-
general.architecture
gemma3n
gemma3n
-
Q8_0
-
gemma3n.activation_sparsity_scale
[1.6448536, 1.6448536, 1.6448536, 1.6448536, 1.6448536, ...]
[1.6448536, 1.6448536, 1.6448536, 1.6448536, 1.6448536, ...]
-
gemma3n.altup.active_idx
0
0
-
gemma3n.altup.num_inputs
4
4
-
gemma3n.attention.head_count
8
8
-
gemma3n.attention.head_count_kv
2
2
-
gemma3n.attention.key_length
256
256
-
gemma3n.attention.layer_norm_rms_epsilon
1e-06
1e-06
-
gemma3n.attention.shared_kv_layers
15
15
-
gemma3n.attention.sliding_window
512
512
-
gemma3n.attention.sliding_window_pattern
[true, true, true, true, false, ...]
[true, true, true, true, false, ...]
-
gemma3n.attention.value_length
256
256
-
35
-
gemma3n.context_length
32768
32768
-
gemma3n.embedding_length
2048
2048
-
gemma3n.embedding_length_per_layer_input
256
256
-
gemma3n.feed_forward_length
[16384, 16384, 16384, 16384, 16384, ...]
[16384, 16384, 16384, 16384, 16384, ...]
-
gemma3n.rope.freq_base
1e+06
1e+06
-
tokenizer.ggml.add_bos_token
true
true
-
tokenizer.ggml.add_eos_token
false
false
-
tokenizer.ggml.add_sep_token
false
false
-
tokenizer.ggml.add_space_prefix
false
false
-
tokenizer.ggml.bos_token_id
2
2
-
tokenizer.ggml.eos_token_id
106
106
-
tokenizer.ggml.model
llama
llama
-
tokenizer.ggml.padding_token_id
0
0
-
tokenizer.ggml.pre
default
default
-
tokenizer.ggml.scores
[-1000, -1000, -1000, -1000, -1000, ...]
[-1000, -1000, -1000, -1000, -1000, ...]
-
tokenizer.ggml.token_type
[3, 3, 3, 3, 3, ...]
[3, 3, 3, 3, 3, ...]
-
tokenizer.ggml.tokens
[<pad>, <eos>, <bos>, <unk>, <mask>, ...]
[<pad>, <eos>, <bos>, <unk>, <mask>, ...]
-
tokenizer.ggml.unknown_token_id
3
3
-
Name
Type
Shape
-
Q8_0
[2048, 262144]
-
Q8_0
[2048, 2048, 3]
-
altup_unembd_proj.weight
Q8_0
Q8_0
[2048, 2048, 3]
-
blk.0.altup_correct_coef.weight
F32
F32
[4, 4]
-
blk.0.altup_correct_scale.weight
F32
F32
[2048]
-
blk.0.altup_predict_coef.weight
F32
F32
[4, 16]
-
blk.0.altup_router.weight
Q8_0
Q8_0
[2048, 4]
-
blk.0.altup_router_norm.weight
F32
F32
[2048]
-
Q8_0
[2048, 512]
-
blk.0.attn_k_norm.weight
F32
F32
[256]
-
blk.0.attn_norm.weight
F32
F32
[2048]
-
blk.0.attn_output.weight
Q8_0
Q8_0
[2048, 2048]
-
Q8_0
[2048, 2048]
-
blk.0.attn_q_norm.weight
F32
F32
[256]
-
Q8_0
[2048, 512]
-
blk.0.ffn_down.weight
Q8_0
Q8_0
[16384, 2048]
-
blk.0.ffn_gate.weight
Q8_0
Q8_0
[2048, 16384]
-
blk.0.ffn_norm.weight
F32
F32
[2048]
-
Q8_0
[2048, 16384]
-
blk.0.inp_gate.weight
Q8_0
Q8_0
[2048, 256]
-
blk.0.laurel_l.weight
Q8_0
Q8_0
[2048, 64]
-
blk.0.laurel_post_norm.weight
F32
F32
[2048]
-
blk.0.laurel_r.weight
Q8_0
Q8_0
[64, 2048]
-
blk.0.post_attention_norm.weight
F32
F32
[2048]
-
blk.0.post_ffw_norm.weight
F32
F32
[2048]
-
blk.0.post_norm.weight
F32
F32
[2048]
-
Q8_0
[256, 2048]
-
blk.1.altup_correct_coef.weight
F32
F32
[4, 4]
-
blk.1.altup_correct_scale.weight
F32
F32
[2048]
-
blk.1.altup_predict_coef.weight
F32
F32
[4, 16]
-
blk.1.altup_router.weight
Q8_0
Q8_0
[2048, 4]
-
blk.1.altup_router_norm.weight
F32
F32
[2048]
-
Q8_0
[2048, 512]
-
blk.1.attn_k_norm.weight
F32
F32
[256]
-
blk.1.attn_norm.weight
F32
F32
[2048]
-
blk.1.attn_output.weight
Q8_0
Q8_0
[2048, 2048]
-
Q8_0
[2048, 2048]
-
blk.1.attn_q_norm.weight
F32
F32
[256]
-
Q8_0
[2048, 512]
-
blk.1.ffn_down.weight
Q8_0
Q8_0
[16384, 2048]
-
blk.1.ffn_gate.weight
Q8_0
Q8_0
[2048, 16384]
-
blk.1.ffn_norm.weight
F32
F32
[2048]
-
Q8_0
[2048, 16384]
-
blk.1.inp_gate.weight
Q8_0
Q8_0
[2048, 256]
-
blk.1.laurel_l.weight
Q8_0
Q8_0
[2048, 64]
-
blk.1.laurel_post_norm.weight
F32
F32
[2048]
-
blk.1.laurel_r.weight
Q8_0
Q8_0
[64, 2048]
-
blk.1.post_attention_norm.weight
F32
F32
[2048]
-
blk.1.post_ffw_norm.weight
F32
F32
[2048]
-
blk.1.post_norm.weight
F32
F32
[2048]
-
Q8_0
[256, 2048]
-
blk.2.altup_correct_coef.weight
F32
F32
[4, 4]
-
blk.2.altup_correct_scale.weight
F32
F32
[2048]
-
blk.2.altup_predict_coef.weight
F32
F32
[4, 16]
-
blk.2.altup_router.weight
Q8_0
Q8_0
[2048, 4]
-
blk.2.altup_router_norm.weight
F32
F32
[2048]
-
Q8_0
[2048, 512]
-
blk.2.attn_k_norm.weight
F32
F32
[256]
-
blk.2.attn_norm.weight
F32
F32
[2048]
-
blk.2.attn_output.weight
Q8_0
Q8_0
[2048, 2048]
-
Q8_0
[2048, 2048]
-
blk.2.attn_q_norm.weight
F32
F32
[256]
-
Q8_0
[2048, 512]
-
blk.2.ffn_down.weight
Q8_0
Q8_0
[16384, 2048]
-
blk.2.ffn_gate.weight
Q8_0
Q8_0
[2048, 16384]
-
blk.2.ffn_norm.weight
F32
F32
[2048]
-
Q8_0
[2048, 16384]
-
blk.2.inp_gate.weight
Q8_0
Q8_0
[2048, 256]
-
blk.2.laurel_l.weight
Q8_0
Q8_0
[2048, 64]
-
blk.2.laurel_post_norm.weight
F32
F32
[2048]
-
blk.2.laurel_r.weight
Q8_0
Q8_0
[64, 2048]
-
blk.2.post_attention_norm.weight
F32
F32
[2048]
-
blk.2.post_ffw_norm.weight
F32
F32
[2048]
-
blk.2.post_norm.weight
F32
F32
[2048]
-
Q8_0
[256, 2048]
-
blk.3.altup_correct_coef.weight
F32
F32
[4, 4]
-
blk.3.altup_correct_scale.weight
F32
F32
[2048]
-
blk.3.altup_predict_coef.weight
F32
F32
[4, 16]
-
blk.3.altup_router.weight
Q8_0
Q8_0
[2048, 4]
-
