GitHub
Discord
Docs
Cloud
Sign in
Download
Download
GitHub
Discord
Docs
Cloud
Sign in
glm-4.7-flash
:q8_0
Updated
2 hours ago
As the strongest model in the 30B class, GLM-4.7-Flash offers a new option for lightweight deployment that balances performance and efficiency.
As the strongest model in the 30B class, GLM-4.7-Flash offers a new option for lightweight deployment that balances performance and efficiency.
Cancel
tools
thinking
glm-4.7-flash:q8_0
...
/
model
ae7f944e7f47 · 32GB
Metadata
general.architecture
glm4moelite
glm4moelite
general.file_type
Q8_0
Q8_0
glm4moelite.attention.head_count
20
20
glm4moelite.attention.head_count_kv
20
20
glm4moelite.attention.key_length
256
256
glm4moelite.attention.kv_lora_rank
512
512
glm4moelite.attention.layer_norm_rms_epsilon
1e-05
1e-05
glm4moelite.attention.q_lora_rank
768
768
glm4moelite.attention.value_length
256
256
glm4moelite.block_count
47
47
glm4moelite.context_length
202752
202752
glm4moelite.embedding_length
2048
2048
glm4moelite.expert_count
64
64
glm4moelite.expert_feed_forward_length
1536
1536
glm4moelite.expert_gating_func
2
2
glm4moelite.expert_shared_count
1
1
glm4moelite.expert_used_count
4
4
glm4moelite.expert_weights_norm
true
true
glm4moelite.expert_weights_scale
1.8
1.8
glm4moelite.feed_forward_length
10240
10240
glm4moelite.leading_dense_block_count
1
1
glm4moelite.rope.dimension_count
64
64
glm4moelite.rope.freq_base
1e+06
1e+06
tokenizer.ggml.add_eos_token
false
false
tokenizer.ggml.add_padding_token
false
false
tokenizer.ggml.eos_token_id
154820
154820
tokenizer.ggml.eos_token_ids
[154820, 154827, 154829]
[154820, 154827, 154829]
tokenizer.ggml.merges
[Ġ Ġ, Ġ ĠĠĠ, ĠĠ ĠĠ, ĠĠĠ Ġ, i n, ...]
[Ġ Ġ, Ġ ĠĠĠ, ĠĠ ĠĠ, ĠĠĠ Ġ, i n, ...]
tokenizer.ggml.model
gpt2
gpt2
tokenizer.ggml.padding_token_id
154820
154820
tokenizer.ggml.pre
glm4
glm4
tokenizer.ggml.scores
[0, 1, 2, 3, 4, ...]
[0, 1, 2, 3, 4, ...]
tokenizer.ggml.token_type
[1, 1, 1, 1, 1, ...]
[1, 1, 1, 1, 1, ...]
tokenizer.ggml.tokens
[!, ", #, $, %, ...]
[!, ", #, $, %, ...]
Tensor
Name
Type
Shape
token_embd.weight
Q8_0
Q8_0
[2048, 154880]
blk.0
blk.0.attn_kv_a_mqa.weight
Q8_0
Q8_0
[2048, 576]
blk.0.attn_kv_a_norm.weight
F32
F32
[512]
blk.0.attn_kv_b.weight
Q8_0
Q8_0
[512, 8960]
blk.0.attn_norm.weight
F32
F32
[2048]
blk.0.attn_output.weight
Q8_0
Q8_0
[5120, 2048]
blk.0.attn_q_a.weight
Q8_0
Q8_0
[2048, 768]
blk.0.attn_q_a_norm.weight
F32
F32
[768]
blk.0.attn_q_b.weight
Q8_0
Q8_0
[768, 5120]
blk.0.ffn_down.weight
Q8_0
Q8_0
[10240, 2048]
blk.0.ffn_gate.weight
Q8_0
Q8_0
[2048, 10240]
blk.0.ffn_norm.weight
F32
F32
[2048]
blk.0.ffn_up.weight
Q8_0
Q8_0
[2048, 10240]
blk.1
blk.1.attn_kv_a_mqa.weight
Q8_0
Q8_0
[2048, 576]
blk.1.attn_kv_a_norm.weight
F32
F32
[512]
blk.1.attn_kv_b.weight
Q8_0
Q8_0
[512, 8960]
blk.1.attn_norm.weight
F32
F32
[2048]
blk.1.attn_output.weight
Q8_0
Q8_0
[5120, 2048]
blk.1.attn_q_a.weight
Q8_0
Q8_0
[2048, 768]
blk.1.attn_q_a_norm.weight
F32
F32
[768]
blk.1.attn_q_b.weight
Q8_0
Q8_0
[768, 5120]
blk.1.exp_probs_b.bias
F32
F32
[64]
blk.1.ffn_down_exps.weight
Q8_0
Q8_0
[1536, 2048, 64]
blk.1.ffn_down_shexp.weight
Q8_0
Q8_0
[1536, 2048]
blk.1.ffn_gate_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.1.ffn_gate_inp.weight
F32
F32
[2048, 64]
blk.1.ffn_gate_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.1.ffn_norm.weight
F32
F32
[2048]
blk.1.ffn_up_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.1.ffn_up_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.2
blk.2.attn_kv_a_mqa.weight
Q8_0
Q8_0
[2048, 576]
blk.2.attn_kv_a_norm.weight
F32
F32
[512]
blk.2.attn_kv_b.weight
Q8_0
Q8_0
[512, 8960]
blk.2.attn_norm.weight
F32
F32
[2048]
blk.2.attn_output.weight
Q8_0
Q8_0
[5120, 2048]
blk.2.attn_q_a.weight
Q8_0
Q8_0
[2048, 768]
blk.2.attn_q_a_norm.weight
F32
F32
[768]
blk.2.attn_q_b.weight
Q8_0
Q8_0
[768, 5120]
blk.2.exp_probs_b.bias
F32
F32
[64]
blk.2.ffn_down_exps.weight
Q8_0
Q8_0
[1536, 2048, 64]
blk.2.ffn_down_shexp.weight
Q8_0
Q8_0
[1536, 2048]
blk.2.ffn_gate_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.2.ffn_gate_inp.weight
F32
F32
[2048, 64]
blk.2.ffn_gate_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.2.ffn_norm.weight
F32
F32
[2048]
blk.2.ffn_up_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.2.ffn_up_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.3
blk.3.attn_kv_a_mqa.weight
Q8_0
Q8_0
[2048, 576]
blk.3.attn_kv_a_norm.weight
F32
F32
[512]
blk.3.attn_kv_b.weight
Q8_0
Q8_0
[512, 8960]
blk.3.attn_norm.weight
F32
F32
[2048]
blk.3.attn_output.weight
Q8_0
Q8_0
[5120, 2048]
blk.3.attn_q_a.weight
Q8_0
Q8_0
[2048, 768]
blk.3.attn_q_a_norm.weight
F32
F32
[768]
blk.3.attn_q_b.weight
Q8_0
Q8_0
[768, 5120]
blk.3.exp_probs_b.bias
F32
F32
[64]
blk.3.ffn_down_exps.weight
Q8_0
Q8_0
[1536, 2048, 64]
blk.3.ffn_down_shexp.weight
Q8_0
Q8_0
[1536, 2048]
blk.3.ffn_gate_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.3.ffn_gate_inp.weight
F32
F32
[2048, 64]
blk.3.ffn_gate_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.3.ffn_norm.weight
F32
F32
[2048]
blk.3.ffn_up_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.3.ffn_up_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.4
blk.4.attn_kv_a_mqa.weight
Q8_0
Q8_0
[2048, 576]
blk.4.attn_kv_a_norm.weight
F32
F32
[512]
