]
blk.6.ffn_down_exps.weight

Q8_0

[1536, 5120, 160]
blk.6.ffn_down_shexp.weight

Q8_0

[3072, 5120]
blk.6.ffn_gate_exps.weight

Q8_0

[5120, 1536, 160]
blk.6.ffn_gate_inp.weight

F32

[5120, 160]
blk.6.ffn_gate_shexp.weight

Q8_0

[5120, 3072]
blk.6.ffn_norm.weight

F32

[5120]
blk.6.ffn_up_exps.weight

Q8_0

[5120, 1536, 160]
blk.6.ffn_up_shexp.weight

Q8_0

[5120, 3072]

blk.7

blk.7.attn_kv_a_mqa.weight

Q8_0

[5120, 576]
blk.7.attn_kv_a_norm.weight

F32

[512]
blk.7.attn_kv_b.weight

Q8_0

[512, 32768]
blk.7.attn_norm.weight

F32

[5120]
blk.7.attn_output.weight

Q8_0

[16384, 5120]
blk.7.attn_q_a.weight

Q8_0

[5120, 1536]
blk.7.attn_q_a_norm.weight

F32

[1536]
blk.7.attn_q_b.weight

Q8_0

[1536, 24576]
blk.7.ffn_down_exps.weight

Q8_0

[1536, 5120, 160]
blk.7.ffn_down_shexp.weight

Q8_0

[3072, 5120]
blk.7.ffn_gate_exps.weight

Q8_0

[5120, 1536, 160]
blk.7.ffn_gate_inp.weight

F32

[5120, 160]
blk.7.ffn_gate_shexp.weight

Q8_0

[5120, 3072]
blk.7.ffn_norm.weight

F32

[5120]
blk.7.ffn_up_exps.weight

Q8_0

[5120, 1536, 160]
blk.7.ffn_up_shexp.weight

Q8_0

[5120, 3072]

blk.8

blk.8.attn_kv_a_mqa.weight

Q8_0

[5120, 576]
blk.8.attn_kv_a_norm.weight

F32

[512]
blk.8.attn_kv_b.weight

Q8_0

[512, 32768]
blk.8.attn_norm.weight

F32

[5120]
blk.8.attn_output.weight

Q8_0

[16384, 5120]
blk.8.attn_q_a.weight

Q8_0

[5120, 1536]
blk.8.attn_q_a_norm.weight

F32

[1536]
blk.8.attn_q_b.weight

Q8_0

[1536, 24576]
blk.8.ffn_down_exps.weight

Q8_0

[1536, 5120, 160]
blk.8.ffn_down_shexp.weight

Q8_0

[3072, 5120]
blk.8.ffn_gate_exps.weight

Q8_0

[5120, 1536, 160]
blk.8.ffn_gate_inp.weight

F32

[5120, 160]
blk.8.ffn_gate_shexp.weight

Q8_0

[5120, 3072]
blk.8.ffn_norm.weight

F32

[5120]
blk.8.ffn_up_exps.weight

Q8_0

[5120, 1536, 160]
blk.8.ffn_up_shexp.weight

Q8_0

[5120, 3072]

blk.9

blk.9.attn_kv_a_mqa.weight

Q8_0

[5120, 576]
blk.9.attn_kv_a_norm.weight

F32

[512]
blk.9.attn_kv_b.weight

Q8_0

[512, 32768]
blk.9.attn_norm.weight

F32

[5120]
blk.9.attn_output.weight

Q8_0

[16384, 5120]
blk.9.attn_q_a.weight

Q8_0

[5120, 1536]
blk.9.attn_q_a_norm.weight

F32

[1536]
blk.9.attn_q_b.weight

Q8_0

[1536, 24576]
blk.9.ffn_down_exps.weight

Q8_0

[1536, 5120, 160]
blk.9.ffn_down_shexp.weight

Q8_0

[3072, 5120]
blk.9.ffn_gate_exps.weight

Q8_0

[5120, 1536, 160]
blk.9.ffn_gate_inp.weight

F32

[5120, 160]
blk.9.ffn_gate_shexp.weight

Q8_0

[5120, 3072]
blk.9.ffn_norm.weight

F32

[5120]
blk.9.ffn_up_exps.weight

Q8_0

[5120, 1536, 160]
blk.9.ffn_up_shexp.weight

Q8_0

[5120, 3072]

blk.10

blk.10.attn_kv_a_mqa.weight

Q8_0

[5120, 576]
blk.10.attn_kv_a_norm.weight

F32

[512]
blk.10.attn_kv_b.weight

Q8_0

[512, 32768]
blk.10.attn_norm.weight

F32

[5120]
blk.10.attn_output.weight

Q8_0

[16384, 5120]
blk.10.attn_q_a.weight

Q8_0

[5120, 1536]
blk.10.attn_q_a_norm.weight

F32

[1536]
blk.10.attn_q_b.weight

Q8_0

[1536, 24576]
blk.10.ffn_down_exps.weight

Q8_0

[1536, 5120, 160]
blk.10.ffn_down_shexp.weight

Q8_0

[3072, 5120]
blk.10.ffn_gate_exps.weight

Q8_0

[5120, 1536, 160]
blk.10.ffn_gate_inp.weight

F32

[5120, 160]
blk.10.ffn_gate_shexp.weight

Q8_0

[5120, 3072]
blk.10.ffn_norm.weight

F32

[5120]
blk.10.ffn_up_exps.weight

Q8_0

[5120, 1536, 160]
blk.10.ffn_up_shexp.weight

Q8_0

[5120, 3072]

blk.11

blk.11.attn_kv_a_mqa.weight

Q8_0

[5120, 576]
blk.11.attn_kv_a_norm.weight

F32

[512]
blk.11.attn_kv_b.weight

Q8_0

[512, 32768]
blk.11.attn_norm.weight

F32

[5120]
blk.11.attn_output.weight

Q8_0

[16384, 5120]
blk.11.attn_q_a.weight

Q8_0

[5120, 1536]
blk.11.attn_q_a_norm.weight

F32

[1536]
blk.11.attn_q_b.weight

Q8_0

[1536, 24576]
blk.11.ffn_down_exps.weight

Q8_0

[1536, 5120, 160]
blk.11.ffn_down_shexp.weight

Q8_0

[3072, 5120]
blk.11.ffn_gate_exps.weight

Q8_0

[5120, 1536, 160]
blk.11.ffn_gate_inp.weight

F32

[5120, 160]
blk.11.ffn_gate_shexp.weight

Q8_0

[5120, 3072]
blk.11.ffn_norm.weight

F32

[5120]
blk.11.ffn_up_exps.weight

Q8_0

[5120, 1536, 160]
blk.11.ffn_up_shexp.weight

Q8_0

[5120, 3072]

blk.12

blk.12.attn_kv_a_mqa.weight

Q8_0

[5120, 576]
blk.12.attn_kv_a_norm.weight

F32

[512]
blk.12.attn_kv_b.weight

Q8_0

[512, 32768]
blk.12.attn_norm.weight

F32

[5120]
blk.12.attn_output.weight

Q8_0

[16384, 5120]
blk.12.attn_q_a.weight

Q8_0

[5120, 1536]
blk.12.attn_q_a_norm.weight

F32

[1536]
blk.12.attn_q_b.weight

Q8_0

[1536, 24576]
blk.12.ffn_down_exps.weight

Q8_0

[1536, 5120, 160]
blk.12.ffn_down_shexp.weight

Q8_0

[3072, 5120]
blk.12.ffn_gate_exps.weight

Q8_0

[5120, 1536, 160]
blk.12.ffn_gate_inp.weight

F32

[5120, 160]
blk.12.ffn_gate_shexp.weight

Q8_0

[5120, 3072]
blk.12.ffn_norm.weight

F32

[5120]
blk.12.ffn_up_exps.weight

Q8_0

[5120, 1536, 160]
blk.12.ffn_up_shexp.weight

Q8_0

[5120, 3072]