blk.3.altup_router_norm.weight
F32
F32
[2048]
-
Q8_0
[2048, 512]
-
blk.3.attn_k_norm.weight
F32
F32
[256]
-
blk.3.attn_norm.weight
F32
F32
[2048]
-
blk.3.attn_output.weight
Q8_0
Q8_0
[2048, 2048]
-
Q8_0
[2048, 2048]
-
blk.3.attn_q_norm.weight
F32
F32
[256]
-
Q8_0
[2048, 512]
-
blk.3.ffn_down.weight
Q8_0
Q8_0
[16384, 2048]
-
blk.3.ffn_gate.weight
Q8_0
Q8_0
[2048, 16384]
-
blk.3.ffn_norm.weight
F32
F32
[2048]
-
Q8_0
[2048, 16384]
-
blk.3.inp_gate.weight
Q8_0
Q8_0
[2048, 256]
-
blk.3.laurel_l.weight
Q8_0
Q8_0
[2048, 64]
-
blk.3.laurel_post_norm.weight
F32
F32
[2048]
-
blk.3.laurel_r.weight
Q8_0
Q8_0
[64, 2048]
-
blk.3.post_attention_norm.weight
F32
F32
[2048]
-
blk.3.post_ffw_norm.weight
F32
F32
[2048]
-
blk.3.post_norm.weight
F32
F32
[2048]
-
Q8_0
[256, 2048]
-
blk.4.altup_correct_coef.weight
F32
F32
[4, 4]
-
blk.4.altup_correct_scale.weight
F32
F32
[2048]
-
blk.4.altup_predict_coef.weight
F32
F32
[4, 16]
-
blk.4.altup_router.weight
Q8_0
Q8_0
[2048, 4]
-
blk.4.altup_router_norm.weight
F32
F32
[2048]
-
Q8_0
[2048, 512]
-
blk.4.attn_k_norm.weight
F32
F32
[256]
-
blk.4.attn_norm.weight
F32
F32
[2048]
-
blk.4.attn_output.weight
Q8_0
Q8_0
[2048, 2048]
-
Q8_0
[2048, 2048]
-
blk.4.attn_q_norm.weight
F32
F32
[256]
-
Q8_0
[2048, 512]
-
blk.4.ffn_down.weight
Q8_0
Q8_0
[16384, 2048]
-
blk.4.ffn_gate.weight
Q8_0
Q8_0
[2048, 16384]
-
blk.4.ffn_norm.weight
F32
F32
[2048]
-
Q8_0
[2048, 16384]
-
blk.4.inp_gate.weight
Q8_0
Q8_0
[2048, 256]
-
blk.4.laurel_l.weight
Q8_0
Q8_0
[2048, 64]
-
blk.4.laurel_post_norm.weight
F32
F32
[2048]
-
blk.4.laurel_r.weight
Q8_0
Q8_0
[64, 2048]
-
blk.4.post_attention_norm.weight
F32
F32
[2048]
-
blk.4.post_ffw_norm.weight
F32
F32
[2048]
-
blk.4.post_norm.weight
F32
F32
[2048]
-
Q8_0
[256, 2048]
-
blk.5.altup_correct_coef.weight
F32
F32
[4, 4]
-
blk.5.altup_correct_scale.weight
F32
F32
[2048]
-
blk.5.altup_predict_coef.weight
F32
F32
[4, 16]
-
blk.5.altup_router.weight
Q8_0
Q8_0
[2048, 4]
-
blk.5.altup_router_norm.weight
F32
F32
[2048]
-
Q8_0
[2048, 512]
-
blk.5.attn_k_norm.weight
F32
F32
[256]
-
blk.5.attn_norm.weight
F32
F32
[2048]
-
blk.5.attn_output.weight
Q8_0
Q8_0
[2048, 2048]
-
Q8_0
[2048, 2048]
-
blk.5.attn_q_norm.weight
F32
F32
[256]
-
Q8_0
[2048, 512]
-
blk.5.ffn_down.weight
Q8_0
Q8_0
[16384, 2048]
-
blk.5.ffn_gate.weight
Q8_0
Q8_0
[2048, 16384]
-
blk.5.ffn_norm.weight
F32
F32
[2048]
-
Q8_0
[2048, 16384]
-
blk.5.inp_gate.weight
Q8_0
Q8_0
[2048, 256]
-
blk.5.laurel_l.weight
Q8_0
Q8_0
[2048, 64]
-
blk.5.laurel_post_norm.weight
F32
F32
[2048]
-
blk.5.laurel_r.weight
Q8_0
Q8_0
[64, 2048]
-
blk.5.post_attention_norm.weight
F32
F32
[2048]
-
blk.5.post_ffw_norm.weight
F32
F32
[2048]
-
blk.5.post_norm.weight
F32
F32
[2048]
-
Q8_0
[256, 2048]
-
blk.6.altup_correct_coef.weight
F32
F32
[4, 4]
-
blk.6.altup_correct_scale.weight
F32
F32
[2048]
-
blk.6.altup_predict_coef.weight
F32
F32
[4, 16]
-
blk.6.altup_router.weight
Q8_0
Q8_0
[2048, 4]
-
blk.6.altup_router_norm.weight
F32
F32
[2048]
-
Q8_0
[2048, 512]
-
blk.6.attn_k_norm.weight
F32
F32
[256]
-
blk.6.attn_norm.weight
F32
F32
[2048]
-
blk.6.attn_output.weight
Q8_0
Q8_0
[2048, 2048]
-
Q8_0
[2048, 2048]
-
blk.6.attn_q_norm.weight
F32
F32
[256]
-
Q8_0
[2048, 512]
-
blk.6.ffn_down.weight
Q8_0
Q8_0
[16384, 2048]
-
blk.6.ffn_gate.weight
Q8_0
Q8_0
[2048, 16384]
-
blk.6.ffn_norm.weight
F32
F32
[2048]
-
Q8_0
[2048, 16384]
-
blk.6.inp_gate.weight
Q8_0
Q8_0
[2048, 256]
-
blk.6.laurel_l.weight
Q8_0
Q8_0
[2048, 64]
-
blk.6.laurel_post_norm.weight
F32
F32
[2048]
-
blk.6.laurel_r.weight
Q8_0
Q8_0
[64, 2048]
-
blk.6.post_attention_norm.weight
F32
F32
[2048]
-
blk.6.post_ffw_norm.weight
F32
F32
[2048]
-
blk.6.post_norm.weight
F32
F32
[2048]
-
Q8_0
[256, 2048]
-
blk.7.altup_correct_coef.weight
F32
F32
[4, 4]
-
blk.7.altup_correct_scale.weight
F32
F32
[2048]
-
blk.7.altup_predict_coef.weight
F32
F32
[4, 16]
-
blk.7.altup_router.weight
Q8_0
Q8_0
[2048, 4]
-
blk.7.altup_router_norm.weight
F32
F32
[2048]
-
Q8_0
[2048, 512]
-
blk.7.attn_k_norm.weight
F32
F32
[256]
-
blk.7.attn_norm.weight
F32
F32
[2048]
-
blk.7.attn_output.weight
Q8_0
Q8_0
[2048, 2048]
-
Q8_0
[2048, 2048]
-
blk.7.attn_q_norm.weight
F32
F32
[256]
-
Q8_0
[2048, 512]
-
blk.7.ffn_down.weight
Q8_0
Q8_0
[16384, 2048]
-
blk.7.ffn_gate.weight
Q8_0
Q8_0
[2048, 16384]
-
blk.7.ffn_norm.weight
F32
F32
[2048]
-
Q8_0
[2048, 16384]
-
blk.7.inp_gate.weight
Q8_0
Q8_0
[2048, 256]
-
blk.7.laurel_l.weight
Q8_0
Q8_0
[2048, 64]
-
blk.7.laurel_post_norm.weight
F32
F32
[2048]
-
blk.7.laurel_r.weight
Q8_0
Q8_0
[64, 2048]
-
blk.7.post_attention_norm.weight
F32
F32
[2048]
-
blk.7.post_ffw_norm.weight
F32
F32
[2048]
-
blk.7.post_norm.weight
F32
F32
[2048]
-
Q8_0
[256, 2048]
-
blk.8.altup_correct_coef.weight
F32
F32
[4, 4]
-
blk.8.altup_correct_scale.weight
F32
F32
[2048]
-
blk.8.altup_predict_coef.weight
F32
F32
[4, 16]
-
blk.8.altup_router.weight
Q8_0
Q8_0
[2048, 4]
-
blk.8.altup_router_norm.weight