blk.4.attn_kv_b.weight
Q8_0
Q8_0
[512, 8960]
blk.4.attn_norm.weight
F32
F32
[2048]
blk.4.attn_output.weight
Q8_0
Q8_0
[5120, 2048]
blk.4.attn_q_a.weight
Q8_0
Q8_0
[2048, 768]
blk.4.attn_q_a_norm.weight
F32
F32
[768]
blk.4.attn_q_b.weight
Q8_0
Q8_0
[768, 5120]
blk.4.exp_probs_b.bias
F32
F32
[64]
blk.4.ffn_down_exps.weight
Q8_0
Q8_0
[1536, 2048, 64]
blk.4.ffn_down_shexp.weight
Q8_0
Q8_0
[1536, 2048]
blk.4.ffn_gate_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.4.ffn_gate_inp.weight
F32
F32
[2048, 64]
blk.4.ffn_gate_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.4.ffn_norm.weight
F32
F32
[2048]
blk.4.ffn_up_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.4.ffn_up_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.5
blk.5.attn_kv_a_mqa.weight
Q8_0
Q8_0
[2048, 576]
blk.5.attn_kv_a_norm.weight
F32
F32
[512]
blk.5.attn_kv_b.weight
Q8_0
Q8_0
[512, 8960]
blk.5.attn_norm.weight
F32
F32
[2048]
blk.5.attn_output.weight
Q8_0
Q8_0
[5120, 2048]
blk.5.attn_q_a.weight
Q8_0
Q8_0
[2048, 768]
blk.5.attn_q_a_norm.weight
F32
F32
[768]
blk.5.attn_q_b.weight
Q8_0
Q8_0
[768, 5120]
blk.5.exp_probs_b.bias
F32
F32
[64]
blk.5.ffn_down_exps.weight
Q8_0
Q8_0
[1536, 2048, 64]
blk.5.ffn_down_shexp.weight
Q8_0
Q8_0
[1536, 2048]
blk.5.ffn_gate_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.5.ffn_gate_inp.weight
F32
F32
[2048, 64]
blk.5.ffn_gate_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.5.ffn_norm.weight
F32
F32
[2048]
blk.5.ffn_up_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.5.ffn_up_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.6
blk.6.attn_kv_a_mqa.weight
Q8_0
Q8_0
[2048, 576]
blk.6.attn_kv_a_norm.weight
F32
F32
[512]
blk.6.attn_kv_b.weight
Q8_0
Q8_0
[512, 8960]
blk.6.attn_norm.weight
F32
F32
[2048]
blk.6.attn_output.weight
Q8_0
Q8_0
[5120, 2048]
blk.6.attn_q_a.weight
Q8_0
Q8_0
[2048, 768]
blk.6.attn_q_a_norm.weight
F32
F32
[768]
blk.6.attn_q_b.weight
Q8_0
Q8_0
[768, 5120]
blk.6.exp_probs_b.bias
F32
F32
[64]
blk.6.ffn_down_exps.weight
Q8_0
Q8_0
[1536, 2048, 64]
blk.6.ffn_down_shexp.weight
Q8_0
Q8_0
[1536, 2048]
blk.6.ffn_gate_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.6.ffn_gate_inp.weight
F32
F32
[2048, 64]
blk.6.ffn_gate_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.6.ffn_norm.weight
F32
F32
[2048]
blk.6.ffn_up_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.6.ffn_up_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.7
blk.7.attn_kv_a_mqa.weight
Q8_0
Q8_0
[2048, 576]
blk.7.attn_kv_a_norm.weight
F32
F32
[512]
blk.7.attn_kv_b.weight
Q8_0
Q8_0
[512, 8960]
blk.7.attn_norm.weight
F32
F32
[2048]
blk.7.attn_output.weight
Q8_0
Q8_0
[5120, 2048]
blk.7.attn_q_a.weight
Q8_0
Q8_0
[2048, 768]
blk.7.attn_q_a_norm.weight
F32
F32
[768]
blk.7.attn_q_b.weight
Q8_0
Q8_0
[768, 5120]
blk.7.exp_probs_b.bias
F32
F32
[64]
blk.7.ffn_down_exps.weight
Q8_0
Q8_0
[1536, 2048, 64]
blk.7.ffn_down_shexp.weight
Q8_0
Q8_0
[1536, 2048]
blk.7.ffn_gate_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.7.ffn_gate_inp.weight
F32
F32
[2048, 64]
blk.7.ffn_gate_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.7.ffn_norm.weight
F32
F32
[2048]
blk.7.ffn_up_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.7.ffn_up_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.8
blk.8.attn_kv_a_mqa.weight
Q8_0
Q8_0
[2048, 576]
blk.8.attn_kv_a_norm.weight
F32
F32
[512]
blk.8.attn_kv_b.weight
Q8_0
Q8_0
[512, 8960]
blk.8.attn_norm.weight
F32
F32
[2048]
blk.8.attn_output.weight
Q8_0
Q8_0
[5120, 2048]
blk.8.attn_q_a.weight
Q8_0
Q8_0
[2048, 768]
blk.8.attn_q_a_norm.weight
F32
F32
[768]
blk.8.attn_q_b.weight
Q8_0
Q8_0
[768, 5120]
blk.8.exp_probs_b.bias
F32
F32
[64]
blk.8.ffn_down_exps.weight
Q8_0
Q8_0
[1536, 2048, 64]
blk.8.ffn_down_shexp.weight
Q8_0
Q8_0
[1536, 2048]
blk.8.ffn_gate_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.8.ffn_gate_inp.weight
F32
F32
[2048, 64]
blk.8.ffn_gate_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.8.ffn_norm.weight
F32
F32
[2048]
blk.8.ffn_up_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.8.ffn_up_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.9
blk.9.attn_kv_a_mqa.weight
Q8_0
Q8_0
[2048, 576]
blk.9.attn_kv_a_norm.weight
F32
F32
[512]
blk.9.attn_kv_b.weight
Q8_0
Q8_0
[512, 8960]
blk.9.attn_norm.weight
F32
F32
[2048]
blk.9.attn_output.weight
Q8_0
Q8_0
[5120, 2048]
blk.9.attn_q_a.weight
Q8_0
Q8_0
[2048, 768]
blk.9.attn_q_a_norm.weight
F32
F32
[768]
blk.9.attn_q_b.weight
Q8_0
Q8_0
[768, 5120]
blk.9.exp_probs_b.bias
F32
F32
[64]
blk.9.ffn_down_exps.weight
Q8_0
Q8_0
[1536, 2048, 64]
blk.9.ffn_down_shexp.weight
Q8_0
Q8_0
[1536, 2048]
blk.9.ffn_gate_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.9.ffn_gate_inp.weight
F32
F32
[2048, 64]
blk.9.ffn_gate_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.9.ffn_norm.weight
F32
F32
[2048]
blk.9.ffn_up_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.9.ffn_up_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.10
blk.10.attn_kv_a_mqa.weight
Q8_0
Q8_0
[2048, 576]
blk.10.attn_kv_a_norm.weight
F32
F32
[512]
blk.10.attn_kv_b.weight
Q8_0
Q8_0
[512, 8960]
blk.10.attn_norm.weight
F32
F32
[2048]
blk.10.attn_output.weight
Q8_0
Q8_0
[5120, 2048]
blk.10.attn_q_a.weight
Q8_0
Q8_0
[2048, 768]
blk.10.attn_q_a_norm.weight