blk.13

blk.13.attn_kv_a_mqa.weight

Q8_0

[5120, 576]
blk.13.attn_kv_a_norm.weight

F32

[512]
blk.13.attn_kv_b.weight

Q8_0

[512, 32768]
blk.13.attn_norm.weight

F32

[5120]
blk.13.attn_output.weight

Q8_0

[16384, 5120]
blk.13.attn_q_a.weight

Q8_0

[5120, 1536]
blk.13.attn_q_a_norm.weight

F32

[1536]
blk.13.attn_q_b.weight

Q8_0

[1536, 24576]
blk.13.ffn_down_exps.weight

Q8_0

[1536, 5120, 160]
blk.13.ffn_down_shexp.weight

Q8_0

[3072, 5120]
blk.13.ffn_gate_exps.weight

Q8_0

[5120, 1536, 160]
blk.13.ffn_gate_inp.weight

F32

[5120, 160]
blk.13.ffn_gate_shexp.weight

Q8_0

[5120, 3072]
blk.13.ffn_norm.weight

F32

[5120]
blk.13.ffn_up_exps.weight

Q8_0

[5120, 1536, 160]
blk.13.ffn_up_shexp.weight

Q8_0

[5120, 3072]

blk.14

blk.14.attn_kv_a_mqa.weight

Q8_0

[5120, 576]
blk.14.attn_kv_a_norm.weight

F32

[512]
blk.14.attn_kv_b.weight

Q8_0

[512, 32768]
blk.14.attn_norm.weight

F32

[5120]
blk.14.attn_output.weight

Q8_0

[16384, 5120]
blk.14.attn_q_a.weight

Q8_0

[5120, 1536]
blk.14.attn_q_a_norm.weight

F32

[1536]
blk.14.attn_q_b.weight

Q8_0

[1536, 24576]
blk.14.ffn_down_exps.weight

Q8_0

[1536, 5120, 160]
blk.14.ffn_down_shexp.weight

Q8_0

[3072, 5120]
blk.14.ffn_gate_exps.weight

Q8_0

[5120, 1536, 160]
blk.14.ffn_gate_inp.weight

F32

[5120, 160]
blk.14.ffn_gate_shexp.weight

Q8_0

[5120, 3072]
blk.14.ffn_norm.weight

F32

[5120]
blk.14.ffn_up_exps.weight

Q8_0

[5120, 1536, 160]
blk.14.ffn_up_shexp.weight

Q8_0

[5120, 3072]

blk.15

blk.15.attn_kv_a_mqa.weight

Q8_0

[5120, 576]
blk.15.attn_kv_a_norm.weight

F32

[512]
blk.15.attn_kv_b.weight

Q8_0

[512, 32768]
blk.15.attn_norm.weight

F32

[5120]
blk.15.attn_output.weight

Q8_0

[16384, 5120]
blk.15.attn_q_a.weight

Q8_0

[5120, 1536]
blk.15.attn_q_a_norm.weight

F32

[1536]
blk.15.attn_q_b.weight

Q8_0

[1536, 24576]
blk.15.ffn_down_exps.weight

Q8_0

[1536, 5120, 160]
blk.15.ffn_down_shexp.weight

Q8_0

[3072, 5120]
blk.15.ffn_gate_exps.weight

Q8_0

[5120, 1536, 160]
blk.15.ffn_gate_inp.weight

F32

[5120, 160]
blk.15.ffn_gate_shexp.weight

Q8_0

[5120, 3072]
blk.15.ffn_norm.weight

F32

[5120]
blk.15.ffn_up_exps.weight

Q8_0

[5120, 1536, 160]
blk.15.ffn_up_shexp.weight

Q8_0

[5120, 3072]

blk.16

blk.16.attn_kv_a_mqa.weight

Q8_0

[5120, 576]
blk.16.attn_kv_a_norm.weight

F32

[512]
blk.16.attn_kv_b.weight

Q8_0

[512, 32768]
blk.16.attn_norm.weight

F32

[5120]
blk.16.attn_output.weight

Q8_0

[16384, 5120]
blk.16.attn_q_a.weight

Q8_0

[5120, 1536]
blk.16.attn_q_a_norm.weight

F32

[1536]
blk.16.attn_q_b.weight

Q8_0

[1536, 24576]
blk.16.ffn_down_exps.weight

Q8_0

[1536, 5120, 160]
blk.16.ffn_down_shexp.weight

Q8_0

[3072, 5120]
blk.16.ffn_gate_exps.weight

Q8_0

[5120, 1536, 160]
blk.16.ffn_gate_inp.weight

F32

[5120, 160]
blk.16.ffn_gate_shexp.weight

Q8_0

[5120, 3072]
blk.16.ffn_norm.weight

F32

[5120]
blk.16.ffn_up_exps.weight

Q8_0

[5120, 1536, 160]
blk.16.ffn_up_shexp.weight

Q8_0

[5120, 3072]

blk.17

blk.17.attn_kv_a_mqa.weight

Q8_0

[5120, 576]
blk.17.attn_kv_a_norm.weight

F32

[512]
blk.17.attn_kv_b.weight

Q8_0

[512, 32768]
blk.17.attn_norm.weight

F32

[5120]
blk.17.attn_output.weight

Q8_0

[16384, 5120]
blk.17.attn_q_a.weight

Q8_0

[5120, 1536]
blk.17.attn_q_a_norm.weight

F32

[1536]
blk.17.attn_q_b.weight

Q8_0

[1536, 24576]
blk.17.ffn_down_exps.weight

Q8_0

[1536, 5120, 160]
blk.17.ffn_down_shexp.weight

Q8_0

[3072, 5120]
blk.17.ffn_gate_exps.weight

Q8_0

[5120, 1536, 160]
blk.17.ffn_gate_inp.weight

F32

[5120, 160]
blk.17.ffn_gate_shexp.weight

Q8_0

[5120, 3072]
blk.17.ffn_norm.weight

F32

[5120]
blk.17.ffn_up_exps.weight

Q8_0

[5120, 1536, 160]
blk.17.ffn_up_shexp.weight

Q8_0

[5120, 3072]

blk.18

blk.18.attn_kv_a_mqa.weight

Q8_0

[5120, 576]
blk.18.attn_kv_a_norm.weight

F32

[512]
blk.18.attn_kv_b.weight

Q8_0

[512, 32768]
blk.18.attn_norm.weight

F32

[5120]
blk.18.attn_output.weight

Q8_0

[16384, 5120]
blk.18.attn_q_a.weight

Q8_0

[5120, 1536]
blk.18.attn_q_a_norm.weight

F32

[1536]
blk.18.attn_q_b.weight

Q8_0

[1536, 24576]
blk.18.ffn_down_exps.weight

Q8_0

[1536, 5120, 160]
blk.18.ffn_down_shexp.weight

Q8_0

[3072, 5120]
blk.18.ffn_gate_exps.weight

Q8_0

[5120, 1536, 160]
blk.18.ffn_gate_inp.weight

F32

[5120, 160]
blk.18.ffn_gate_shexp.weight

Q8_0

[5120, 3072]
blk.18.ffn_norm.weight

F32

[5120]
blk.18.ffn_up_exps.weight

Q8_0

[5120, 1536, 160]
blk.18.ffn_up_shexp.weight

Q8_0

[5120, 3072]

blk.19

blk.19.attn_kv_a_mqa.weight

Q8_0

[5120, 576]
blk.19.attn_kv_a_norm.weight

F32

[512]
blk.19.attn_kv_b.weight

Q8_0

[512, 32768]
blk.19.attn_norm.weight

F32

[5120]
blk.19.attn_output.weight

Q8_0

[16384, 5120]
blk.19.attn_q_a.weight

Q8_0

[5120, 1536]
blk.19.attn_q_a_norm.weight

F32

[1536]
blk.19.attn_q_b.weight

Q8_0

[1536, 24576]
blk.19.ffn_down_exps.weight

Q8_0

[1536, 5120, 160]
blk.19.ffn_down_shexp.weight

Q8_0

[3072, 5120]
blk.19.ffn_gate_exps.weight

Q8_0

[5120, 1536, 160]
blk.19.ffn_gate_inp.weight

F32

[5120, 160]
blk.19.ffn_gate_shexp.weight

Q8_0

[5120, 3072]
blk.19.ffn_norm.weight

F32

[5120]
blk.19.ffn_up_exps.weight

Q8_0

[5120, 1536, 160]
blk.19.ffn_up_shexp.weight

Q8_0

[5120, 3072]