F32
F32
[2048]
-
Q8_0
[2048, 512]
-
blk.8.attn_k_norm.weight
F32
F32
[256]
-
blk.8.attn_norm.weight
F32
F32
[2048]
-
blk.8.attn_output.weight
Q8_0
Q8_0
[2048, 2048]
-
Q8_0
[2048, 2048]
-
blk.8.attn_q_norm.weight
F32
F32
[256]
-
Q8_0
[2048, 512]
-
blk.8.ffn_down.weight
Q8_0
Q8_0
[16384, 2048]
-
blk.8.ffn_gate.weight
Q8_0
Q8_0
[2048, 16384]
-
blk.8.ffn_norm.weight
F32
F32
[2048]
-
Q8_0
[2048, 16384]
-
blk.8.inp_gate.weight
Q8_0
Q8_0
[2048, 256]
-
blk.8.laurel_l.weight
Q8_0
Q8_0
[2048, 64]
-
blk.8.laurel_post_norm.weight
F32
F32
[2048]
-
blk.8.laurel_r.weight
Q8_0
Q8_0
[64, 2048]
-
blk.8.post_attention_norm.weight
F32
F32
[2048]
-
blk.8.post_ffw_norm.weight
F32
F32
[2048]
-
blk.8.post_norm.weight
F32
F32
[2048]
-
Q8_0
[256, 2048]
-
blk.9.altup_correct_coef.weight
F32
F32
[4, 4]
-
blk.9.altup_correct_scale.weight
F32
F32
[2048]
-
blk.9.altup_predict_coef.weight
F32
F32
[4, 16]
-
blk.9.altup_router.weight
Q8_0
Q8_0
[2048, 4]
-
blk.9.altup_router_norm.weight
F32
F32
[2048]
-
Q8_0
[2048, 512]
-
blk.9.attn_k_norm.weight
F32
F32
[256]
-
blk.9.attn_norm.weight
F32
F32
[2048]
-
blk.9.attn_output.weight
Q8_0
Q8_0
[2048, 2048]
-
Q8_0
[2048, 2048]
-
blk.9.attn_q_norm.weight
F32
F32
[256]
-
Q8_0
[2048, 512]
-
blk.9.ffn_down.weight
Q8_0
Q8_0
[16384, 2048]
-
blk.9.ffn_gate.weight
Q8_0
Q8_0
[2048, 16384]
-
blk.9.ffn_norm.weight
F32
F32
[2048]
-
Q8_0
[2048, 16384]
-
blk.9.inp_gate.weight
Q8_0
Q8_0
[2048, 256]
-
blk.9.laurel_l.weight
Q8_0
Q8_0
[2048, 64]
-
blk.9.laurel_post_norm.weight
F32
F32
[2048]
-
blk.9.laurel_r.weight
Q8_0
Q8_0
[64, 2048]
-
blk.9.post_attention_norm.weight
F32
F32
[2048]
-
blk.9.post_ffw_norm.weight
F32
F32
[2048]
-
blk.9.post_norm.weight
F32
F32
[2048]
-
Q8_0
[256, 2048]
-
blk.10.altup_correct_coef.weight
F32
F32
[4, 4]
-
blk.10.altup_correct_scale.weight
F32
F32
[2048]
-
blk.10.altup_predict_coef.weight
F32
F32
[4, 16]
-
blk.10.altup_router.weight
Q8_0
Q8_0
[2048, 4]
-
blk.10.altup_router_norm.weight
F32
F32
[2048]
-
blk.10.attn_k.weight
Q8_0
Q8_0
[2048, 512]
-
blk.10.attn_k_norm.weight
F32
F32
[256]
-
blk.10.attn_norm.weight
F32
F32
[2048]
-
blk.10.attn_output.weight
Q8_0
Q8_0
[2048, 2048]
-
blk.10.attn_q.weight
Q8_0
Q8_0
[2048, 2048]
-
blk.10.attn_q_norm.weight
F32
F32
[256]
-
blk.10.attn_v.weight
Q8_0
Q8_0
[2048, 512]
-
blk.10.ffn_down.weight
Q8_0
Q8_0
[16384, 2048]
-
blk.10.ffn_gate.weight
Q8_0
Q8_0
[2048, 16384]
-
blk.10.ffn_norm.weight
F32
F32
[2048]
-
blk.10.ffn_up.weight
Q8_0
Q8_0
[2048, 16384]
-
blk.10.inp_gate.weight
Q8_0
Q8_0
[2048, 256]
-
blk.10.laurel_l.weight
Q8_0
Q8_0
[2048, 64]
-
blk.10.laurel_post_norm.weight
F32
F32
[2048]
-
blk.10.laurel_r.weight
Q8_0
Q8_0
[64, 2048]
-
blk.10.post_attention_norm.weight
F32
F32
[2048]
-
blk.10.post_ffw_norm.weight
F32
F32
[2048]
-
blk.10.post_norm.weight
F32
F32
[2048]
-
Q8_0
[256, 2048]
-
blk.11.altup_correct_coef.weight
F32
F32
[4, 4]
-
blk.11.altup_correct_scale.weight
F32
F32
[2048]
-
blk.11.altup_predict_coef.weight
F32
F32
[4, 16]
-
blk.11.altup_router.weight
Q8_0
Q8_0
[2048, 4]
-
blk.11.altup_router_norm.weight
F32
F32
[2048]
-
blk.11.attn_k.weight
Q8_0
Q8_0
[2048, 512]
-
blk.11.attn_k_norm.weight
F32
F32
[256]
-
blk.11.attn_norm.weight
F32
F32
[2048]
-
blk.11.attn_output.weight
Q8_0
Q8_0
[2048, 2048]
-
blk.11.attn_q.weight
Q8_0
Q8_0
[2048, 2048]
-
blk.11.attn_q_norm.weight
F32
F32
[256]
-
blk.11.attn_v.weight
Q8_0
Q8_0
[2048, 512]
-
blk.11.ffn_down.weight
Q8_0
Q8_0
[16384, 2048]
-
blk.11.ffn_gate.weight
Q8_0
Q8_0
[2048, 16384]
-
blk.11.ffn_norm.weight
F32
F32
[2048]
-
blk.11.ffn_up.weight
Q8_0
Q8_0
[2048, 16384]
-
blk.11.inp_gate.weight
Q8_0
Q8_0
[2048, 256]
-
blk.11.laurel_l.weight
Q8_0
Q8_0
[2048, 64]
-
blk.11.laurel_post_norm.weight
F32
F32
[2048]
-
blk.11.laurel_r.weight
Q8_0
Q8_0
[64, 2048]
-
blk.11.post_attention_norm.weight
F32
F32
[2048]
-
blk.11.post_ffw_norm.weight
F32
F32
[2048]
-
blk.11.post_norm.weight
F32
F32
[2048]
-
Q8_0
[256, 2048]
-
blk.12.altup_correct_coef.weight
F32
F32
[4, 4]
-
blk.12.altup_correct_scale.weight
F32
F32
[2048]
-
blk.12.altup_predict_coef.weight
F32
F32
[4, 16]
-
blk.12.altup_router.weight
Q8_0
Q8_0
[2048, 4]
-
blk.12.altup_router_norm.weight
F32
F32
[2048]
-
blk.12.attn_k.weight
Q8_0
Q8_0
[2048, 512]
-
blk.12.attn_k_norm.weight
F32
F32
[256]
-
blk.12.attn_norm.weight
F32
F32
[2048]
-
blk.12.attn_output.weight
Q8_0
Q8_0
[2048, 2048]
-
blk.12.attn_q.weight
Q8_0
Q8_0
[2048, 2048]
-
blk.12.attn_q_norm.weight
F32
F32
[256]
-
blk.12.attn_v.weight
Q8_0
Q8_0
[2048, 512]
-
blk.12.ffn_down.weight
Q8_0
Q8_0
[16384, 2048]
-
blk.12.ffn_gate.weight
Q8_0
Q8_0
[2048, 16384]
-
blk.12.ffn_norm.weight
F32
F32
[2048]
-
blk.12.ffn_up.weight
Q8_0
Q8_0
[2048, 16384]
-
blk.12.inp_gate.weight
Q8_0
Q8_0
[2048, 256]
-
blk.12.laurel_l.weight
Q8_0
Q8_0
[2048, 64]
-
blk.12.laurel_post_norm.weight
F32
F32
[2048]
-
blk.12.laurel_r.weight
Q8_0
Q8_0
[64, 2048]
-
blk.12.post_attention_norm.weight