F32
F32
[768]
blk.10.attn_q_b.weight
Q8_0
Q8_0
[768, 5120]
blk.10.exp_probs_b.bias
F32
F32
[64]
blk.10.ffn_down_exps.weight
Q8_0
Q8_0
[1536, 2048, 64]
blk.10.ffn_down_shexp.weight
Q8_0
Q8_0
[1536, 2048]
blk.10.ffn_gate_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.10.ffn_gate_inp.weight
F32
F32
[2048, 64]
blk.10.ffn_gate_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.10.ffn_norm.weight
F32
F32
[2048]
blk.10.ffn_up_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.10.ffn_up_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.11
blk.11.attn_kv_a_mqa.weight
Q8_0
Q8_0
[2048, 576]
blk.11.attn_kv_a_norm.weight
F32
F32
[512]
blk.11.attn_kv_b.weight
Q8_0
Q8_0
[512, 8960]
blk.11.attn_norm.weight
F32
F32
[2048]
blk.11.attn_output.weight
Q8_0
Q8_0
[5120, 2048]
blk.11.attn_q_a.weight
Q8_0
Q8_0
[2048, 768]
blk.11.attn_q_a_norm.weight
F32
F32
[768]
blk.11.attn_q_b.weight
Q8_0
Q8_0
[768, 5120]
blk.11.exp_probs_b.bias
F32
F32
[64]
blk.11.ffn_down_exps.weight
Q8_0
Q8_0
[1536, 2048, 64]
blk.11.ffn_down_shexp.weight
Q8_0
Q8_0
[1536, 2048]
blk.11.ffn_gate_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.11.ffn_gate_inp.weight
F32
F32
[2048, 64]
blk.11.ffn_gate_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.11.ffn_norm.weight
F32
F32
[2048]
blk.11.ffn_up_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.11.ffn_up_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.12
blk.12.attn_kv_a_mqa.weight
Q8_0
Q8_0
[2048, 576]
blk.12.attn_kv_a_norm.weight
F32
F32
[512]
blk.12.attn_kv_b.weight
Q8_0
Q8_0
[512, 8960]
blk.12.attn_norm.weight
F32
F32
[2048]
blk.12.attn_output.weight
Q8_0
Q8_0
[5120, 2048]
blk.12.attn_q_a.weight
Q8_0
Q8_0
[2048, 768]
blk.12.attn_q_a_norm.weight
F32
F32
[768]
blk.12.attn_q_b.weight
Q8_0
Q8_0
[768, 5120]
blk.12.exp_probs_b.bias
F32
F32
[64]
blk.12.ffn_down_exps.weight
Q8_0
Q8_0
[1536, 2048, 64]
blk.12.ffn_down_shexp.weight
Q8_0
Q8_0
[1536, 2048]
blk.12.ffn_gate_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.12.ffn_gate_inp.weight
F32
F32
[2048, 64]
blk.12.ffn_gate_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.12.ffn_norm.weight
F32
F32
[2048]
blk.12.ffn_up_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.12.ffn_up_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.13
blk.13.attn_kv_a_mqa.weight
Q8_0
Q8_0
[2048, 576]
blk.13.attn_kv_a_norm.weight
F32
F32
[512]
blk.13.attn_kv_b.weight
Q8_0
Q8_0
[512, 8960]
blk.13.attn_norm.weight
F32
F32
[2048]
blk.13.attn_output.weight
Q8_0
Q8_0
[5120, 2048]
blk.13.attn_q_a.weight
Q8_0
Q8_0
[2048, 768]
blk.13.attn_q_a_norm.weight
F32
F32
[768]
blk.13.attn_q_b.weight
Q8_0
Q8_0
[768, 5120]
blk.13.exp_probs_b.bias
F32
F32
[64]
blk.13.ffn_down_exps.weight
Q8_0
Q8_0
[1536, 2048, 64]
blk.13.ffn_down_shexp.weight
Q8_0
Q8_0
[1536, 2048]
blk.13.ffn_gate_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.13.ffn_gate_inp.weight
F32
F32
[2048, 64]
blk.13.ffn_gate_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.13.ffn_norm.weight
F32
F32
[2048]
blk.13.ffn_up_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.13.ffn_up_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.14
blk.14.attn_kv_a_mqa.weight
Q8_0
Q8_0
[2048, 576]
blk.14.attn_kv_a_norm.weight
F32
F32
[512]
blk.14.attn_kv_b.weight
Q8_0
Q8_0
[512, 8960]
blk.14.attn_norm.weight
F32
F32
[2048]
blk.14.attn_output.weight
Q8_0
Q8_0
[5120, 2048]
blk.14.attn_q_a.weight
Q8_0
Q8_0
[2048, 768]
blk.14.attn_q_a_norm.weight
F32
F32
[768]
blk.14.attn_q_b.weight
Q8_0
Q8_0
[768, 5120]
blk.14.exp_probs_b.bias
F32
F32
[64]
blk.14.ffn_down_exps.weight
Q8_0
Q8_0
[1536, 2048, 64]
blk.14.ffn_down_shexp.weight
Q8_0
Q8_0
[1536, 2048]
blk.14.ffn_gate_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.14.ffn_gate_inp.weight
F32
F32
[2048, 64]
blk.14.ffn_gate_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.14.ffn_norm.weight
F32
F32
[2048]
blk.14.ffn_up_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.14.ffn_up_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.15
blk.15.attn_kv_a_mqa.weight
Q8_0
Q8_0
[2048, 576]
blk.15.attn_kv_a_norm.weight
F32
F32
[512]
blk.15.attn_kv_b.weight
Q8_0
Q8_0
[512, 8960]
blk.15.attn_norm.weight
F32
F32
[2048]
blk.15.attn_output.weight
Q8_0
Q8_0
[5120, 2048]
blk.15.attn_q_a.weight
Q8_0
Q8_0
[2048, 768]
blk.15.attn_q_a_norm.weight
F32
F32
[768]
blk.15.attn_q_b.weight
Q8_0
Q8_0
[768, 5120]
blk.15.exp_probs_b.bias
F32
F32
[64]
blk.15.ffn_down_exps.weight
Q8_0
Q8_0
[1536, 2048, 64]
blk.15.ffn_down_shexp.weight
Q8_0
Q8_0
[1536, 2048]
blk.15.ffn_gate_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.15.ffn_gate_inp.weight
F32
F32
[2048, 64]
blk.15.ffn_gate_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.15.ffn_norm.weight
F32
F32
[2048]
blk.15.ffn_up_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.15.ffn_up_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.16
blk.16.attn_kv_a_mqa.weight
Q8_0
Q8_0
[2048, 576]
blk.16.attn_kv_a_norm.weight
F32
F32
[512]
blk.16.attn_kv_b.weight
Q8_0
Q8_0
[512, 8960]
blk.16.attn_norm.weight
F32
F32
[2048]
blk.16.attn_output.weight
Q8_0
Q8_0
[5120, 2048]
blk.16.attn_q_a.weight
Q8_0
Q8_0
[2048, 768]
blk.16.attn_q_a_norm.weight
F32
F32
[768]
blk.16.attn_q_b.weight
Q8_0
Q8_0
[768, 5120]
blk.16.exp_probs_b.bias
F32
F32