blk.20

blk.20.attn_kv_a_mqa.weight

Q8_0

[5120, 576]
blk.20.attn_kv_a_norm.weight

F32

[512]
blk.20.attn_kv_b.weight

Q8_0

[512, 32768]
blk.20.attn_norm.weight

F32

[5120]
blk.20.attn_output.weight

Q8_0

[16384, 5120]
blk.20.attn_q_a.weight

Q8_0

[5120, 1536]
blk.20.attn_q_a_norm.weight

F32

[1536]
blk.20.attn_q_b.weight

Q8_0

[1536, 24576]
blk.20.ffn_down_exps.weight

Q8_0

[1536, 5120, 160]
blk.20.ffn_down_shexp.weight

Q8_0

[3072, 5120]
blk.20.ffn_gate_exps.weight

Q8_0

[5120, 1536, 160]
blk.20.ffn_gate_inp.weight

F32

[5120, 160]
blk.20.ffn_gate_shexp.weight

Q8_0

[5120, 3072]
blk.20.ffn_norm.weight

F32

[5120]
blk.20.ffn_up_exps.weight

Q8_0

[5120, 1536, 160]
blk.20.ffn_up_shexp.weight

Q8_0

[5120, 3072]

blk.21

blk.21.attn_kv_a_mqa.weight

Q8_0

[5120, 576]
blk.21.attn_kv_a_norm.weight

F32

[512]
blk.21.attn_kv_b.weight

Q8_0

[512, 32768]
blk.21.attn_norm.weight

F32

[5120]
blk.21.attn_output.weight

Q8_0

[16384, 5120]
blk.21.attn_q_a.weight

Q8_0

[5120, 1536]
blk.21.attn_q_a_norm.weight

F32

[1536]
blk.21.attn_q_b.weight

Q8_0

[1536, 24576]
blk.21.ffn_down_exps.weight

Q8_0

[1536, 5120, 160]
blk.21.ffn_down_shexp.weight

Q8_0

[3072, 5120]
blk.21.ffn_gate_exps.weight

Q8_0

[5120, 1536, 160]
blk.21.ffn_gate_inp.weight

F32

[5120, 160]
blk.21.ffn_gate_shexp.weight

Q8_0

[5120, 3072]
blk.21.ffn_norm.weight

F32

[5120]
blk.21.ffn_up_exps.weight

Q8_0

[5120, 1536, 160]
blk.21.ffn_up_shexp.weight

Q8_0

[5120, 3072]

blk.22

blk.22.attn_kv_a_mqa.weight

Q8_0

[5120, 576]
blk.22.attn_kv_a_norm.weight

F32

[512]
blk.22.attn_kv_b.weight

Q8_0

[512, 32768]
blk.22.attn_norm.weight

F32

[5120]
blk.22.attn_output.weight

Q8_0

[16384, 5120]
blk.22.attn_q_a.weight

Q8_0

[5120, 1536]
blk.22.attn_q_a_norm.weight

F32

[1536]
blk.22.attn_q_b.weight

Q8_0

[1536, 24576]
blk.22.ffn_down_exps.weight

Q8_0

[1536, 5120, 160]
blk.22.ffn_down_shexp.weight

Q8_0

[3072, 5120]
blk.22.ffn_gate_exps.weight

Q8_0

[5120, 1536, 160]
blk.22.ffn_gate_inp.weight

F32

[5120, 160]
blk.22.ffn_gate_shexp.weight

Q8_0

[5120, 3072]
blk.22.ffn_norm.weight

F32

[5120]
blk.22.ffn_up_exps.weight

Q8_0

[5120, 1536, 160]
blk.22.ffn_up_shexp.weight

Q8_0

[5120, 3072]

blk.23

blk.23.attn_kv_a_mqa.weight

Q8_0

[5120, 576]
blk.23.attn_kv_a_norm.weight

F32

[512]
blk.23.attn_kv_b.weight

Q8_0

[512, 32768]
blk.23.attn_norm.weight

F32

[5120]
blk.23.attn_output.weight

Q8_0

[16384, 5120]
blk.23.attn_q_a.weight

Q8_0

[5120, 1536]
blk.23.attn_q_a_norm.weight

F32

[1536]
blk.23.attn_q_b.weight

Q8_0

[1536, 24576]
blk.23.ffn_down_exps.weight

Q8_0

[1536, 5120, 160]
blk.23.ffn_down_shexp.weight

Q8_0

[3072, 5120]
blk.23.ffn_gate_exps.weight

Q8_0

[5120, 1536, 160]
blk.23.ffn_gate_inp.weight

F32

[5120, 160]
blk.23.ffn_gate_shexp.weight

Q8_0

[5120, 3072]
blk.23.ffn_norm.weight

F32

[5120]
blk.23.ffn_up_exps.weight

Q8_0

[5120, 1536, 160]
blk.23.ffn_up_shexp.weight

Q8_0

[5120, 3072]

blk.24

blk.24.attn_kv_a_mqa.weight

Q8_0

[5120, 576]
blk.24.attn_kv_a_norm.weight

F32

[512]
blk.24.attn_kv_b.weight

Q8_0

[512, 32768]
blk.24.attn_norm.weight

F32

[5120]
blk.24.attn_output.weight

Q8_0

[16384, 5120]
blk.24.attn_q_a.weight

Q8_0

[5120, 1536]
blk.24.attn_q_a_norm.weight

F32

[1536]
blk.24.attn_q_b.weight

Q8_0

[1536, 24576]
blk.24.ffn_down_exps.weight

Q8_0

[1536, 5120, 160]
blk.24.ffn_down_shexp.weight

Q8_0

[3072, 5120]
blk.24.ffn_gate_exps.weight

Q8_0

[5120, 1536, 160]
blk.24.ffn_gate_inp.weight

F32

[5120, 160]
blk.24.ffn_gate_shexp.weight

Q8_0

[5120, 3072]
blk.24.ffn_norm.weight

F32

[5120]
blk.24.ffn_up_exps.weight

Q8_0

[5120, 1536, 160]
blk.24.ffn_up_shexp.weight

Q8_0

[5120, 3072]

blk.25

blk.25.attn_kv_a_mqa.weight

Q8_0

[5120, 576]
blk.25.attn_kv_a_norm.weight

F32

[512]
blk.25.attn_kv_b.weight

Q8_0

[512, 32768]
blk.25.attn_norm.weight

F32

[5120]
blk.25.attn_output.weight

Q8_0

[16384, 5120]
blk.25.attn_q_a.weight

Q8_0

[5120, 1536]
blk.25.attn_q_a_norm.weight

F32

[1536]
blk.25.attn_q_b.weight

Q8_0

[1536, 24576]
blk.25.ffn_down_exps.weight

Q8_0

[1536, 5120, 160]
blk.25.ffn_down_shexp.weight

Q8_0

[3072, 5120]
blk.25.ffn_gate_exps.weight

Q8_0

[5120, 1536, 160]
blk.25.ffn_gate_inp.weight

F32

[5120, 160]
blk.25.ffn_gate_shexp.weight

Q8_0

[5120, 3072]
blk.25.ffn_norm.weight

F32

[5120]
blk.25.ffn_up_exps.weight

Q8_0

[5120, 1536, 160]
blk.25.ffn_up_shexp.weight

Q8_0

[5120, 3072]

blk.26

blk.26.attn_kv_a_mqa.weight

Q8_0

[5120, 576]
blk.26.attn_kv_a_norm.weight

F32

[512]
blk.26.attn_kv_b.weight

Q8_0

[512, 32768]
blk.26.attn_norm.weight

F32

[5120]
blk.26.attn_output.weight

Q8_0

[16384, 5120]
blk.26.attn_q_a.weight

Q8_0

[5120, 1536]
blk.26.attn_q_a_norm.weight

F32

[1536]
blk.26.attn_q_b.weight

Q8_0

[1536, 24576]
blk.26.ffn_down_exps.weight

Q8_0

[1536, 5120, 160]
blk.26.ffn_down_shexp.weight

Q8_0

[3072, 5120]
blk.26.ffn_gate_exps.weight

Q8_0

[5120, 1536, 160]
blk.26.ffn_gate_inp.weight

F32

[5120, 160]
blk.26.ffn_gate_shexp.weight

Q8_0

[5120, 3072]
blk.26.ffn_norm.weight

F32

[5120]
blk.26.ffn_up_exps.weight

Q8_0

[5120, 1536, 160]
blk.26.ffn_up_shexp.weight

Q8_0

[5120, 3072]