F32
F32
[2048]
-
blk.12.post_ffw_norm.weight
F32
F32
[2048]
-
blk.12.post_norm.weight
F32
F32
[2048]
-
Q8_0
[256, 2048]
-
blk.13.altup_correct_coef.weight
F32
F32
[4, 4]
-
blk.13.altup_correct_scale.weight
F32
F32
[2048]
-
blk.13.altup_predict_coef.weight
F32
F32
[4, 16]
-
blk.13.altup_router.weight
Q8_0
Q8_0
[2048, 4]
-
blk.13.altup_router_norm.weight
F32
F32
[2048]
-
blk.13.attn_k.weight
Q8_0
Q8_0
[2048, 512]
-
blk.13.attn_k_norm.weight
F32
F32
[256]
-
blk.13.attn_norm.weight
F32
F32
[2048]
-
blk.13.attn_output.weight
Q8_0
Q8_0
[2048, 2048]
-
blk.13.attn_q.weight
Q8_0
Q8_0
[2048, 2048]
-
blk.13.attn_q_norm.weight
F32
F32
[256]
-
blk.13.attn_v.weight
Q8_0
Q8_0
[2048, 512]
-
blk.13.ffn_down.weight
Q8_0
Q8_0
[16384, 2048]
-
blk.13.ffn_gate.weight
Q8_0
Q8_0
[2048, 16384]
-
blk.13.ffn_norm.weight
F32
F32
[2048]
-
blk.13.ffn_up.weight
Q8_0
Q8_0
[2048, 16384]
-
blk.13.inp_gate.weight
Q8_0
Q8_0
[2048, 256]
-
blk.13.laurel_l.weight
Q8_0
Q8_0
[2048, 64]
-
blk.13.laurel_post_norm.weight
F32
F32
[2048]
-
blk.13.laurel_r.weight
Q8_0
Q8_0
[64, 2048]
-
blk.13.post_attention_norm.weight
F32
F32
[2048]
-
blk.13.post_ffw_norm.weight
F32
F32
[2048]
-
blk.13.post_norm.weight
F32
F32
[2048]
-
Q8_0
[256, 2048]
-
blk.14.altup_correct_coef.weight
F32
F32
[4, 4]
-
blk.14.altup_correct_scale.weight
F32
F32
[2048]
-
blk.14.altup_predict_coef.weight
F32
F32
[4, 16]
-
blk.14.altup_router.weight
Q8_0
Q8_0
[2048, 4]
-
blk.14.altup_router_norm.weight
F32
F32
[2048]
-
blk.14.attn_k.weight
Q8_0
Q8_0
[2048, 512]
-
blk.14.attn_k_norm.weight
F32
F32
[256]
-
blk.14.attn_norm.weight
F32
F32
[2048]
-
blk.14.attn_output.weight
Q8_0
Q8_0
[2048, 2048]
-
blk.14.attn_q.weight
Q8_0
Q8_0
[2048, 2048]
-
blk.14.attn_q_norm.weight
F32
F32
[256]
-
blk.14.attn_v.weight
Q8_0
Q8_0
[2048, 512]
-
blk.14.ffn_down.weight
Q8_0
Q8_0
[16384, 2048]
-
blk.14.ffn_gate.weight
Q8_0
Q8_0
[2048, 16384]
-
blk.14.ffn_norm.weight
F32
F32
[2048]
-
blk.14.ffn_up.weight
Q8_0
Q8_0
[2048, 16384]
-
blk.14.inp_gate.weight
Q8_0
Q8_0
[2048, 256]
-
blk.14.laurel_l.weight
Q8_0
Q8_0
[2048, 64]
-
blk.14.laurel_post_norm.weight
F32
F32
[2048]
-
blk.14.laurel_r.weight
Q8_0
Q8_0
[64, 2048]
-
blk.14.post_attention_norm.weight
F32
F32
[2048]
-
blk.14.post_ffw_norm.weight
F32
F32
[2048]
-
blk.14.post_norm.weight
F32
F32
[2048]
-
Q8_0
[256, 2048]
-
blk.15.altup_correct_coef.weight
F32
F32
[4, 4]
-
blk.15.altup_correct_scale.weight
F32
F32
[2048]
-
blk.15.altup_predict_coef.weight
F32
F32
[4, 16]
-
blk.15.altup_router.weight
Q8_0
Q8_0
[2048, 4]
-
blk.15.altup_router_norm.weight
F32
F32
[2048]
-
blk.15.attn_k.weight
Q8_0
Q8_0
[2048, 512]
-
blk.15.attn_k_norm.weight
F32
F32
[256]
-
blk.15.attn_norm.weight
F32
F32
[2048]
-
blk.15.attn_output.weight
Q8_0
Q8_0
[2048, 2048]
-
blk.15.attn_q.weight
Q8_0
Q8_0
[2048, 2048]
-
blk.15.attn_q_norm.weight
F32
F32
[256]
-
blk.15.attn_v.weight
Q8_0
Q8_0
[2048, 512]
-
blk.15.ffn_down.weight
Q8_0
Q8_0
[16384, 2048]
-
blk.15.ffn_gate.weight
Q8_0
Q8_0
[2048, 16384]
-
blk.15.ffn_norm.weight
F32
F32
[2048]
-
blk.15.ffn_up.weight
Q8_0
Q8_0
[2048, 16384]
-
blk.15.inp_gate.weight
Q8_0
Q8_0
[2048, 256]
-
blk.15.laurel_l.weight
Q8_0
Q8_0
[2048, 64]
-
blk.15.laurel_post_norm.weight
F32
F32
[2048]
-
blk.15.laurel_r.weight
Q8_0
Q8_0
[64, 2048]
-
blk.15.post_attention_norm.weight
F32
F32
[2048]
-
blk.15.post_ffw_norm.weight
F32
F32
[2048]
-
blk.15.post_norm.weight
F32
F32
[2048]
-
Q8_0
[256, 2048]
-
blk.16.altup_correct_coef.weight
F32
F32
[4, 4]
-
blk.16.altup_correct_scale.weight
F32
F32
[2048]
-
blk.16.altup_predict_coef.weight
F32
F32
[4, 16]
-
blk.16.altup_router.weight
Q8_0
Q8_0
[2048, 4]
-
blk.16.altup_router_norm.weight
F32
F32
[2048]
-
blk.16.attn_k.weight
Q8_0
Q8_0
[2048, 512]
-
blk.16.attn_k_norm.weight
F32
F32
[256]
-
blk.16.attn_norm.weight
F32
F32
[2048]
-
blk.16.attn_output.weight
Q8_0
Q8_0
[2048, 2048]
-
blk.16.attn_q.weight
Q8_0
Q8_0
[2048, 2048]
-
blk.16.attn_q_norm.weight
F32
F32
[256]
-
blk.16.attn_v.weight
Q8_0
Q8_0
[2048, 512]
-
blk.16.ffn_down.weight
Q8_0
Q8_0
[16384, 2048]
-
blk.16.ffn_gate.weight
Q8_0
Q8_0
[2048, 16384]
-
blk.16.ffn_norm.weight
F32
F32
[2048]
-
blk.16.ffn_up.weight
Q8_0
Q8_0
[2048, 16384]
-
blk.16.inp_gate.weight
Q8_0
Q8_0
[2048, 256]
-
blk.16.laurel_l.weight
Q8_0
Q8_0
[2048, 64]
-
blk.16.laurel_post_norm.weight
F32
F32
[2048]
-
blk.16.laurel_r.weight
Q8_0
Q8_0
[64, 2048]
-
blk.16.post_attention_norm.weight
F32
F32
[2048]
-
blk.16.post_ffw_norm.weight
F32
F32
[2048]
-
blk.16.post_norm.weight
F32
F32
[2048]
-
Q8_0
[256, 2048]
-
blk.17.altup_correct_coef.weight
F32
F32
[4, 4]
-
blk.17.altup_correct_scale.weight
F32
F32
[2048]
-
blk.17.altup_predict_coef.weight
F32
F32
[4, 16]
-
blk.17.altup_router.weight
Q8_0
Q8_0
[2048, 4]
-
blk.17.altup_router_norm.weight
F32
F32
[2048]
-
blk.17.attn_k.weight
Q8_0
Q8_0
[2048, 512]
-
blk.17.attn_k_norm.weight
F32
F32
[256]
-
blk.17.attn_norm.weight
F32