[64]
blk.16.ffn_down_exps.weight
Q8_0
Q8_0
[1536, 2048, 64]
blk.16.ffn_down_shexp.weight
Q8_0
Q8_0
[1536, 2048]
blk.16.ffn_gate_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.16.ffn_gate_inp.weight
F32
F32
[2048, 64]
blk.16.ffn_gate_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.16.ffn_norm.weight
F32
F32
[2048]
blk.16.ffn_up_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.16.ffn_up_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.17
blk.17.attn_kv_a_mqa.weight
Q8_0
Q8_0
[2048, 576]
blk.17.attn_kv_a_norm.weight
F32
F32
[512]
blk.17.attn_kv_b.weight
Q8_0
Q8_0
[512, 8960]
blk.17.attn_norm.weight
F32
F32
[2048]
blk.17.attn_output.weight
Q8_0
Q8_0
[5120, 2048]
blk.17.attn_q_a.weight
Q8_0
Q8_0
[2048, 768]
blk.17.attn_q_a_norm.weight
F32
F32
[768]
blk.17.attn_q_b.weight
Q8_0
Q8_0
[768, 5120]
blk.17.exp_probs_b.bias
F32
F32
[64]
blk.17.ffn_down_exps.weight
Q8_0
Q8_0
[1536, 2048, 64]
blk.17.ffn_down_shexp.weight
Q8_0
Q8_0
[1536, 2048]
blk.17.ffn_gate_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.17.ffn_gate_inp.weight
F32
F32
[2048, 64]
blk.17.ffn_gate_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.17.ffn_norm.weight
F32
F32
[2048]
blk.17.ffn_up_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.17.ffn_up_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.18
blk.18.attn_kv_a_mqa.weight
Q8_0
Q8_0
[2048, 576]
blk.18.attn_kv_a_norm.weight
F32
F32
[512]
blk.18.attn_kv_b.weight
Q8_0
Q8_0
[512, 8960]
blk.18.attn_norm.weight
F32
F32
[2048]
blk.18.attn_output.weight
Q8_0
Q8_0
[5120, 2048]
blk.18.attn_q_a.weight
Q8_0
Q8_0
[2048, 768]
blk.18.attn_q_a_norm.weight
F32
F32
[768]
blk.18.attn_q_b.weight
Q8_0
Q8_0
[768, 5120]
blk.18.exp_probs_b.bias
F32
F32
[64]
blk.18.ffn_down_exps.weight
Q8_0
Q8_0
[1536, 2048, 64]
blk.18.ffn_down_shexp.weight
Q8_0
Q8_0
[1536, 2048]
blk.18.ffn_gate_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.18.ffn_gate_inp.weight
F32
F32
[2048, 64]
blk.18.ffn_gate_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.18.ffn_norm.weight
F32
F32
[2048]
blk.18.ffn_up_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.18.ffn_up_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.19
blk.19.attn_kv_a_mqa.weight
Q8_0
Q8_0
[2048, 576]
blk.19.attn_kv_a_norm.weight
F32
F32
[512]
blk.19.attn_kv_b.weight
Q8_0
Q8_0
[512, 8960]
blk.19.attn_norm.weight
F32
F32
[2048]
blk.19.attn_output.weight
Q8_0
Q8_0
[5120, 2048]
blk.19.attn_q_a.weight
Q8_0
Q8_0
[2048, 768]
blk.19.attn_q_a_norm.weight
F32
F32
[768]
blk.19.attn_q_b.weight
Q8_0
Q8_0
[768, 5120]
blk.19.exp_probs_b.bias
F32
F32
[64]
blk.19.ffn_down_exps.weight
Q8_0
Q8_0
[1536, 2048, 64]
blk.19.ffn_down_shexp.weight
Q8_0
Q8_0
[1536, 2048]
blk.19.ffn_gate_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.19.ffn_gate_inp.weight
F32
F32
[2048, 64]
blk.19.ffn_gate_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.19.ffn_norm.weight
F32
F32
[2048]
blk.19.ffn_up_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.19.ffn_up_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.20
blk.20.attn_kv_a_mqa.weight
Q8_0
Q8_0
[2048, 576]
blk.20.attn_kv_a_norm.weight
F32
F32
[512]
blk.20.attn_kv_b.weight
Q8_0
Q8_0
[512, 8960]
blk.20.attn_norm.weight
F32
F32
[2048]
blk.20.attn_output.weight
Q8_0
Q8_0
[5120, 2048]
blk.20.attn_q_a.weight
Q8_0
Q8_0
[2048, 768]
blk.20.attn_q_a_norm.weight
F32
F32
[768]
blk.20.attn_q_b.weight
Q8_0
Q8_0
[768, 5120]
blk.20.exp_probs_b.bias
F32
F32
[64]
blk.20.ffn_down_exps.weight
Q8_0
Q8_0
[1536, 2048, 64]
blk.20.ffn_down_shexp.weight
Q8_0
Q8_0
[1536, 2048]
blk.20.ffn_gate_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.20.ffn_gate_inp.weight
F32
F32
[2048, 64]
blk.20.ffn_gate_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.20.ffn_norm.weight
F32
F32
[2048]
blk.20.ffn_up_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.20.ffn_up_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.21
blk.21.attn_kv_a_mqa.weight
Q8_0
Q8_0
[2048, 576]
blk.21.attn_kv_a_norm.weight
F32
F32
[512]
blk.21.attn_kv_b.weight
Q8_0
Q8_0
[512, 8960]
blk.21.attn_norm.weight
F32
F32
[2048]
blk.21.attn_output.weight
Q8_0
Q8_0
[5120, 2048]
blk.21.attn_q_a.weight
Q8_0
Q8_0
[2048, 768]
blk.21.attn_q_a_norm.weight
F32
F32
[768]
blk.21.attn_q_b.weight
Q8_0
Q8_0
[768, 5120]
blk.21.exp_probs_b.bias
F32
F32
[64]
blk.21.ffn_down_exps.weight
Q8_0
Q8_0
[1536, 2048, 64]
blk.21.ffn_down_shexp.weight
Q8_0
Q8_0
[1536, 2048]
blk.21.ffn_gate_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.21.ffn_gate_inp.weight
F32
F32
[2048, 64]
blk.21.ffn_gate_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.21.ffn_norm.weight
F32
F32
[2048]
blk.21.ffn_up_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.21.ffn_up_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.22
blk.22.attn_kv_a_mqa.weight
Q8_0
Q8_0
[2048, 576]
blk.22.attn_kv_a_norm.weight
F32
F32
[512]
blk.22.attn_kv_b.weight
Q8_0
Q8_0
[512, 8960]
blk.22.attn_norm.weight
F32
F32
[2048]
blk.22.attn_output.weight
Q8_0
Q8_0
[5120, 2048]
blk.22.attn_q_a.weight
Q8_0
Q8_0
[2048, 768]
blk.22.attn_q_a_norm.weight
F32
F32
[768]
blk.22.attn_q_b.weight
Q8_0
Q8_0
[768, 5120]
blk.22.exp_probs_b.bias
F32
F32
[64]
blk.22.ffn_down_exps.weight
Q8_0
Q8_0
[1536, 2048, 64]
blk.22.ffn_down_shexp.weight
Q8_0
Q8_0