blk.27

blk.27.attn_kv_a_mqa.weight

Q8_0

[5120, 576]
blk.27.attn_kv_a_norm.weight

F32

[512]
blk.27.attn_kv_b.weight

Q8_0

[512, 32768]
blk.27.attn_norm.weight

F32

[5120]
blk.27.attn_output.weight

Q8_0

[16384, 5120]
blk.27.attn_q_a.weight

Q8_0

[5120, 1536]
blk.27.attn_q_a_norm.weight

F32

[1536]
blk.27.attn_q_b.weight

Q8_0

[1536, 24576]
blk.27.ffn_down_exps.weight

Q8_0

[1536, 5120, 160]
blk.27.ffn_down_shexp.weight

Q8_0

[3072, 5120]
blk.27.ffn_gate_exps.weight

Q8_0

[5120, 1536, 160]
blk.27.ffn_gate_inp.weight

F32

[5120, 160]
blk.27.ffn_gate_shexp.weight

Q8_0

[5120, 3072]
blk.27.ffn_norm.weight

F32

[5120]
blk.27.ffn_up_exps.weight

Q8_0

[5120, 1536, 160]
blk.27.ffn_up_shexp.weight

Q8_0

[5120, 3072]

blk.28

blk.28.attn_kv_a_mqa.weight

Q8_0

[5120, 576]
blk.28.attn_kv_a_norm.weight

F32

[512]
blk.28.attn_kv_b.weight

Q8_0

[512, 32768]
blk.28.attn_norm.weight

F32

[5120]
blk.28.attn_output.weight

Q8_0

[16384, 5120]
blk.28.attn_q_a.weight

Q8_0

[5120, 1536]
blk.28.attn_q_a_norm.weight

F32

[1536]
blk.28.attn_q_b.weight

Q8_0

[1536, 24576]
blk.28.ffn_down_exps.weight

Q8_0

[1536, 5120, 160]
blk.28.ffn_down_shexp.weight

Q8_0

[3072, 5120]
blk.28.ffn_gate_exps.weight

Q8_0

[5120, 1536, 160]
blk.28.ffn_gate_inp.weight

F32

[5120, 160]
blk.28.ffn_gate_shexp.weight

Q8_0

[5120, 3072]
blk.28.ffn_norm.weight

F32

[5120]
blk.28.ffn_up_exps.weight

Q8_0

[5120, 1536, 160]
blk.28.ffn_up_shexp.weight

Q8_0

[5120, 3072]

blk.29

blk.29.attn_kv_a_mqa.weight

Q8_0

[5120, 576]
blk.29.attn_kv_a_norm.weight

F32

[512]
blk.29.attn_kv_b.weight

Q8_0

[512, 32768]
blk.29.attn_norm.weight

F32

[5120]
blk.29.attn_output.weight

Q8_0

[16384, 5120]
blk.29.attn_q_a.weight

Q8_0

[5120, 1536]
blk.29.attn_q_a_norm.weight

F32

[1536]
blk.29.attn_q_b.weight

Q8_0

[1536, 24576]
blk.29.ffn_down_exps.weight

Q8_0

[1536, 5120, 160]
blk.29.ffn_down_shexp.weight

Q8_0

[3072, 5120]
blk.29.ffn_gate_exps.weight

Q8_0

[5120, 1536, 160]
blk.29.ffn_gate_inp.weight

F32

[5120, 160]
blk.29.ffn_gate_shexp.weight

Q8_0

[5120, 3072]
blk.29.ffn_norm.weight

F32

[5120]
blk.29.ffn_up_exps.weight

Q8_0

[5120, 1536, 160]
blk.29.ffn_up_shexp.weight

Q8_0

[5120, 3072]

blk.30

blk.30.attn_kv_a_mqa.weight

Q8_0

[5120, 576]
blk.30.attn_kv_a_norm.weight

F32

[512]
blk.30.attn_kv_b.weight

Q8_0

[512, 32768]
blk.30.attn_norm.weight

F32

[5120]
blk.30.attn_output.weight

Q8_0

[16384, 5120]
blk.30.attn_q_a.weight

Q8_0

[5120, 1536]
blk.30.attn_q_a_norm.weight

F32

[1536]
blk.30.attn_q_b.weight

Q8_0

[1536, 24576]
blk.30.ffn_down_exps.weight

Q8_0

[1536, 5120, 160]
blk.30.ffn_down_shexp.weight

Q8_0

[3072, 5120]
blk.30.ffn_gate_exps.weight

Q8_0

[5120, 1536, 160]
blk.30.ffn_gate_inp.weight

F32

[5120, 160]
blk.30.ffn_gate_shexp.weight

Q8_0

[5120, 3072]
blk.30.ffn_norm.weight

F32

[5120]
blk.30.ffn_up_exps.weight

Q8_0

[5120, 1536, 160]
blk.30.ffn_up_shexp.weight

Q8_0

[5120, 3072]

blk.31

blk.31.attn_kv_a_mqa.weight

Q8_0

[5120, 576]
blk.31.attn_kv_a_norm.weight

F32

[512]
blk.31.attn_kv_b.weight

Q8_0

[512, 32768]
blk.31.attn_norm.weight

F32

[5120]
blk.31.attn_output.weight

Q8_0

[16384, 5120]
blk.31.attn_q_a.weight

Q8_0

[5120, 1536]
blk.31.attn_q_a_norm.weight

F32

[1536]
blk.31.attn_q_b.weight

Q8_0

[1536, 24576]
blk.31.ffn_down_exps.weight

Q8_0

[1536, 5120, 160]
blk.31.ffn_down_shexp.weight

Q8_0

[3072, 5120]
blk.31.ffn_gate_exps.weight

Q8_0

[5120, 1536, 160]
blk.31.ffn_gate_inp.weight

F32

[5120, 160]
blk.31.ffn_gate_shexp.weight

Q8_0

[5120, 3072]
blk.31.ffn_norm.weight

F32

[5120]
blk.31.ffn_up_exps.weight

Q8_0

[5120, 1536, 160]
blk.31.ffn_up_shexp.weight

Q8_0

[5120, 3072]

blk.32

blk.32.attn_kv_a_mqa.weight

Q8_0

[5120, 576]
blk.32.attn_kv_a_norm.weight

F32

[512]
blk.32.attn_kv_b.weight

Q8_0

[512, 32768]
blk.32.attn_norm.weight

F32

[5120]
blk.32.attn_output.weight

Q8_0

[16384, 5120]
blk.32.attn_q_a.weight

Q8_0

[5120, 1536]
blk.32.attn_q_a_norm.weight

F32

[1536]
blk.32.attn_q_b.weight

Q8_0

[1536, 24576]
blk.32.ffn_down_exps.weight

Q8_0

[1536, 5120, 160]
blk.32.ffn_down_shexp.weight

Q8_0

[3072, 5120]
blk.32.ffn_gate_exps.weight

Q8_0

[5120, 1536, 160]
blk.32.ffn_gate_inp.weight

F32

[5120, 160]
blk.32.ffn_gate_shexp.weight

Q8_0

[5120, 3072]
blk.32.ffn_norm.weight

F32

[5120]
blk.32.ffn_up_exps.weight

Q8_0

[5120, 1536, 160]
blk.32.ffn_up_shexp.weight

Q8_0

[5120, 3072]