F32
[2048]
-
blk.17.attn_output.weight
Q8_0
Q8_0
[2048, 2048]
-
blk.17.attn_q.weight
Q8_0
Q8_0
[2048, 2048]
-
blk.17.attn_q_norm.weight
F32
F32
[256]
-
blk.17.attn_v.weight
Q8_0
Q8_0
[2048, 512]
-
blk.17.ffn_down.weight
Q8_0
Q8_0
[16384, 2048]
-
blk.17.ffn_gate.weight
Q8_0
Q8_0
[2048, 16384]
-
blk.17.ffn_norm.weight
F32
F32
[2048]
-
blk.17.ffn_up.weight
Q8_0
Q8_0
[2048, 16384]
-
blk.17.inp_gate.weight
Q8_0
Q8_0
[2048, 256]
-
blk.17.laurel_l.weight
Q8_0
Q8_0
[2048, 64]
-
blk.17.laurel_post_norm.weight
F32
F32
[2048]
-
blk.17.laurel_r.weight
Q8_0
Q8_0
[64, 2048]
-
blk.17.post_attention_norm.weight
F32
F32
[2048]
-
blk.17.post_ffw_norm.weight
F32
F32
[2048]
-
blk.17.post_norm.weight
F32
F32
[2048]
-
Q8_0
[256, 2048]
-
blk.18.altup_correct_coef.weight
F32
F32
[4, 4]
-
blk.18.altup_correct_scale.weight
F32
F32
[2048]
-
blk.18.altup_predict_coef.weight
F32
F32
[4, 16]
-
blk.18.altup_router.weight
Q8_0
Q8_0
[2048, 4]
-
blk.18.altup_router_norm.weight
F32
F32
[2048]
-
blk.18.attn_k.weight
Q8_0
Q8_0
[2048, 512]
-
blk.18.attn_k_norm.weight
F32
F32
[256]
-
blk.18.attn_norm.weight
F32
F32
[2048]
-
blk.18.attn_output.weight
Q8_0
Q8_0
[2048, 2048]
-
blk.18.attn_q.weight
Q8_0
Q8_0
[2048, 2048]
-
blk.18.attn_q_norm.weight
F32
F32
[256]
-
blk.18.attn_v.weight
Q8_0
Q8_0
[2048, 512]
-
blk.18.ffn_down.weight
Q8_0
Q8_0
[16384, 2048]
-
blk.18.ffn_gate.weight
Q8_0
Q8_0
[2048, 16384]
-
blk.18.ffn_norm.weight
F32
F32
[2048]
-
blk.18.ffn_up.weight
Q8_0
Q8_0
[2048, 16384]
-
blk.18.inp_gate.weight
Q8_0
Q8_0
[2048, 256]
-
blk.18.laurel_l.weight
Q8_0
Q8_0
[2048, 64]
-
blk.18.laurel_post_norm.weight
F32
F32
[2048]
-
blk.18.laurel_r.weight
Q8_0
Q8_0
[64, 2048]
-
blk.18.post_attention_norm.weight
F32
F32
[2048]
-
blk.18.post_ffw_norm.weight
F32
F32
[2048]
-
blk.18.post_norm.weight
F32
F32
[2048]
-
Q8_0
[256, 2048]
-
blk.19.altup_correct_coef.weight
F32
F32
[4, 4]
-
blk.19.altup_correct_scale.weight
F32
F32
[2048]
-
blk.19.altup_predict_coef.weight
F32
F32
[4, 16]
-
blk.19.altup_router.weight
Q8_0
Q8_0
[2048, 4]
-
blk.19.altup_router_norm.weight
F32
F32
[2048]
-
blk.19.attn_k.weight
Q8_0
Q8_0
[2048, 512]
-
blk.19.attn_k_norm.weight
F32
F32
[256]
-
blk.19.attn_norm.weight
F32
F32
[2048]
-
blk.19.attn_output.weight
Q8_0
Q8_0
[2048, 2048]
-
blk.19.attn_q.weight
Q8_0
Q8_0
[2048, 2048]
-
blk.19.attn_q_norm.weight
F32
F32
[256]
-
blk.19.attn_v.weight
Q8_0
Q8_0
[2048, 512]
-
blk.19.ffn_down.weight
Q8_0
Q8_0
[16384, 2048]
-
blk.19.ffn_gate.weight
Q8_0
Q8_0
[2048, 16384]
-
blk.19.ffn_norm.weight
F32
F32
[2048]
-
blk.19.ffn_up.weight
Q8_0
Q8_0
[2048, 16384]
-
blk.19.inp_gate.weight
Q8_0
Q8_0
[2048, 256]
-
blk.19.laurel_l.weight
Q8_0
Q8_0
[2048, 64]
-
blk.19.laurel_post_norm.weight
F32
F32
[2048]
-
blk.19.laurel_r.weight
Q8_0
Q8_0
[64, 2048]
-
blk.19.post_attention_norm.weight
F32
F32
[2048]
-
blk.19.post_ffw_norm.weight
F32
F32
[2048]
-
blk.19.post_norm.weight
F32
F32
[2048]
-
Q8_0
[256, 2048]
-
blk.20.altup_correct_coef.weight
F32
F32
[4, 4]
-
blk.20.altup_correct_scale.weight
F32
F32
[2048]
-
blk.20.altup_predict_coef.weight
F32
F32
[4, 16]
-
blk.20.altup_router.weight
Q8_0
Q8_0
[2048, 4]
-
blk.20.altup_router_norm.weight
F32
F32
[2048]
-
blk.20.attn_k.weight
Q8_0
Q8_0
[2048, 512]
-
blk.20.attn_k_norm.weight
F32
F32
[256]
-
blk.20.attn_norm.weight
F32
F32
[2048]
-
blk.20.attn_output.weight
Q8_0
Q8_0
[2048, 2048]
-
blk.20.attn_q.weight
Q8_0
Q8_0
[2048, 2048]
-
blk.20.attn_q_norm.weight
F32
F32
[256]
-
blk.20.attn_v.weight
Q8_0
Q8_0
[2048, 512]
-
blk.20.ffn_down.weight
Q8_0
Q8_0
[16384, 2048]
-
blk.20.ffn_gate.weight
Q8_0
Q8_0
[2048, 16384]
-
blk.20.ffn_norm.weight
F32
F32
[2048]
-
blk.20.ffn_up.weight
Q8_0
Q8_0
[2048, 16384]
-
blk.20.inp_gate.weight
Q8_0
Q8_0
[2048, 256]
-
blk.20.laurel_l.weight
Q8_0
Q8_0
[2048, 64]
-
blk.20.laurel_post_norm.weight
F32
F32
[2048]
-
blk.20.laurel_r.weight
Q8_0
Q8_0
[64, 2048]
-
blk.20.post_attention_norm.weight
F32
F32
[2048]
-
blk.20.post_ffw_norm.weight
F32
F32
[2048]
-
blk.20.post_norm.weight
F32
F32
[2048]
-
Q8_0
[256, 2048]
-
blk.21.altup_correct_coef.weight
F32
F32
[4, 4]
-
blk.21.altup_correct_scale.weight
F32
F32
[2048]
-
blk.21.altup_predict_coef.weight
F32
F32
[4, 16]
-
blk.21.altup_router.weight
Q8_0
Q8_0
[2048, 4]
-
blk.21.altup_router_norm.weight
F32
F32
[2048]
-
blk.21.attn_k.weight
Q8_0
Q8_0
[2048, 512]
-
blk.21.attn_k_norm.weight
F32
F32
[256]
-
blk.21.attn_norm.weight
F32
F32
[2048]
-
blk.21.attn_output.weight
Q8_0
Q8_0
[2048, 2048]
-
blk.21.attn_q.weight
Q8_0
Q8_0
[2048, 2048]
-
blk.21.attn_q_norm.weight
F32
F32
[256]
-
blk.21.attn_v.weight
Q8_0
Q8_0
[2048, 512]
-
blk.21.ffn_down.weight
Q8_0
Q8_0
[16384, 2048]
-
blk.21.ffn_gate.weight
Q8_0
Q8_0
[2048, 16384]
-
blk.21.ffn_norm.weight
F32
F32
[2048]
-
blk.21.ffn_up.weight
Q8_0
Q8_0
[2048, 16384]
-
blk.21.inp_gate.weight
Q8_0
Q8_0
[2048, 256]
-
blk.21.laurel_l.weight
Q8_0
Q8_0
[2048, 64]
-
blk.21.laurel_post_norm.weight