[1536, 2048]
blk.22.ffn_gate_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.22.ffn_gate_inp.weight
F32
F32
[2048, 64]
blk.22.ffn_gate_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.22.ffn_norm.weight
F32
F32
[2048]
blk.22.ffn_up_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.22.ffn_up_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.23
blk.23.attn_kv_a_mqa.weight
Q8_0
Q8_0
[2048, 576]
blk.23.attn_kv_a_norm.weight
F32
F32
[512]
blk.23.attn_kv_b.weight
Q8_0
Q8_0
[512, 8960]
blk.23.attn_norm.weight
F32
F32
[2048]
blk.23.attn_output.weight
Q8_0
Q8_0
[5120, 2048]
blk.23.attn_q_a.weight
Q8_0
Q8_0
[2048, 768]
blk.23.attn_q_a_norm.weight
F32
F32
[768]
blk.23.attn_q_b.weight
Q8_0
Q8_0
[768, 5120]
blk.23.exp_probs_b.bias
F32
F32
[64]
blk.23.ffn_down_exps.weight
Q8_0
Q8_0
[1536, 2048, 64]
blk.23.ffn_down_shexp.weight
Q8_0
Q8_0
[1536, 2048]
blk.23.ffn_gate_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.23.ffn_gate_inp.weight
F32
F32
[2048, 64]
blk.23.ffn_gate_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.23.ffn_norm.weight
F32
F32
[2048]
blk.23.ffn_up_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.23.ffn_up_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.24
blk.24.attn_kv_a_mqa.weight
Q8_0
Q8_0
[2048, 576]
blk.24.attn_kv_a_norm.weight
F32
F32
[512]
blk.24.attn_kv_b.weight
Q8_0
Q8_0
[512, 8960]
blk.24.attn_norm.weight
F32
F32
[2048]
blk.24.attn_output.weight
Q8_0
Q8_0
[5120, 2048]
blk.24.attn_q_a.weight
Q8_0
Q8_0
[2048, 768]
blk.24.attn_q_a_norm.weight
F32
F32
[768]
blk.24.attn_q_b.weight
Q8_0
Q8_0
[768, 5120]
blk.24.exp_probs_b.bias
F32
F32
[64]
blk.24.ffn_down_exps.weight
Q8_0
Q8_0
[1536, 2048, 64]
blk.24.ffn_down_shexp.weight
Q8_0
Q8_0
[1536, 2048]
blk.24.ffn_gate_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.24.ffn_gate_inp.weight
F32
F32
[2048, 64]
blk.24.ffn_gate_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.24.ffn_norm.weight
F32
F32
[2048]
blk.24.ffn_up_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.24.ffn_up_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.25
blk.25.attn_kv_a_mqa.weight
Q8_0
Q8_0
[2048, 576]
blk.25.attn_kv_a_norm.weight
F32
F32
[512]
blk.25.attn_kv_b.weight
Q8_0
Q8_0
[512, 8960]
blk.25.attn_norm.weight
F32
F32
[2048]
blk.25.attn_output.weight
Q8_0
Q8_0
[5120, 2048]
blk.25.attn_q_a.weight
Q8_0
Q8_0
[2048, 768]
blk.25.attn_q_a_norm.weight
F32
F32
[768]
blk.25.attn_q_b.weight
Q8_0
Q8_0
[768, 5120]
blk.25.exp_probs_b.bias
F32
F32
[64]
blk.25.ffn_down_exps.weight
Q8_0
Q8_0
[1536, 2048, 64]
blk.25.ffn_down_shexp.weight
Q8_0
Q8_0
[1536, 2048]
blk.25.ffn_gate_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.25.ffn_gate_inp.weight
F32
F32
[2048, 64]
blk.25.ffn_gate_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.25.ffn_norm.weight
F32
F32
[2048]
blk.25.ffn_up_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.25.ffn_up_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.26
blk.26.attn_kv_a_mqa.weight
Q8_0
Q8_0
[2048, 576]
blk.26.attn_kv_a_norm.weight
F32
F32
[512]
blk.26.attn_kv_b.weight
Q8_0
Q8_0
[512, 8960]
blk.26.attn_norm.weight
F32
F32
[2048]
blk.26.attn_output.weight
Q8_0
Q8_0
[5120, 2048]
blk.26.attn_q_a.weight
Q8_0
Q8_0
[2048, 768]
blk.26.attn_q_a_norm.weight
F32
F32
[768]
blk.26.attn_q_b.weight
Q8_0
Q8_0
[768, 5120]
blk.26.exp_probs_b.bias
F32
F32
[64]
blk.26.ffn_down_exps.weight
Q8_0
Q8_0
[1536, 2048, 64]
blk.26.ffn_down_shexp.weight
Q8_0
Q8_0
[1536, 2048]
blk.26.ffn_gate_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.26.ffn_gate_inp.weight
F32
F32
[2048, 64]
blk.26.ffn_gate_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.26.ffn_norm.weight
F32
F32
[2048]
blk.26.ffn_up_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.26.ffn_up_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.27
blk.27.attn_kv_a_mqa.weight
Q8_0
Q8_0
[2048, 576]
blk.27.attn_kv_a_norm.weight
F32
F32
[512]
blk.27.attn_kv_b.weight
Q8_0
Q8_0
[512, 8960]
blk.27.attn_norm.weight
F32
F32
[2048]
blk.27.attn_output.weight
Q8_0
Q8_0
[5120, 2048]
blk.27.attn_q_a.weight
Q8_0
Q8_0
[2048, 768]
blk.27.attn_q_a_norm.weight
F32
F32
[768]
blk.27.attn_q_b.weight
Q8_0
Q8_0
[768, 5120]
blk.27.exp_probs_b.bias
F32
F32
[64]
blk.27.ffn_down_exps.weight
Q8_0
Q8_0
[1536, 2048, 64]
blk.27.ffn_down_shexp.weight
Q8_0
Q8_0
[1536, 2048]
blk.27.ffn_gate_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.27.ffn_gate_inp.weight
F32
F32
[2048, 64]
blk.27.ffn_gate_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.27.ffn_norm.weight
F32
F32
[2048]
blk.27.ffn_up_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.27.ffn_up_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.28
blk.28.attn_kv_a_mqa.weight
Q8_0
Q8_0
[2048, 576]
blk.28.attn_kv_a_norm.weight
F32
F32
[512]
blk.28.attn_kv_b.weight
Q8_0
Q8_0
[512, 8960]
blk.28.attn_norm.weight
F32
F32
[2048]
blk.28.attn_output.weight
Q8_0
Q8_0
[5120, 2048]
blk.28.attn_q_a.weight
Q8_0
Q8_0
[2048, 768]
blk.28.attn_q_a_norm.weight
F32
F32
[768]
blk.28.attn_q_b.weight
Q8_0
Q8_0
[768, 5120]
blk.28.exp_probs_b.bias
F32
F32
[64]
blk.28.ffn_down_exps.weight
Q8_0
Q8_0
[1536, 2048, 64]
blk.28.ffn_down_shexp.weight
Q8_0
Q8_0
[1536, 2048]
blk.28.ffn_gate_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.28.ffn_gate_inp.weight