blk.33

blk.33.attn_kv_a_mqa.weight

Q8_0

[5120, 576]
blk.33.attn_kv_a_norm.weight

F32

[512]
blk.33.attn_kv_b.weight

Q8_0

[512, 32768]
blk.33.attn_norm.weight

F32

[5120]
blk.33.attn_output.weight

Q8_0

[16384, 5120]
blk.33.attn_q_a.weight

Q8_0

[5120, 1536]
blk.33.attn_q_a_norm.weight

F32

[1536]
blk.33.attn_q_b.weight

Q8_0

[1536, 24576]
blk.33.ffn_down_exps.weight

Q8_0

[1536, 5120, 160]
blk.33.ffn_down_shexp.weight

Q8_0

[3072, 5120]
blk.33.ffn_gate_exps.weight

Q8_0

[5120, 1536, 160]
blk.33.ffn_gate_inp.weight

F32

[5120, 160]
blk.33.ffn_gate_shexp.weight

Q8_0

[5120, 3072]
blk.33.ffn_norm.weight

F32

[5120]
blk.33.ffn_up_exps.weight

Q8_0

[5120, 1536, 160]
blk.33.ffn_up_shexp.weight

Q8_0

[5120, 3072]

blk.34

blk.34.attn_kv_a_mqa.weight

Q8_0

[5120, 576]
blk.34.attn_kv_a_norm.weight

F32

[512]
blk.34.attn_kv_b.weight

Q8_0

[512, 32768]
blk.34.attn_norm.weight

F32

[5120]
blk.34.attn_output.weight

Q8_0

[16384, 5120]
blk.34.attn_q_a.weight

Q8_0

[5120, 1536]
blk.34.attn_q_a_norm.weight

F32

[1536]
blk.34.attn_q_b.weight

Q8_0

[1536, 24576]
blk.34.ffn_down_exps.weight

Q8_0

[1536, 5120, 160]
blk.34.ffn_down_shexp.weight

Q8_0

[3072, 5120]
blk.34.ffn_gate_exps.weight

Q8_0

[5120, 1536, 160]
blk.34.ffn_gate_inp.weight

F32

[5120, 160]
blk.34.ffn_gate_shexp.weight

Q8_0

[5120, 3072]
blk.34.ffn_norm.weight

F32

[5120]
blk.34.ffn_up_exps.weight

Q8_0

[5120, 1536, 160]
blk.34.ffn_up_shexp.weight

Q8_0

[5120, 3072]

blk.35

blk.35.attn_kv_a_mqa.weight

Q8_0

[5120, 576]
blk.35.attn_kv_a_norm.weight

F32

[512]
blk.35.attn_kv_b.weight

Q8_0

[512, 32768]
blk.35.attn_norm.weight

F32

[5120]
blk.35.attn_output.weight

Q8_0

[16384, 5120]
blk.35.attn_q_a.weight

Q8_0

[5120, 1536]
blk.35.attn_q_a_norm.weight

F32

[1536]
blk.35.attn_q_b.weight

Q8_0

[1536, 24576]
blk.35.ffn_down_exps.weight

Q8_0

[1536, 5120, 160]
blk.35.ffn_down_shexp.weight

Q8_0

[3072, 5120]
blk.35.ffn_gate_exps.weight

Q8_0

[5120, 1536, 160]
blk.35.ffn_gate_inp.weight

F32

[5120, 160]
blk.35.ffn_gate_shexp.weight

Q8_0

[5120, 3072]
blk.35.ffn_norm.weight

F32

[5120]
blk.35.ffn_up_exps.weight

Q8_0

[5120, 1536, 160]
blk.35.ffn_up_shexp.weight

Q8_0

[5120, 3072]

blk.36

blk.36.attn_kv_a_mqa.weight

Q8_0

[5120, 576]
blk.36.attn_kv_a_norm.weight

F32

[512]
blk.36.attn_kv_b.weight

Q8_0

[512, 32768]
blk.36.attn_norm.weight

F32

[5120]
blk.36.attn_output.weight

Q8_0

[16384, 5120]
blk.36.attn_q_a.weight

Q8_0

[5120, 1536]
blk.36.attn_q_a_norm.weight

F32

[1536]
blk.36.attn_q_b.weight

Q8_0

[1536, 24576]
blk.36.ffn_down_exps.weight

Q8_0

[1536, 5120, 160]
blk.36.ffn_down_shexp.weight

Q8_0

[3072, 5120]
blk.36.ffn_gate_exps.weight

Q8_0

[5120, 1536, 160]
blk.36.ffn_gate_inp.weight

F32

[5120, 160]
blk.36.ffn_gate_shexp.weight

Q8_0

[5120, 3072]
blk.36.ffn_norm.weight

F32

[5120]
blk.36.ffn_up_exps.weight

Q8_0

[5120, 1536, 160]
blk.36.ffn_up_shexp.weight

Q8_0

[5120, 3072]

blk.37

blk.37.attn_kv_a_mqa.weight

Q8_0

[5120, 576]
blk.37.attn_kv_a_norm.weight

F32

[512]
blk.37.attn_kv_b.weight

Q8_0

[512, 32768]
blk.37.attn_norm.weight

F32

[5120]
blk.37.attn_output.weight

Q8_0

[16384, 5120]
blk.37.attn_q_a.weight

Q8_0

[5120, 1536]
blk.37.attn_q_a_norm.weight

F32

[1536]
blk.37.attn_q_b.weight

Q8_0

[1536, 24576]
blk.37.ffn_down_exps.weight

Q8_0

[1536, 5120, 160]
blk.37.ffn_down_shexp.weight

Q8_0

[3072, 5120]
blk.37.ffn_gate_exps.weight

Q8_0

[5120, 1536, 160]
blk.37.ffn_gate_inp.weight

F32

[5120, 160]
blk.37.ffn_gate_shexp.weight

Q8_0

[5120, 3072]
blk.37.ffn_norm.weight

F32

[5120]
blk.37.ffn_up_exps.weight

Q8_0

[5120, 1536, 160]
blk.37.ffn_up_shexp.weight

Q8_0

[5120, 3072]

blk.38

blk.38.attn_kv_a_mqa.weight

Q8_0

[5120, 576]
blk.38.attn_kv_a_norm.weight

F32

[512]
blk.38.attn_kv_b.weight

Q8_0

[512, 32768]
blk.38.attn_norm.weight

F32

[5120]
blk.38.attn_output.weight

Q8_0

[16384, 5120]
blk.38.attn_q_a.weight

Q8_0

[5120, 1536]
blk.38.attn_q_a_norm.weight

F32

[1536]
blk.38.attn_q_b.weight

Q8_0

[1536, 24576]
blk.38.ffn_down_exps.weight

Q8_0

[1536, 5120, 160]
blk.38.ffn_down_shexp.weight

Q8_0

[3072, 5120]
blk.38.ffn_gate_exps.weight

Q8_0

[5120, 1536, 160]
blk.38.ffn_gate_inp.weight

F32

[5120, 160]
blk.38.ffn_gate_shexp.weight

Q8_0

[5120, 3072]
blk.38.ffn_norm.weight

F32

[5120]
blk.38.ffn_up_exps.weight

Q8_0

[5120, 1536, 160]
blk.38.ffn_up_shexp.weight

Q8_0

[5120, 3072]

blk.39

blk.39.attn_kv_a_mqa.weight

Q8_0

[5120, 576]
blk.39.attn_kv_a_norm.weight

F32

[512]
blk.39.attn_kv_b.weight

Q8_0

[512, 32768]
blk.39.attn_norm.weight

F32

[5120]
blk.39.attn_output.weight

Q8_0

[16384, 5120]
blk.39.attn_q_a.weight

Q8_0

[5120, 1536]
blk.39.attn_q_a_norm.weight

F32

[1536]
blk.39.attn_q_b.weight

Q8_0

[1536, 24576]
blk.39.ffn_down_exps.weight

Q8_0

[1536, 5120, 160]
blk.39.ffn_down_shexp.weight

Q8_0

[3072, 5120]
blk.39.ffn_gate_exps.weight

Q8_0

[5120, 1536, 160]
blk.39.ffn_gate_inp.weight

F32

[5120, 160]
blk.39.ffn_gate_shexp.weight

Q8_0

[5120, 3072]
blk.39.ffn_norm.weight

F32

[5120]
blk.39.ffn_up_exps.weight

Q8_0

[5120, 1536, 160]
blk.39.ffn_up_shexp.weight

Q8_0

[5120, 3072]