F32
F32
[2048]
-
blk.21.laurel_r.weight
Q8_0
Q8_0
[64, 2048]
-
blk.21.post_attention_norm.weight
F32
F32
[2048]
-
blk.21.post_ffw_norm.weight
F32
F32
[2048]
-
blk.21.post_norm.weight
F32
F32
[2048]
-
Q8_0
[256, 2048]
-
blk.22.altup_correct_coef.weight
F32
F32
[4, 4]
-
blk.22.altup_correct_scale.weight
F32
F32
[2048]
-
blk.22.altup_predict_coef.weight
F32
F32
[4, 16]
-
blk.22.altup_router.weight
Q8_0
Q8_0
[2048, 4]
-
blk.22.altup_router_norm.weight
F32
F32
[2048]
-
blk.22.attn_k.weight
Q8_0
Q8_0
[2048, 512]
-
blk.22.attn_k_norm.weight
F32
F32
[256]
-
blk.22.attn_norm.weight
F32
F32
[2048]
-
blk.22.attn_output.weight
Q8_0
Q8_0
[2048, 2048]
-
blk.22.attn_q.weight
Q8_0
Q8_0
[2048, 2048]
-
blk.22.attn_q_norm.weight
F32
F32
[256]
-
blk.22.attn_v.weight
Q8_0
Q8_0
[2048, 512]
-
blk.22.ffn_down.weight
Q8_0
Q8_0
[16384, 2048]
-
blk.22.ffn_gate.weight
Q8_0
Q8_0
[2048, 16384]
-
blk.22.ffn_norm.weight
F32
F32
[2048]
-
blk.22.ffn_up.weight
Q8_0
Q8_0
[2048, 16384]
-
blk.22.inp_gate.weight
Q8_0
Q8_0
[2048, 256]
-
blk.22.laurel_l.weight
Q8_0
Q8_0
[2048, 64]
-
blk.22.laurel_post_norm.weight
F32
F32
[2048]
-
blk.22.laurel_r.weight
Q8_0
Q8_0
[64, 2048]
-
blk.22.post_attention_norm.weight
F32
F32
[2048]
-
blk.22.post_ffw_norm.weight
F32
F32
[2048]
-
blk.22.post_norm.weight
F32
F32
[2048]
-
Q8_0
[256, 2048]
-
blk.23.altup_correct_coef.weight
F32
F32
[4, 4]
-
blk.23.altup_correct_scale.weight
F32
F32
[2048]
-
blk.23.altup_predict_coef.weight
F32
F32
[4, 16]
-
blk.23.altup_router.weight
Q8_0
Q8_0
[2048, 4]
-
blk.23.altup_router_norm.weight
F32
F32
[2048]
-
blk.23.attn_k.weight
Q8_0
Q8_0
[2048, 512]
-
blk.23.attn_k_norm.weight
F32
F32
[256]
-
blk.23.attn_norm.weight
F32
F32
[2048]
-
blk.23.attn_output.weight
Q8_0
Q8_0
[2048, 2048]
-
blk.23.attn_q.weight
Q8_0
Q8_0
[2048, 2048]
-
blk.23.attn_q_norm.weight
F32
F32
[256]
-
blk.23.attn_v.weight
Q8_0
Q8_0
[2048, 512]
-
blk.23.ffn_down.weight
Q8_0
Q8_0
[16384, 2048]
-
blk.23.ffn_gate.weight
Q8_0
Q8_0
[2048, 16384]
-
blk.23.ffn_norm.weight
F32
F32
[2048]
-
blk.23.ffn_up.weight
Q8_0
Q8_0
[2048, 16384]
-
blk.23.inp_gate.weight
Q8_0
Q8_0
[2048, 256]
-
blk.23.laurel_l.weight
Q8_0
Q8_0
[2048, 64]
-
blk.23.laurel_post_norm.weight
F32
F32
[2048]
-
blk.23.laurel_r.weight
Q8_0
Q8_0
[64, 2048]
-
blk.23.post_attention_norm.weight
F32
F32
[2048]
-
blk.23.post_ffw_norm.weight
F32
F32
[2048]
-
blk.23.post_norm.weight
F32
F32
[2048]
-
Q8_0
[256, 2048]
-
blk.24.altup_correct_coef.weight
F32
F32
[4, 4]
-
blk.24.altup_correct_scale.weight
F32
F32
[2048]
-
blk.24.altup_predict_coef.weight
F32
F32
[4, 16]
-
blk.24.altup_router.weight
Q8_0
Q8_0
[2048, 4]
-
blk.24.altup_router_norm.weight
F32
F32
[2048]
-
blk.24.attn_k.weight
Q8_0
Q8_0
[2048, 512]
-
blk.24.attn_k_norm.weight
F32
F32
[256]
-
blk.24.attn_norm.weight
F32
F32
[2048]
-
blk.24.attn_output.weight
Q8_0
Q8_0
[2048, 2048]
-
blk.24.attn_q.weight
Q8_0
Q8_0
[2048, 2048]
-
blk.24.attn_q_norm.weight
F32
F32
[256]
-
blk.24.attn_v.weight
Q8_0
Q8_0
[2048, 512]
-
blk.24.ffn_down.weight
Q8_0
Q8_0
[16384, 2048]
-
blk.24.ffn_gate.weight
Q8_0
Q8_0
[2048, 16384]
-
blk.24.ffn_norm.weight
F32
F32
[2048]
-
blk.24.ffn_up.weight
Q8_0
Q8_0
[2048, 16384]
-
blk.24.inp_gate.weight
Q8_0
Q8_0
[2048, 256]
-
blk.24.laurel_l.weight
Q8_0
Q8_0
[2048, 64]
-
blk.24.laurel_post_norm.weight
F32
F32
[2048]
-
blk.24.laurel_r.weight
Q8_0
Q8_0
[64, 2048]
-
blk.24.post_attention_norm.weight
F32
F32
[2048]
-
blk.24.post_ffw_norm.weight
F32
F32
[2048]
-
blk.24.post_norm.weight
F32
F32
[2048]
-
Q8_0
[256, 2048]
-
blk.25.altup_correct_coef.weight
F32
F32
[4, 4]
-
blk.25.altup_correct_scale.weight
F32
F32
[2048]
-
blk.25.altup_predict_coef.weight
F32
F32
[4, 16]
-
blk.25.altup_router.weight
Q8_0
Q8_0
[2048, 4]
-
blk.25.altup_router_norm.weight
F32
F32
[2048]
-
blk.25.attn_k.weight
Q8_0
Q8_0
[2048, 512]
-
blk.25.attn_k_norm.weight
F32
F32
[256]
-
blk.25.attn_norm.weight
F32
F32
[2048]
-
blk.25.attn_output.weight
Q8_0
Q8_0
[2048, 2048]
-
blk.25.attn_q.weight
Q8_0
Q8_0
[2048, 2048]
-
blk.25.attn_q_norm.weight
F32
F32
[256]
-
blk.25.attn_v.weight
Q8_0
Q8_0
[2048, 512]
-
blk.25.ffn_down.weight
Q8_0
Q8_0
[16384, 2048]
-
blk.25.ffn_gate.weight
Q8_0
Q8_0
[2048, 16384]
-
blk.25.ffn_norm.weight
F32
F32
[2048]
-
blk.25.ffn_up.weight
Q8_0
Q8_0
[2048, 16384]
-
blk.25.inp_gate.weight
Q8_0
Q8_0
[2048, 256]
-
blk.25.laurel_l.weight
Q8_0
Q8_0
[2048, 64]
-
blk.25.laurel_post_norm.weight
F32
F32
[2048]
-
blk.25.laurel_r.weight
Q8_0
Q8_0
[64, 2048]
-
blk.25.post_attention_norm.weight
F32
F32
[2048]
-
blk.25.post_ffw_norm.weight
F32
F32
[2048]
-
blk.25.post_norm.weight
F32
F32
[2048]
-
Q8_0
[256, 2048]
-
blk.26.altup_correct_coef.weight
F32
F32
[4, 4]
-
blk.26.altup_correct_scale.weight
F32
F32
[2048]
-
blk.26.altup_predict_coef.weight
F32
F32
[4, 16]
-
blk.26.altup_router.weight
Q8_0
Q8_0
[2048, 4]
-
blk.26.altup_router_norm.weight
F32
F32
[2048]
-
blk.26.attn_k.weight