F32
F32
[2048, 64]
blk.28.ffn_gate_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.28.ffn_norm.weight
F32
F32
[2048]
blk.28.ffn_up_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.28.ffn_up_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.29
blk.29.attn_kv_a_mqa.weight
Q8_0
Q8_0
[2048, 576]
blk.29.attn_kv_a_norm.weight
F32
F32
[512]
blk.29.attn_kv_b.weight
Q8_0
Q8_0
[512, 8960]
blk.29.attn_norm.weight
F32
F32
[2048]
blk.29.attn_output.weight
Q8_0
Q8_0
[5120, 2048]
blk.29.attn_q_a.weight
Q8_0
Q8_0
[2048, 768]
blk.29.attn_q_a_norm.weight
F32
F32
[768]
blk.29.attn_q_b.weight
Q8_0
Q8_0
[768, 5120]
blk.29.exp_probs_b.bias
F32
F32
[64]
blk.29.ffn_down_exps.weight
Q8_0
Q8_0
[1536, 2048, 64]
blk.29.ffn_down_shexp.weight
Q8_0
Q8_0
[1536, 2048]
blk.29.ffn_gate_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.29.ffn_gate_inp.weight
F32
F32
[2048, 64]
blk.29.ffn_gate_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.29.ffn_norm.weight
F32
F32
[2048]
blk.29.ffn_up_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.29.ffn_up_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.30
blk.30.attn_kv_a_mqa.weight
Q8_0
Q8_0
[2048, 576]
blk.30.attn_kv_a_norm.weight
F32
F32
[512]
blk.30.attn_kv_b.weight
Q8_0
Q8_0
[512, 8960]
blk.30.attn_norm.weight
F32
F32
[2048]
blk.30.attn_output.weight
Q8_0
Q8_0
[5120, 2048]
blk.30.attn_q_a.weight
Q8_0
Q8_0
[2048, 768]
blk.30.attn_q_a_norm.weight
F32
F32
[768]
blk.30.attn_q_b.weight
Q8_0
Q8_0
[768, 5120]
blk.30.exp_probs_b.bias
F32
F32
[64]
blk.30.ffn_down_exps.weight
Q8_0
Q8_0
[1536, 2048, 64]
blk.30.ffn_down_shexp.weight
Q8_0
Q8_0
[1536, 2048]
blk.30.ffn_gate_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.30.ffn_gate_inp.weight
F32
F32
[2048, 64]
blk.30.ffn_gate_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.30.ffn_norm.weight
F32
F32
[2048]
blk.30.ffn_up_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.30.ffn_up_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.31
blk.31.attn_kv_a_mqa.weight
Q8_0
Q8_0
[2048, 576]
blk.31.attn_kv_a_norm.weight
F32
F32
[512]
blk.31.attn_kv_b.weight
Q8_0
Q8_0
[512, 8960]
blk.31.attn_norm.weight
F32
F32
[2048]
blk.31.attn_output.weight
Q8_0
Q8_0
[5120, 2048]
blk.31.attn_q_a.weight
Q8_0
Q8_0
[2048, 768]
blk.31.attn_q_a_norm.weight
F32
F32
[768]
blk.31.attn_q_b.weight
Q8_0
Q8_0
[768, 5120]
blk.31.exp_probs_b.bias
F32
F32
[64]
blk.31.ffn_down_exps.weight
Q8_0
Q8_0
[1536, 2048, 64]
blk.31.ffn_down_shexp.weight
Q8_0
Q8_0
[1536, 2048]
blk.31.ffn_gate_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.31.ffn_gate_inp.weight
F32
F32
[2048, 64]
blk.31.ffn_gate_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.31.ffn_norm.weight
F32
F32
[2048]
blk.31.ffn_up_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.31.ffn_up_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.32
blk.32.attn_kv_a_mqa.weight
Q8_0
Q8_0
[2048, 576]
blk.32.attn_kv_a_norm.weight
F32
F32
[512]
blk.32.attn_kv_b.weight
Q8_0
Q8_0
[512, 8960]
blk.32.attn_norm.weight
F32
F32
[2048]
blk.32.attn_output.weight
Q8_0
Q8_0
[5120, 2048]
blk.32.attn_q_a.weight
Q8_0
Q8_0
[2048, 768]
blk.32.attn_q_a_norm.weight
F32
F32
[768]
blk.32.attn_q_b.weight
Q8_0
Q8_0
[768, 5120]
blk.32.exp_probs_b.bias
F32
F32
[64]
blk.32.ffn_down_exps.weight
Q8_0
Q8_0
[1536, 2048, 64]
blk.32.ffn_down_shexp.weight
Q8_0
Q8_0
[1536, 2048]
blk.32.ffn_gate_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.32.ffn_gate_inp.weight
F32
F32
[2048, 64]
blk.32.ffn_gate_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.32.ffn_norm.weight
F32
F32
[2048]
blk.32.ffn_up_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.32.ffn_up_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.33
blk.33.attn_kv_a_mqa.weight
Q8_0
Q8_0
[2048, 576]
blk.33.attn_kv_a_norm.weight
F32
F32
[512]
blk.33.attn_kv_b.weight
Q8_0
Q8_0
[512, 8960]
blk.33.attn_norm.weight
F32
F32
[2048]
blk.33.attn_output.weight
Q8_0
Q8_0
[5120, 2048]
blk.33.attn_q_a.weight
Q8_0
Q8_0
[2048, 768]
blk.33.attn_q_a_norm.weight
F32
F32
[768]
blk.33.attn_q_b.weight
Q8_0
Q8_0
[768, 5120]
blk.33.exp_probs_b.bias
F32
F32
[64]
blk.33.ffn_down_exps.weight
Q8_0
Q8_0
[1536, 2048, 64]
blk.33.ffn_down_shexp.weight
Q8_0
Q8_0
[1536, 2048]
blk.33.ffn_gate_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.33.ffn_gate_inp.weight
F32
F32
[2048, 64]
blk.33.ffn_gate_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.33.ffn_norm.weight
F32
F32
[2048]
blk.33.ffn_up_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.33.ffn_up_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.34
blk.34.attn_kv_a_mqa.weight
Q8_0
Q8_0
[2048, 576]
blk.34.attn_kv_a_norm.weight
F32
F32
[512]
blk.34.attn_kv_b.weight
Q8_0
Q8_0
[512, 8960]
blk.34.attn_norm.weight
F32
F32
[2048]
blk.34.attn_output.weight
Q8_0
Q8_0
[5120, 2048]
blk.34.attn_q_a.weight
Q8_0
Q8_0
[2048, 768]
blk.34.attn_q_a_norm.weight
F32
F32
[768]
blk.34.attn_q_b.weight
Q8_0
Q8_0
[768, 5120]
blk.34.exp_probs_b.bias
F32
F32
[64]
blk.34.ffn_down_exps.weight
Q8_0
Q8_0
[1536, 2048, 64]
blk.34.ffn_down_shexp.weight
Q8_0
Q8_0
[1536, 2048]
blk.34.ffn_gate_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.34.ffn_gate_inp.weight
F32
F32
[2048, 64]
blk.34.ffn_gate_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.34.ffn_norm.weight