blk.40

blk.40.attn_kv_a_mqa.weight

Q8_0

[5120, 576]
blk.40.attn_kv_a_norm.weight

F32

[512]
blk.40.attn_kv_b.weight

Q8_0

[512, 32768]
blk.40.attn_norm.weight

F32

[5120]
blk.40.attn_output.weight

Q8_0

[16384, 5120]
blk.40.attn_q_a.weight

Q8_0

[5120, 1536]
blk.40.attn_q_a_norm.weight

F32

[1536]
blk.40.attn_q_b.weight

Q8_0

[1536, 24576]
blk.40.ffn_down_exps.weight

Q8_0

[1536, 5120, 160]
blk.40.ffn_down_shexp.weight

Q8_0

[3072, 5120]
blk.40.ffn_gate_exps.weight

Q8_0

[5120, 1536, 160]
blk.40.ffn_gate_inp.weight

F32

[5120, 160]
blk.40.ffn_gate_shexp.weight

Q8_0

[5120, 3072]
blk.40.ffn_norm.weight

F32

[5120]
blk.40.ffn_up_exps.weight

Q8_0

[5120, 1536, 160]
blk.40.ffn_up_shexp.weight

Q8_0

[5120, 3072]

blk.41

blk.41.attn_kv_a_mqa.weight

Q8_0

[5120, 576]
blk.41.attn_kv_a_norm.weight

F32

[512]
blk.41.attn_kv_b.weight

Q8_0

[512, 32768]
blk.41.attn_norm.weight

F32

[5120]
blk.41.attn_output.weight

Q8_0

[16384, 5120]
blk.41.attn_q_a.weight

Q8_0

[5120, 1536]
blk.41.attn_q_a_norm.weight

F32

[1536]
blk.41.attn_q_b.weight

Q8_0

[1536, 24576]
blk.41.ffn_down_exps.weight

Q8_0

[1536, 5120, 160]
blk.41.ffn_down_shexp.weight

Q8_0

[3072, 5120]
blk.41.ffn_gate_exps.weight

Q8_0

[5120, 1536, 160]
blk.41.ffn_gate_inp.weight

F32

[5120, 160]
blk.41.ffn_gate_shexp.weight

Q8_0

[5120, 3072]
blk.41.ffn_norm.weight

F32

[5120]
blk.41.ffn_up_exps.weight

Q8_0

[5120, 1536, 160]
blk.41.ffn_up_shexp.weight

Q8_0

[5120, 3072]

blk.42

blk.42.attn_kv_a_mqa.weight

Q8_0

[5120, 576]
blk.42.attn_kv_a_norm.weight

F32

[512]
blk.42.attn_kv_b.weight

Q8_0

[512, 32768]
blk.42.attn_norm.weight

F32

[5120]
blk.42.attn_output.weight

Q8_0

[16384, 5120]
blk.42.attn_q_a.weight

Q8_0

[5120, 1536]
blk.42.attn_q_a_norm.weight

F32

[1536]
blk.42.attn_q_b.weight

Q8_0

[1536, 24576]
blk.42.ffn_down_exps.weight

Q8_0

[1536, 5120, 160]
blk.42.ffn_down_shexp.weight

Q8_0

[3072, 5120]
blk.42.ffn_gate_exps.weight

Q8_0

[5120, 1536, 160]
blk.42.ffn_gate_inp.weight

F32

[5120, 160]
blk.42.ffn_gate_shexp.weight

Q8_0

[5120, 3072]
blk.42.ffn_norm.weight

F32

[5120]
blk.42.ffn_up_exps.weight

Q8_0

[5120, 1536, 160]
blk.42.ffn_up_shexp.weight

Q8_0

[5120, 3072]

blk.43

blk.43.attn_kv_a_mqa.weight

Q8_0

[5120, 576]
blk.43.attn_kv_a_norm.weight

F32

[512]
blk.43.attn_kv_b.weight

Q8_0

[512, 32768]
blk.43.attn_norm.weight

F32

[5120]
blk.43.attn_output.weight

Q8_0

[16384, 5120]
blk.43.attn_q_a.weight

Q8_0

[5120, 1536]
blk.43.attn_q_a_norm.weight

F32

[1536]
blk.43.attn_q_b.weight

Q8_0

[1536, 24576]
blk.43.ffn_down_exps.weight

Q8_0

[1536, 5120, 160]
blk.43.ffn_down_shexp.weight

Q8_0

[3072, 5120]
blk.43.ffn_gate_exps.weight

Q8_0

[5120, 1536, 160]
blk.43.ffn_gate_inp.weight

F32

[5120, 160]
blk.43.ffn_gate_shexp.weight

Q8_0

[5120, 3072]
blk.43.ffn_norm.weight

F32

[5120]
blk.43.ffn_up_exps.weight

Q8_0

[5120, 1536, 160]
blk.43.ffn_up_shexp.weight

Q8_0

[5120, 3072]

blk.44

blk.44.attn_kv_a_mqa.weight

Q8_0

[5120, 576]
blk.44.attn_kv_a_norm.weight

F32

[512]
blk.44.attn_kv_b.weight

Q8_0

[512, 32768]
blk.44.attn_norm.weight

F32

[5120]
blk.44.attn_output.weight

Q8_0

[16384, 5120]
blk.44.attn_q_a.weight

Q8_0

[5120, 1536]
blk.44.attn_q_a_norm.weight

F32

[1536]
blk.44.attn_q_b.weight

Q8_0

[1536, 24576]
blk.44.ffn_down_exps.weight

Q8_0

[1536, 5120, 160]
blk.44.ffn_down_shexp.weight

Q8_0

[3072, 5120]
blk.44.ffn_gate_exps.weight

Q8_0

[5120, 1536, 160]
blk.44.ffn_gate_inp.weight

F32

[5120, 160]
blk.44.ffn_gate_shexp.weight

Q8_0

[5120, 3072]
blk.44.ffn_norm.weight

F32

[5120]
blk.44.ffn_up_exps.weight

Q8_0

[5120, 1536, 160]
blk.44.ffn_up_shexp.weight

Q8_0

[5120, 3072]

blk.45

blk.45.attn_kv_a_mqa.weight

Q8_0

[5120, 576]
blk.45.attn_kv_a_norm.weight

F32

[512]
blk.45.attn_kv_b.weight

Q8_0

[512, 32768]
blk.45.attn_norm.weight

F32

[5120]
blk.45.attn_output.weight

Q8_0

[16384, 5120]
blk.45.attn_q_a.weight

Q8_0

[5120, 1536]
blk.45.attn_q_a_norm.weight

F32

[1536]
blk.45.attn_q_b.weight

Q8_0

[1536, 24576]
blk.45.ffn_down_exps.weight

Q8_0

[1536, 5120, 160]
blk.45.ffn_down_shexp.weight

Q8_0

[3072, 5120]
blk.45.ffn_gate_exps.weight

Q8_0

[5120, 1536, 160]
blk.45.ffn_gate_inp.weight

F32

[5120, 160]
blk.45.ffn_gate_shexp.weight

Q8_0

[5120, 3072]
blk.45.ffn_norm.weight

F32

[5120]
blk.45.ffn_up_exps.weight

Q8_0

[5120, 1536, 160]
blk.45.ffn_up_shexp.weight

Q8_0

[5120, 3072]

blk.46

blk.46.attn_kv_a_mqa.weight

Q8_0

[5120, 576]
blk.46.attn_kv_a_norm.weight

F32

[512]
blk.46.attn_kv_b.weight

Q8_0

[512, 32768]
blk.46.attn_norm.weight

F32

[5120]
blk.46.attn_output.weight

Q8_0

[16384, 5120]
blk.46.attn_q_a.weight

Q8_0

[5120, 1536]
blk.46.attn_q_a_norm.weight

F32

[1536]
blk.46.attn_q_b.weight

Q8_0

[1536, 24576]
blk.46.ffn_down_exps.weight

Q8_0

[1536, 5120, 160]
blk.46.ffn_down_shexp.weight

Q8_0

[3072, 5120]
blk.46.ffn_gate_exps.weight

Q8_0

[5120, 1536, 160]
blk.46.ffn_gate_inp.weight

F32

[5120, 160]
blk.46.ffn_gate_shexp.weight

Q8_0

[5120, 3072]
blk.46.ffn_norm.weight

F32

[5120]
blk.46.ffn_up_exps.weight

Q8_0

[5120, 1536, 160]
blk.46.ffn_up_shexp.weight

Q8_0

[5120, 3072]