Q8_0
Q8_0
[2048, 512]
-
blk.26.attn_k_norm.weight
F32
F32
[256]
-
blk.26.attn_norm.weight
F32
F32
[2048]
-
blk.26.attn_output.weight
Q8_0
Q8_0
[2048, 2048]
-
blk.26.attn_q.weight
Q8_0
Q8_0
[2048, 2048]
-
blk.26.attn_q_norm.weight
F32
F32
[256]
-
blk.26.attn_v.weight
Q8_0
Q8_0
[2048, 512]
-
blk.26.ffn_down.weight
Q8_0
Q8_0
[16384, 2048]
-
blk.26.ffn_gate.weight
Q8_0
Q8_0
[2048, 16384]
-
blk.26.ffn_norm.weight
F32
F32
[2048]
-
blk.26.ffn_up.weight
Q8_0
Q8_0
[2048, 16384]
-
blk.26.inp_gate.weight
Q8_0
Q8_0
[2048, 256]
-
blk.26.laurel_l.weight
Q8_0
Q8_0
[2048, 64]
-
blk.26.laurel_post_norm.weight
F32
F32
[2048]
-
blk.26.laurel_r.weight
Q8_0
Q8_0
[64, 2048]
-
blk.26.post_attention_norm.weight
F32
F32
[2048]
-
blk.26.post_ffw_norm.weight
F32
F32
[2048]
-
blk.26.post_norm.weight
F32
F32
[2048]
-
Q8_0
[256, 2048]
-
blk.27.altup_correct_coef.weight
F32
F32
[4, 4]
-
blk.27.altup_correct_scale.weight
F32
F32
[2048]
-
blk.27.altup_predict_coef.weight
F32
F32
[4, 16]
-
blk.27.altup_router.weight
Q8_0
Q8_0
[2048, 4]
-
blk.27.altup_router_norm.weight
F32
F32
[2048]
-
blk.27.attn_k.weight
Q8_0
Q8_0
[2048, 512]
-
blk.27.attn_k_norm.weight
F32
F32
[256]
-
blk.27.attn_norm.weight
F32
F32
[2048]
-
blk.27.attn_output.weight
Q8_0
Q8_0
[2048, 2048]
-
blk.27.attn_q.weight
Q8_0
Q8_0
[2048, 2048]
-
blk.27.attn_q_norm.weight
F32
F32
[256]
-
blk.27.attn_v.weight
Q8_0
Q8_0
[2048, 512]
-
blk.27.ffn_down.weight
Q8_0
Q8_0
[16384, 2048]
-
blk.27.ffn_gate.weight
Q8_0
Q8_0
[2048, 16384]
-
blk.27.ffn_norm.weight
F32
F32
[2048]
-
blk.27.ffn_up.weight
Q8_0
Q8_0
[2048, 16384]
-
blk.27.inp_gate.weight
Q8_0
Q8_0
[2048, 256]
-
blk.27.laurel_l.weight
Q8_0
Q8_0
[2048, 64]
-
blk.27.laurel_post_norm.weight
F32
F32
[2048]
-
blk.27.laurel_r.weight
Q8_0
Q8_0
[64, 2048]
-
blk.27.post_attention_norm.weight
F32
F32
[2048]
-
blk.27.post_ffw_norm.weight
F32
F32
[2048]
-
blk.27.post_norm.weight
F32
F32
[2048]
-
Q8_0
[256, 2048]
-
blk.28.altup_correct_coef.weight
F32
F32
[4, 4]
-
blk.28.altup_correct_scale.weight
F32
F32
[2048]
-
blk.28.altup_predict_coef.weight
F32
F32
[4, 16]
-
blk.28.altup_router.weight
Q8_0
Q8_0
[2048, 4]
-
blk.28.altup_router_norm.weight
F32
F32
[2048]
-
blk.28.attn_k.weight
Q8_0
Q8_0
[2048, 512]
-
blk.28.attn_k_norm.weight
F32
F32
[256]
-
blk.28.attn_norm.weight
F32
F32
[2048]
-
blk.28.attn_output.weight
Q8_0
Q8_0
[2048, 2048]
-
blk.28.attn_q.weight
Q8_0
Q8_0
[2048, 2048]
-
blk.28.attn_q_norm.weight
F32
F32
[256]
-
blk.28.attn_v.weight
Q8_0
Q8_0
[2048, 512]
-
blk.28.ffn_down.weight
Q8_0
Q8_0
[16384, 2048]
-
blk.28.ffn_gate.weight
Q8_0
Q8_0
[2048, 16384]
-
blk.28.ffn_norm.weight
F32
F32
[2048]
-
blk.28.ffn_up.weight
Q8_0
Q8_0
[2048, 16384]
-
blk.28.inp_gate.weight
Q8_0
Q8_0
[2048, 256]
-
blk.28.laurel_l.weight
Q8_0
Q8_0
[2048, 64]
-
blk.28.laurel_post_norm.weight
F32
F32
[2048]
-
blk.28.laurel_r.weight
Q8_0
Q8_0
[64, 2048]
-
blk.28.post_attention_norm.weight
F32
F32
[2048]
-
blk.28.post_ffw_norm.weight
F32
F32
[2048]
-
blk.28.post_norm.weight
F32
F32
[2048]
-
Q8_0
[256, 2048]
-
blk.29.altup_correct_coef.weight
F32
F32
[4, 4]
-
blk.29.altup_correct_scale.weight
F32
F32
[2048]
-
blk.29.altup_predict_coef.weight
F32
F32
[4, 16]
-
blk.29.altup_router.weight
Q8_0
Q8_0
[2048, 4]
-
blk.29.altup_router_norm.weight
F32
F32
[2048]
-
blk.29.attn_k.weight
Q8_0
Q8_0
[2048, 512]
-
blk.29.attn_k_norm.weight
F32
F32
[256]
-
blk.29.attn_norm.weight
F32
F32
[2048]
-
blk.29.attn_output.weight
Q8_0
Q8_0
[2048, 2048]
-
blk.29.attn_q.weight
Q8_0
Q8_0
[2048, 2048]
-
blk.29.attn_q_norm.weight
F32
F32
[256]
-
blk.29.attn_v.weight
Q8_0
Q8_0
[2048, 512]
-
blk.29.ffn_down.weight
Q8_0
Q8_0
[16384, 2048]
-
blk.29.ffn_gate.weight
Q8_0
Q8_0
[2048, 16384]
-
blk.29.ffn_norm.weight
F32
F32
[2048]
-
blk.29.ffn_up.weight
Q8_0
Q8_0
[2048, 16384]
-
blk.29.inp_gate.weight
Q8_0
Q8_0
[2048, 256]
-
blk.29.laurel_l.weight
Q8_0
Q8_0
[2048, 64]
-
blk.29.laurel_post_norm.weight
F32
F32
[2048]
-
blk.29.laurel_r.weight
Q8_0
Q8_0
[64, 2048]
-
blk.29.post_attention_norm.weight
F32
F32
[2048]
-
blk.29.post_ffw_norm.weight
F32
F32
[2048]
-
blk.29.post_norm.weight
F32
F32
[2048]
-
Q8_0
[256, 2048]
-
blk.30.altup_correct_coef.weight
F32
F32
[4, 4]
-
blk.30.altup_correct_scale.weight
F32
F32
[2048]
-
blk.30.altup_predict_coef.weight
F32
F32
[4, 16]
-
blk.30.altup_router.weight
Q8_0
Q8_0
[2048, 4]
-
blk.30.altup_router_norm.weight
F32
F32
[2048]
-
blk.30.attn_k.weight
Q8_0
Q8_0
[2048, 512]
-
blk.30.attn_k_norm.weight
F32
F32
[256]
-
blk.30.attn_norm.weight
F32
F32
[2048]
-
blk.30.attn_output.weight
Q8_0
Q8_0
[2048, 2048]
-
blk.30.attn_q.weight
Q8_0
Q8_0
[2048, 2048]
-
blk.30.attn_q_norm.weight
F32
F32
[256]
-
blk.30.attn_v.weight
Q8_0
Q8_0
[2048, 512]
-
blk.30.ffn_down.weight
Q8_0
Q8_0
[16384, 2048]
-
blk.30.ffn_gate.weight
Q8_0
Q8_0
[2048, 16384]
-
blk.30.ffn_norm.weight
F32
F32
[2048]
-
blk.30.ffn_up.weight
Q8_0
Q8_0
[2048, 16384]
-