F32
F32
[2048]
blk.34.ffn_up_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.34.ffn_up_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.35
blk.35.attn_kv_a_mqa.weight
Q8_0
Q8_0
[2048, 576]
blk.35.attn_kv_a_norm.weight
F32
F32
[512]
blk.35.attn_kv_b.weight
Q8_0
Q8_0
[512, 8960]
blk.35.attn_norm.weight
F32
F32
[2048]
blk.35.attn_output.weight
Q8_0
Q8_0
[5120, 2048]
blk.35.attn_q_a.weight
Q8_0
Q8_0
[2048, 768]
blk.35.attn_q_a_norm.weight
F32
F32
[768]
blk.35.attn_q_b.weight
Q8_0
Q8_0
[768, 5120]
blk.35.exp_probs_b.bias
F32
F32
[64]
blk.35.ffn_down_exps.weight
Q8_0
Q8_0
[1536, 2048, 64]
blk.35.ffn_down_shexp.weight
Q8_0
Q8_0
[1536, 2048]
blk.35.ffn_gate_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.35.ffn_gate_inp.weight
F32
F32
[2048, 64]
blk.35.ffn_gate_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.35.ffn_norm.weight
F32
F32
[2048]
blk.35.ffn_up_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.35.ffn_up_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.36
blk.36.attn_kv_a_mqa.weight
Q8_0
Q8_0
[2048, 576]
blk.36.attn_kv_a_norm.weight
F32
F32
[512]
blk.36.attn_kv_b.weight
Q8_0
Q8_0
[512, 8960]
blk.36.attn_norm.weight
F32
F32
[2048]
blk.36.attn_output.weight
Q8_0
Q8_0
[5120, 2048]
blk.36.attn_q_a.weight
Q8_0
Q8_0
[2048, 768]
blk.36.attn_q_a_norm.weight
F32
F32
[768]
blk.36.attn_q_b.weight
Q8_0
Q8_0
[768, 5120]
blk.36.exp_probs_b.bias
F32
F32
[64]
blk.36.ffn_down_exps.weight
Q8_0
Q8_0
[1536, 2048, 64]
blk.36.ffn_down_shexp.weight
Q8_0
Q8_0
[1536, 2048]
blk.36.ffn_gate_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.36.ffn_gate_inp.weight
F32
F32
[2048, 64]
blk.36.ffn_gate_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.36.ffn_norm.weight
F32
F32
[2048]
blk.36.ffn_up_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.36.ffn_up_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.37
blk.37.attn_kv_a_mqa.weight
Q8_0
Q8_0
[2048, 576]
blk.37.attn_kv_a_norm.weight
F32
F32
[512]
blk.37.attn_kv_b.weight
Q8_0
Q8_0
[512, 8960]
blk.37.attn_norm.weight
F32
F32
[2048]
blk.37.attn_output.weight
Q8_0
Q8_0
[5120, 2048]
blk.37.attn_q_a.weight
Q8_0
Q8_0
[2048, 768]
blk.37.attn_q_a_norm.weight
F32
F32
[768]
blk.37.attn_q_b.weight
Q8_0
Q8_0
[768, 5120]
blk.37.exp_probs_b.bias
F32
F32
[64]
blk.37.ffn_down_exps.weight
Q8_0
Q8_0
[1536, 2048, 64]
blk.37.ffn_down_shexp.weight
Q8_0
Q8_0
[1536, 2048]
blk.37.ffn_gate_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.37.ffn_gate_inp.weight
F32
F32
[2048, 64]
blk.37.ffn_gate_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.37.ffn_norm.weight
F32
F32
[2048]
blk.37.ffn_up_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.37.ffn_up_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.38
blk.38.attn_kv_a_mqa.weight
Q8_0
Q8_0
[2048, 576]
blk.38.attn_kv_a_norm.weight
F32
F32
[512]
blk.38.attn_kv_b.weight
Q8_0
Q8_0
[512, 8960]
blk.38.attn_norm.weight
F32
F32
[2048]
blk.38.attn_output.weight
Q8_0
Q8_0
[5120, 2048]
blk.38.attn_q_a.weight
Q8_0
Q8_0
[2048, 768]
blk.38.attn_q_a_norm.weight
F32
F32
[768]
blk.38.attn_q_b.weight
Q8_0
Q8_0
[768, 5120]
blk.38.exp_probs_b.bias
F32
F32
[64]
blk.38.ffn_down_exps.weight
Q8_0
Q8_0
[1536, 2048, 64]
blk.38.ffn_down_shexp.weight
Q8_0
Q8_0
[1536, 2048]
blk.38.ffn_gate_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.38.ffn_gate_inp.weight
F32
F32
[2048, 64]
blk.38.ffn_gate_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.38.ffn_norm.weight
F32
F32
[2048]
blk.38.ffn_up_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.38.ffn_up_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.39
blk.39.attn_kv_a_mqa.weight
Q8_0
Q8_0
[2048, 576]
blk.39.attn_kv_a_norm.weight
F32
F32
[512]
blk.39.attn_kv_b.weight
Q8_0
Q8_0
[512, 8960]
blk.39.attn_norm.weight
F32
F32
[2048]
blk.39.attn_output.weight
Q8_0
Q8_0
[5120, 2048]
blk.39.attn_q_a.weight
Q8_0
Q8_0
[2048, 768]
blk.39.attn_q_a_norm.weight
F32
F32
[768]
blk.39.attn_q_b.weight
Q8_0
Q8_0
[768, 5120]
blk.39.exp_probs_b.bias
F32
F32
[64]
blk.39.ffn_down_exps.weight
Q8_0
Q8_0
[1536, 2048, 64]
blk.39.ffn_down_shexp.weight
Q8_0
Q8_0
[1536, 2048]
blk.39.ffn_gate_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.39.ffn_gate_inp.weight
F32
F32
[2048, 64]
blk.39.ffn_gate_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.39.ffn_norm.weight
F32
F32
[2048]
blk.39.ffn_up_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.39.ffn_up_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.40
blk.40.attn_kv_a_mqa.weight
Q8_0
Q8_0
[2048, 576]
blk.40.attn_kv_a_norm.weight
F32
F32
[512]
blk.40.attn_kv_b.weight
Q8_0
Q8_0
[512, 8960]
blk.40.attn_norm.weight
F32
F32
[2048]
blk.40.attn_output.weight
Q8_0
Q8_0
[5120, 2048]
blk.40.attn_q_a.weight
Q8_0
Q8_0
[2048, 768]
blk.40.attn_q_a_norm.weight
F32
F32
[768]
blk.40.attn_q_b.weight
Q8_0
Q8_0
[768, 5120]
blk.40.exp_probs_b.bias
F32
F32
[64]
blk.40.ffn_down_exps.weight
Q8_0
Q8_0
[1536, 2048, 64]
blk.40.ffn_down_shexp.weight
Q8_0
Q8_0
[1536, 2048]
blk.40.ffn_gate_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.40.ffn_gate_inp.weight
F32
F32
[2048, 64]
blk.40.ffn_gate_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.40.ffn_norm.weight
F32
F32
[2048]