blk.47

blk.47.attn_kv_a_mqa.weight

Q8_0

[5120, 576]
blk.47.attn_kv_a_norm.weight

F32

[512]
blk.47.attn_kv_b.weight

Q8_0

[512, 32768]
blk.47.attn_norm.weight

F32

[5120]
blk.47.attn_output.weight

Q8_0

[16384, 5120]
blk.47.attn_q_a.weight

Q8_0

[5120, 1536]
blk.47.attn_q_a_norm.weight

F32

[1536]
blk.47.attn_q_b.weight

Q8_0

[1536, 24576]
blk.47.ffn_down_exps.weight

Q8_0

[1536, 5120, 160]
blk.47.ffn_down_shexp.weight

Q8_0

[3072, 5120]
blk.47.ffn_gate_exps.weight

Q8_0

[5120, 1536, 160]
blk.47.ffn_gate_inp.weight

F32

[5120, 160]
blk.47.ffn_gate_shexp.weight

Q8_0

[5120, 3072]
blk.47.ffn_norm.weight

F32

[5120]
blk.47.ffn_up_exps.weight

Q8_0

[5120, 1536, 160]
blk.47.ffn_up_shexp.weight

Q8_0

[5120, 3072]

blk.48

blk.48.attn_kv_a_mqa.weight

Q8_0

[5120, 576]
blk.48.attn_kv_a_norm.weight

F32

[512]
blk.48.attn_kv_b.weight

Q8_0

[512, 32768]
blk.48.attn_norm.weight

F32

[5120]
blk.48.attn_output.weight

Q8_0

[16384, 5120]
blk.48.attn_q_a.weight

Q8_0

[5120, 1536]
blk.48.attn_q_a_norm.weight

F32

[1536]
blk.48.attn_q_b.weight

Q8_0

[1536, 24576]
blk.48.ffn_down_exps.weight

Q8_0

[1536, 5120, 160]
blk.48.ffn_down_shexp.weight

Q8_0

[3072, 5120]
blk.48.ffn_gate_exps.weight

Q8_0

[5120, 1536, 160]
blk.48.ffn_gate_inp.weight

F32

[5120, 160]
blk.48.ffn_gate_shexp.weight

Q8_0

[5120, 3072]
blk.48.ffn_norm.weight

F32

[5120]
blk.48.ffn_up_exps.weight

Q8_0

[5120, 1536, 160]
blk.48.ffn_up_shexp.weight

Q8_0

[5120, 3072]

blk.49

blk.49.attn_kv_a_mqa.weight

Q8_0

[5120, 576]
blk.49.attn_kv_a_norm.weight

F32

[512]
blk.49.attn_kv_b.weight

Q8_0

[512, 32768]
blk.49.attn_norm.weight

F32

[5120]
blk.49.attn_output.weight

Q8_0

[16384, 5120]
blk.49.attn_q_a.weight

Q8_0

[5120, 1536]
blk.49.attn_q_a_norm.weight

F32

[1536]
blk.49.attn_q_b.weight

Q8_0

[1536, 24576]
blk.49.ffn_down_exps.weight

Q8_0

[1536, 5120, 160]
blk.49.ffn_down_shexp.weight

Q8_0

[3072, 5120]
blk.49.ffn_gate_exps.weight

Q8_0

[5120, 1536, 160]
blk.49.ffn_gate_inp.weight

F32

[5120, 160]
blk.49.ffn_gate_shexp.weight

Q8_0

[5120, 3072]
blk.49.ffn_norm.weight

F32

[5120]
blk.49.ffn_up_exps.weight

Q8_0

[5120, 1536, 160]
blk.49.ffn_up_shexp.weight

Q8_0

[5120, 3072]

blk.50

blk.50.attn_kv_a_mqa.weight

Q8_0

[5120, 576]
blk.50.attn_kv_a_norm.weight

F32

[512]
blk.50.attn_kv_b.weight

Q8_0

[512, 32768]
blk.50.attn_norm.weight

F32

[5120]
blk.50.attn_output.weight

Q8_0

[16384, 5120]
blk.50.attn_q_a.weight

Q8_0

[5120, 1536]
blk.50.attn_q_a_norm.weight

F32

[1536]
blk.50.attn_q_b.weight

Q8_0

[1536, 24576]
blk.50.ffn_down_exps.weight

Q8_0

[1536, 5120, 160]
blk.50.ffn_down_shexp.weight

Q8_0

[3072, 5120]
blk.50.ffn_gate_exps.weight

Q8_0

[5120, 1536, 160]
blk.50.ffn_gate_inp.weight

F32

[5120, 160]
blk.50.ffn_gate_shexp.weight

Q8_0

[5120, 3072]
blk.50.ffn_norm.weight

F32

[5120]
blk.50.ffn_up_exps.weight

Q8_0

[5120, 1536, 160]
blk.50.ffn_up_shexp.weight

Q8_0

[5120, 3072]

blk.51

blk.51.attn_kv_a_mqa.weight

Q8_0

[5120, 576]
blk.51.attn_kv_a_norm.weight

F32

[512]
blk.51.attn_kv_b.weight

Q8_0

[512, 32768]
blk.51.attn_norm.weight

F32

[5120]
blk.51.attn_output.weight

Q8_0

[16384, 5120]
blk.51.attn_q_a.weight

Q8_0

[5120, 1536]
blk.51.attn_q_a_norm.weight

F32

[1536]
blk.51.attn_q_b.weight

Q8_0

[1536, 24576]
blk.51.ffn_down_exps.weight

Q8_0

[1536, 5120, 160]
blk.51.ffn_down_shexp.weight

Q8_0

[3072, 5120]
blk.51.ffn_gate_exps.weight

Q8_0

[5120, 1536, 160]
blk.51.ffn_gate_inp.weight

F32

[5120, 160]
blk.51.ffn_gate_shexp.weight

Q8_0

[5120, 3072]
blk.51.ffn_norm.weight

F32

[5120]
blk.51.ffn_up_exps.weight

Q8_0

[5120, 1536, 160]
blk.51.ffn_up_shexp.weight

Q8_0

[5120, 3072]

blk.52

blk.52.attn_kv_a_mqa.weight

Q8_0

[5120, 576]
blk.52.attn_kv_a_norm.weight

F32

[512]
blk.52.attn_kv_b.weight

Q8_0

[512, 32768]
blk.52.attn_norm.weight

F32

[5120]
blk.52.attn_output.weight

Q8_0

[16384, 5120]
blk.52.attn_q_a.weight

Q8_0

[5120, 1536]
blk.52.attn_q_a_norm.weight

F32

[1536]
blk.52.attn_q_b.weight

Q8_0

[1536, 24576]
blk.52.ffn_down_exps.weight

Q8_0

[1536, 5120, 160]
blk.52.ffn_down_shexp.weight

Q8_0

[3072, 5120]
blk.52.ffn_gate_exps.weight

Q8_0

[5120, 1536, 160]
blk.52.ffn_gate_inp.weight

F32

[5120, 160]
blk.52.ffn_gate_shexp.weight

Q8_0

[5120, 3072]
blk.52.ffn_norm.weight

F32

[5120]
blk.52.ffn_up_exps.weight

Q8_0

[5120, 1536, 160]
blk.52.ffn_up_shexp.weight

Q8_0

[5120, 3072]