blk.30.inp_gate.weight
Q8_0
Q8_0
[2048, 256]
-
blk.30.laurel_l.weight
Q8_0
Q8_0
[2048, 64]
-
blk.30.laurel_post_norm.weight
F32
F32
[2048]
-
blk.30.laurel_r.weight
Q8_0
Q8_0
[64, 2048]
-
blk.30.post_attention_norm.weight
F32
F32
[2048]
-
blk.30.post_ffw_norm.weight
F32
F32
[2048]
-
blk.30.post_norm.weight
F32
F32
[2048]
-
Q8_0
[256, 2048]
-
blk.31.altup_correct_coef.weight
F32
F32
[4, 4]
-
blk.31.altup_correct_scale.weight
F32
F32
[2048]
-
blk.31.altup_predict_coef.weight
F32
F32
[4, 16]
-
blk.31.altup_router.weight
Q8_0
Q8_0
[2048, 4]
-
blk.31.altup_router_norm.weight
F32
F32
[2048]
-
blk.31.attn_k.weight
Q8_0
Q8_0
[2048, 512]
-
blk.31.attn_k_norm.weight
F32
F32
[256]
-
blk.31.attn_norm.weight
F32
F32
[2048]
-
blk.31.attn_output.weight
Q8_0
Q8_0
[2048, 2048]
-
blk.31.attn_q.weight
Q8_0
Q8_0
[2048, 2048]
-
blk.31.attn_q_norm.weight
F32
F32
[256]
-
blk.31.attn_v.weight
Q8_0
Q8_0
[2048, 512]
-
blk.31.ffn_down.weight
Q8_0
Q8_0
[16384, 2048]
-
blk.31.ffn_gate.weight
Q8_0
Q8_0
[2048, 16384]
-
blk.31.ffn_norm.weight
F32
F32
[2048]
-
blk.31.ffn_up.weight
Q8_0
Q8_0
[2048, 16384]
-
blk.31.inp_gate.weight
Q8_0
Q8_0
[2048, 256]
-
blk.31.laurel_l.weight
Q8_0
Q8_0
[2048, 64]
-
blk.31.laurel_post_norm.weight
F32
F32
[2048]
-
blk.31.laurel_r.weight
Q8_0
Q8_0
[64, 2048]
-
blk.31.post_attention_norm.weight
F32
F32
[2048]
-
blk.31.post_ffw_norm.weight
F32
F32
[2048]
-
blk.31.post_norm.weight
F32
F32
[2048]
-
Q8_0
[256, 2048]
-
blk.32.altup_correct_coef.weight
F32
F32
[4, 4]
-
blk.32.altup_correct_scale.weight
F32
F32
[2048]
-
blk.32.altup_predict_coef.weight
F32
F32
[4, 16]
-
blk.32.altup_router.weight
Q8_0
Q8_0
[2048, 4]
-
blk.32.altup_router_norm.weight
F32
F32
[2048]
-
blk.32.attn_k.weight
Q8_0
Q8_0
[2048, 512]
-
blk.32.attn_k_norm.weight
F32
F32
[256]
-
blk.32.attn_norm.weight
F32
F32
[2048]
-
blk.32.attn_output.weight
Q8_0
Q8_0
[2048, 2048]
-
blk.32.attn_q.weight
Q8_0
Q8_0
[2048, 2048]
-
blk.32.attn_q_norm.weight
F32
F32
[256]
-
blk.32.attn_v.weight
Q8_0
Q8_0
[2048, 512]
-
blk.32.ffn_down.weight
Q8_0
Q8_0
[16384, 2048]
-
blk.32.ffn_gate.weight
Q8_0
Q8_0
[2048, 16384]
-
blk.32.ffn_norm.weight
F32
F32
[2048]
-
blk.32.ffn_up.weight
Q8_0
Q8_0
[2048, 16384]
-
blk.32.inp_gate.weight
Q8_0
Q8_0
[2048, 256]
-
blk.32.laurel_l.weight
Q8_0
Q8_0
[2048, 64]
-
blk.32.laurel_post_norm.weight
F32
F32
[2048]
-
blk.32.laurel_r.weight
Q8_0
Q8_0
[64, 2048]
-
blk.32.post_attention_norm.weight
F32
F32
[2048]
-
blk.32.post_ffw_norm.weight
F32
F32
[2048]
-
blk.32.post_norm.weight
F32
F32
[2048]
-
Q8_0
[256, 2048]
-
blk.33.altup_correct_coef.weight
F32
F32
[4, 4]
-
blk.33.altup_correct_scale.weight
F32
F32
[2048]
-
blk.33.altup_predict_coef.weight
F32
F32
[4, 16]
-
blk.33.altup_router.weight
Q8_0
Q8_0
[2048, 4]
-
blk.33.altup_router_norm.weight
F32
F32
[2048]
-
blk.33.attn_k.weight
Q8_0
Q8_0
[2048, 512]
-
blk.33.attn_k_norm.weight
F32
F32
[256]
-
blk.33.attn_norm.weight
F32
F32
[2048]
-
blk.33.attn_output.weight
Q8_0
Q8_0
[2048, 2048]
-
blk.33.attn_q.weight
Q8_0
Q8_0
[2048, 2048]
-
blk.33.attn_q_norm.weight
F32
F32
[256]
-
blk.33.attn_v.weight
Q8_0
Q8_0
[2048, 512]
-
blk.33.ffn_down.weight
Q8_0
Q8_0
[16384, 2048]
-
blk.33.ffn_gate.weight
Q8_0
Q8_0
[2048, 16384]
-
blk.33.ffn_norm.weight
F32
F32
[2048]
-
blk.33.ffn_up.weight
Q8_0
Q8_0
[2048, 16384]
-
blk.33.inp_gate.weight
Q8_0
Q8_0
[2048, 256]
-
blk.33.laurel_l.weight
Q8_0
Q8_0
[2048, 64]
-
blk.33.laurel_post_norm.weight
F32
F32
[2048]
-
blk.33.laurel_r.weight
Q8_0
Q8_0
[64, 2048]
-
blk.33.post_attention_norm.weight
F32
F32
[2048]
-
blk.33.post_ffw_norm.weight
F32
F32
[2048]
-
blk.33.post_norm.weight
F32
F32
[2048]
-
Q8_0
[256, 2048]
-
blk.34.altup_correct_coef.weight
F32
F32
[4, 4]
-
blk.34.altup_correct_scale.weight
F32
F32
[2048]
-
blk.34.altup_predict_coef.weight
F32
F32
[4, 16]
-
blk.34.altup_router.weight
Q8_0
Q8_0
[2048, 4]
-
blk.34.altup_router_norm.weight
F32
F32
[2048]
-
blk.34.attn_k.weight
Q8_0
Q8_0
[2048, 512]
-
blk.34.attn_k_norm.weight
F32
F32
[256]
-
blk.34.attn_norm.weight
F32
F32
[2048]
-
blk.34.attn_output.weight
Q8_0
Q8_0
[2048, 2048]
-
blk.34.attn_q.weight
Q8_0
Q8_0
[2048, 2048]
-
blk.34.attn_q_norm.weight
F32
F32
[256]
-
blk.34.attn_v.weight
Q8_0
Q8_0
[2048, 512]
-
blk.34.ffn_down.weight
Q8_0
Q8_0
[16384, 2048]
-
blk.34.ffn_gate.weight
Q8_0
Q8_0
[2048, 16384]
-
blk.34.ffn_norm.weight
F32
F32
[2048]
-
blk.34.ffn_up.weight
Q8_0
Q8_0
[2048, 16384]
-
blk.34.inp_gate.weight
Q8_0
Q8_0
[2048, 256]
-
blk.34.laurel_l.weight
Q8_0
Q8_0
[2048, 64]
-
blk.34.laurel_post_norm.weight
F32
F32
[2048]
-
blk.34.laurel_r.weight
Q8_0
Q8_0
[64, 2048]
-
blk.34.post_attention_norm.weight
F32
F32
[2048]
-
blk.34.post_ffw_norm.weight
F32
F32
[2048]
-
blk.34.post_norm.weight
F32
F32
[2048]
-
Q8_0
[256, 2048]
-
per_layer_model_proj.weight
Q8_0
Q8_0
[2048, 8960]
-
per_layer_proj_norm.weight
F32
F32
[256]
-
per_layer_token_embd.weight
Q8_0
Q8_0
[8960, 262144]
-
F32
[2048]