blk.40.ffn_up_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.40.ffn_up_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.41
blk.41.attn_kv_a_mqa.weight
Q8_0
Q8_0
[2048, 576]
blk.41.attn_kv_a_norm.weight
F32
F32
[512]
blk.41.attn_kv_b.weight
Q8_0
Q8_0
[512, 8960]
blk.41.attn_norm.weight
F32
F32
[2048]
blk.41.attn_output.weight
Q8_0
Q8_0
[5120, 2048]
blk.41.attn_q_a.weight
Q8_0
Q8_0
[2048, 768]
blk.41.attn_q_a_norm.weight
F32
F32
[768]
blk.41.attn_q_b.weight
Q8_0
Q8_0
[768, 5120]
blk.41.exp_probs_b.bias
F32
F32
[64]
blk.41.ffn_down_exps.weight
Q8_0
Q8_0
[1536, 2048, 64]
blk.41.ffn_down_shexp.weight
Q8_0
Q8_0
[1536, 2048]
blk.41.ffn_gate_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.41.ffn_gate_inp.weight
F32
F32
[2048, 64]
blk.41.ffn_gate_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.41.ffn_norm.weight
F32
F32
[2048]
blk.41.ffn_up_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.41.ffn_up_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.42
blk.42.attn_kv_a_mqa.weight
Q8_0
Q8_0
[2048, 576]
blk.42.attn_kv_a_norm.weight
F32
F32
[512]
blk.42.attn_kv_b.weight
Q8_0
Q8_0
[512, 8960]
blk.42.attn_norm.weight
F32
F32
[2048]
blk.42.attn_output.weight
Q8_0
Q8_0
[5120, 2048]
blk.42.attn_q_a.weight
Q8_0
Q8_0
[2048, 768]
blk.42.attn_q_a_norm.weight
F32
F32
[768]
blk.42.attn_q_b.weight
Q8_0
Q8_0
[768, 5120]
blk.42.exp_probs_b.bias
F32
F32
[64]
blk.42.ffn_down_exps.weight
Q8_0
Q8_0
[1536, 2048, 64]
blk.42.ffn_down_shexp.weight
Q8_0
Q8_0
[1536, 2048]
blk.42.ffn_gate_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.42.ffn_gate_inp.weight
F32
F32
[2048, 64]
blk.42.ffn_gate_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.42.ffn_norm.weight
F32
F32
[2048]
blk.42.ffn_up_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.42.ffn_up_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.43
blk.43.attn_kv_a_mqa.weight
Q8_0
Q8_0
[2048, 576]
blk.43.attn_kv_a_norm.weight
F32
F32
[512]
blk.43.attn_kv_b.weight
Q8_0
Q8_0
[512, 8960]
blk.43.attn_norm.weight
F32
F32
[2048]
blk.43.attn_output.weight
Q8_0
Q8_0
[5120, 2048]
blk.43.attn_q_a.weight
Q8_0
Q8_0
[2048, 768]
blk.43.attn_q_a_norm.weight
F32
F32
[768]
blk.43.attn_q_b.weight
Q8_0
Q8_0
[768, 5120]
blk.43.exp_probs_b.bias
F32
F32
[64]
blk.43.ffn_down_exps.weight
Q8_0
Q8_0
[1536, 2048, 64]
blk.43.ffn_down_shexp.weight
Q8_0
Q8_0
[1536, 2048]
blk.43.ffn_gate_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.43.ffn_gate_inp.weight
F32
F32
[2048, 64]
blk.43.ffn_gate_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.43.ffn_norm.weight
F32
F32
[2048]
blk.43.ffn_up_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.43.ffn_up_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.44
blk.44.attn_kv_a_mqa.weight
Q8_0
Q8_0
[2048, 576]
blk.44.attn_kv_a_norm.weight
F32
F32
[512]
blk.44.attn_kv_b.weight
Q8_0
Q8_0
[512, 8960]
blk.44.attn_norm.weight
F32
F32
[2048]
blk.44.attn_output.weight
Q8_0
Q8_0
[5120, 2048]
blk.44.attn_q_a.weight
Q8_0
Q8_0
[2048, 768]
blk.44.attn_q_a_norm.weight
F32
F32
[768]
blk.44.attn_q_b.weight
Q8_0
Q8_0
[768, 5120]
blk.44.exp_probs_b.bias
F32
F32
[64]
blk.44.ffn_down_exps.weight
Q8_0
Q8_0
[1536, 2048, 64]
blk.44.ffn_down_shexp.weight
Q8_0
Q8_0
[1536, 2048]
blk.44.ffn_gate_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.44.ffn_gate_inp.weight
F32
F32
[2048, 64]
blk.44.ffn_gate_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.44.ffn_norm.weight
F32
F32
[2048]
blk.44.ffn_up_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.44.ffn_up_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.45
blk.45.attn_kv_a_mqa.weight
Q8_0
Q8_0
[2048, 576]
blk.45.attn_kv_a_norm.weight
F32
F32
[512]
blk.45.attn_kv_b.weight
Q8_0
Q8_0
[512, 8960]
blk.45.attn_norm.weight
F32
F32
[2048]
blk.45.attn_output.weight
Q8_0
Q8_0
[5120, 2048]
blk.45.attn_q_a.weight
Q8_0
Q8_0
[2048, 768]
blk.45.attn_q_a_norm.weight
F32
F32
[768]
blk.45.attn_q_b.weight
Q8_0
Q8_0
[768, 5120]
blk.45.exp_probs_b.bias
F32
F32
[64]
blk.45.ffn_down_exps.weight
Q8_0
Q8_0
[1536, 2048, 64]
blk.45.ffn_down_shexp.weight
Q8_0
Q8_0
[1536, 2048]
blk.45.ffn_gate_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.45.ffn_gate_inp.weight
F32
F32
[2048, 64]
blk.45.ffn_gate_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.45.ffn_norm.weight
F32
F32
[2048]
blk.45.ffn_up_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.45.ffn_up_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.46
blk.46.attn_kv_a_mqa.weight
Q8_0
Q8_0
[2048, 576]
blk.46.attn_kv_a_norm.weight
F32
F32
[512]
blk.46.attn_kv_b.weight
Q8_0
Q8_0
[512, 8960]
blk.46.attn_norm.weight
F32
F32
[2048]
blk.46.attn_output.weight
Q8_0
Q8_0
[5120, 2048]
blk.46.attn_q_a.weight
Q8_0
Q8_0
[2048, 768]
blk.46.attn_q_a_norm.weight
F32
F32
[768]
blk.46.attn_q_b.weight
Q8_0
Q8_0
[768, 5120]
blk.46.exp_probs_b.bias
F32
F32
[64]
blk.46.ffn_down_exps.weight
Q8_0
Q8_0
[1536, 2048, 64]
blk.46.ffn_down_shexp.weight
Q8_0
Q8_0
[1536, 2048]
blk.46.ffn_gate_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.46.ffn_gate_inp.weight
F32
F32
[2048, 64]
blk.46.ffn_gate_shexp.weight
Q8_0
Q8_0
[2048, 1536]
blk.46.ffn_norm.weight
F32
F32
[2048]
blk.46.ffn_up_exps.weight
Q8_0
Q8_0
[2048, 1536, 64]
blk.46.ffn_up_shexp.weight
Q8_0
Q8_0
[2048, 1536]
output.weight
Q8_0
Q8_0
[2048, 154880]
output_norm.weight
F32
F32
[2048]