blk.53

blk.53.attn_kv_a_mqa.weight

Q8_0

[5120, 576]
blk.53.attn_kv_a_norm.weight

F32

[512]
blk.53.attn_kv_b.weight

Q8_0

[512, 32768]
blk.53.attn_norm.weight

F32

[5120]
blk.53.attn_output.weight

Q8_0

[16384, 5120]
blk.53.attn_q_a.weight

Q8_0

[5120, 1536]
blk.53.attn_q_a_norm.weight

F32

[1536]
blk.53.attn_q_b.weight

Q8_0

[1536, 24576]
blk.53.ffn_down_exps.weight

Q8_0

[1536, 5120, 160]
blk.53.ffn_down_shexp.weight

Q8_0

[3072, 5120]
blk.53.ffn_gate_exps.weight

Q8_0

[5120, 1536, 160]
blk.53.ffn_gate_inp.weight

F32

[5120, 160]
blk.53.ffn_gate_shexp.weight

Q8_0

[5120, 3072]
blk.53.ffn_norm.weight

F32

[5120]
blk.53.ffn_up_exps.weight

Q8_0

[5120, 1536, 160]
blk.53.ffn_up_shexp.weight

Q8_0

[5120, 3072]

blk.54

blk.54.attn_kv_a_mqa.weight

Q8_0

[5120, 576]
blk.54.attn_kv_a_norm.weight

F32

[512]
blk.54.attn_kv_b.weight

Q8_0

[512, 32768]
blk.54.attn_norm.weight

F32

[5120]
blk.54.attn_output.weight

Q8_0

[16384, 5120]
blk.54.attn_q_a.weight

Q8_0

[5120, 1536]
blk.54.attn_q_a_norm.weight

F32

[1536]
blk.54.attn_q_b.weight

Q8_0

[1536, 24576]
blk.54.ffn_down_exps.weight

Q8_0

[1536, 5120, 160]
blk.54.ffn_down_shexp.weight

Q8_0

[3072, 5120]
blk.54.ffn_gate_exps.weight

Q8_0

[5120, 1536, 160]
blk.54.ffn_gate_inp.weight

F32

[5120, 160]
blk.54.ffn_gate_shexp.weight

Q8_0

[5120, 3072]
blk.54.ffn_norm.weight

F32

[5120]
blk.54.ffn_up_exps.weight

Q8_0

[5120, 1536, 160]
blk.54.ffn_up_shexp.weight

Q8_0

[5120, 3072]

blk.55

blk.55.attn_kv_a_mqa.weight

Q8_0

[5120, 576]
blk.55.attn_kv_a_norm.weight

F32

[512]
blk.55.attn_kv_b.weight

Q8_0

[512, 32768]
blk.55.attn_norm.weight

F32

[5120]
blk.55.attn_output.weight

Q8_0

[16384, 5120]
blk.55.attn_q_a.weight

Q8_0

[5120, 1536]
blk.55.attn_q_a_norm.weight

F32

[1536]
blk.55.attn_q_b.weight

Q8_0

[1536, 24576]
blk.55.ffn_down_exps.weight

Q8_0

[1536, 5120, 160]
blk.55.ffn_down_shexp.weight

Q8_0

[3072, 5120]
blk.55.ffn_gate_exps.weight

Q8_0

[5120, 1536, 160]
blk.55.ffn_gate_inp.weight

F32

[5120, 160]
blk.55.ffn_gate_shexp.weight

Q8_0

[5120, 3072]
blk.55.ffn_norm.weight

F32

[5120]
blk.55.ffn_up_exps.weight

Q8_0

[5120, 1536, 160]
blk.55.ffn_up_shexp.weight

Q8_0

[5120, 3072]

blk.56

blk.56.attn_kv_a_mqa.weight

Q8_0

[5120, 576]
blk.56.attn_kv_a_norm.weight

F32

[512]
blk.56.attn_kv_b.weight

Q8_0

[512, 32768]
blk.56.attn_norm.weight

F32

[5120]
blk.56.attn_output.weight

Q8_0

[16384, 5120]
blk.56.attn_q_a.weight

Q8_0

[5120, 1536]
blk.56.attn_q_a_norm.weight

F32

[1536]
blk.56.attn_q_b.weight

Q8_0

[1536, 24576]
blk.56.ffn_down_exps.weight

Q8_0

[1536, 5120, 160]
blk.56.ffn_down_shexp.weight

Q8_0

[3072, 5120]
blk.56.ffn_gate_exps.weight

Q8_0

[5120, 1536, 160]
blk.56.ffn_gate_inp.weight

F32

[5120, 160]
blk.56.ffn_gate_shexp.weight

Q8_0

[5120, 3072]
blk.56.ffn_norm.weight

F32

[5120]
blk.56.ffn_up_exps.weight

Q8_0

[5120, 1536, 160]
blk.56.ffn_up_shexp.weight

Q8_0

[5120, 3072]

blk.57

blk.57.attn_kv_a_mqa.weight

Q8_0

[5120, 576]
blk.57.attn_kv_a_norm.weight

F32

[512]
blk.57.attn_kv_b.weight

Q8_0

[512, 32768]
blk.57.attn_norm.weight

F32

[5120]
blk.57.attn_output.weight

Q8_0

[16384, 5120]
blk.57.attn_q_a.weight

Q8_0

[5120, 1536]
blk.57.attn_q_a_norm.weight

F32

[1536]
blk.57.attn_q_b.weight

Q8_0

[1536, 24576]
blk.57.ffn_down_exps.weight

Q8_0

[1536, 5120, 160]
blk.57.ffn_down_shexp.weight

Q8_0

[3072, 5120]
blk.57.ffn_gate_exps.weight

Q8_0

[5120, 1536, 160]
blk.57.ffn_gate_inp.weight

F32

[5120, 160]
blk.57.ffn_gate_shexp.weight

Q8_0

[5120, 3072]
blk.57.ffn_norm.weight

F32

[5120]
blk.57.ffn_up_exps.weight

Q8_0

[5120, 1536, 160]
blk.57.ffn_up_shexp.weight

Q8_0

[5120, 3072]

blk.58

blk.58.attn_kv_a_mqa.weight

Q8_0

[5120, 576]
blk.58.attn_kv_a_norm.weight

F32

[512]
blk.58.attn_kv_b.weight

Q8_0

[512, 32768]
blk.58.attn_norm.weight

F32

[5120]
blk.58.attn_output.weight

Q8_0

[16384, 5120]
blk.58.attn_q_a.weight

Q8_0

[5120, 1536]
blk.58.attn_q_a_norm.weight

F32

[1536]
blk.58.attn_q_b.weight

Q8_0

[1536, 24576]
blk.58.ffn_down_exps.weight

Q8_0

[1536, 5120, 160]
blk.58.ffn_down_shexp.weight

Q8_0

[3072, 5120]
blk.58.ffn_gate_exps.weight

Q8_0

[5120, 1536, 160]
blk.58.ffn_gate_inp.weight

F32

[5120, 160]
blk.58.ffn_gate_shexp.weight

Q8_0

[5120, 3072]
blk.58.ffn_norm.weight

F32

[5120]
blk.58.ffn_up_exps.weight

Q8_0

[5120, 1536, 160]
blk.58.ffn_up_shexp.weight

Q8_0

[5120, 3072]

blk.59

blk.59.attn_kv_a_mqa.weight

Q8_0

[5120, 576]
blk.59.attn_kv_a_norm.weight

F32

[512]
blk.59.attn_kv_b.weight

Q8_0

[512, 32768]
blk.59.attn_norm.weight

F32

[5120]
blk.59.attn_output.weight

Q8_0

[16384, 5120]
blk.59.attn_q_a.weight

Q8_0

[5120, 1536]
blk.59.attn_q_a_norm.weight

F32

[1536]
blk.59.attn_q_b.weight

Q8_0

[1536, 24576]
blk.59.ffn_down_exps.weight

Q8_0

[1536, 5120, 160]
blk.59.ffn_down_shexp.weight

Q8_0

[3072, 5120]
blk.59.ffn_gate_exps.weight

Q8_0

[5120, 1536, 160]
blk.59.ffn_gate_inp.weight

F32

[5120, 160]
blk.59.ffn_gate_shexp.weight

Q8_0

[5120, 3072]
blk.59.ffn_norm.weight

F32

[5120]
blk.59.ffn_up_exps.weight

Q8_0

[5120, 1536, 160]
blk.59.ffn_up_shexp.weight

Q8_0

[5120, 3072]

output.weight

Q8_0

[5120, 102400]
output_norm.weight

F32

[5120]