Models
Docs
Pricing
Sign in
Download
Models
Download
Docs
Pricing
Sign in
frob
/
deepseek-v4-flash
:284b-a13b-mxfp4
149
Downloads
Updated
1 week ago
Cancel
tools
thinking
deepseek-v4-flash:284b-a13b-mxfp4
...
/
model
1b716cb91a3d · 156GB
Metadata
general.architecture
deepseek4
deepseek4
general.file_type
(!unknown_file_type 38!)
(!unknown_file_type 38!)
deepseek4.attention.compress_ratios
[0, 0, 4, 128, 4, ...]
[0, 0, 4, 128, 4, ...]
deepseek4.attention.compress_rope_freq_base
160000
160000
deepseek4.attention.head_count
64
64
deepseek4.attention.head_count_kv
1
1
deepseek4.attention.indexer.head_count
64
64
deepseek4.attention.indexer.key_length
128
128
deepseek4.attention.indexer.top_k
512
512
deepseek4.attention.key_length
512
512
deepseek4.attention.layer_norm_rms_epsilon
1e-06
1e-06
deepseek4.attention.output_group_count
8
8
deepseek4.attention.output_lora_rank
1024
1024
deepseek4.attention.q_lora_rank
1024
1024
deepseek4.attention.sliding_window
128
128
deepseek4.attention.value_length
512
512
deepseek4.block_count
43
43
deepseek4.context_length
1048576
1048576
deepseek4.embedding_length
4096
4096
deepseek4.expert_count
256
256
deepseek4.expert_feed_forward_length
2048
2048
deepseek4.expert_gating_func
4
4
deepseek4.expert_shared_count
1
1
deepseek4.expert_used_count
6
6
deepseek4.expert_weights_norm
true
true
deepseek4.expert_weights_scale
1.5
1.5
deepseek4.hash_layer_count
3
3
deepseek4.hyper_connection.count
4
4
deepseek4.hyper_connection.epsilon
1e-06
1e-06
deepseek4.hyper_connection.sinkhorn_iterations
20
20
deepseek4.rope.dimension_count
64
64
deepseek4.rope.freq_base
10000
10000
deepseek4.rope.scaling.factor
16
16
deepseek4.rope.scaling.original_context_length
65536
65536
deepseek4.rope.scaling.type
yarn
yarn
deepseek4.rope.scaling.yarn_beta_fast
32
32
deepseek4.rope.scaling.yarn_beta_slow
1
1
deepseek4.swiglu_clamp_exp
[10, 10, 10, 10, 10, ...]
[10, 10, 10, 10, 10, ...]
deepseek4.swiglu_clamp_shexp
[10, 10, 10, 10, 10, ...]
[10, 10, 10, 10, 10, ...]
tokenizer.ggml.add_bos_token
false
false
tokenizer.ggml.add_eos_token
false
false
tokenizer.ggml.bos_token_id
0
0
tokenizer.ggml.eos_token_id
1
1
tokenizer.ggml.merges
[Ġ t, Ġ a, i n, Ġ Ġ, h e, ...]
[Ġ t, Ġ a, i n, Ġ Ġ, h e, ...]
tokenizer.ggml.model
gpt2
gpt2
tokenizer.ggml.padding_token_id
1
1
tokenizer.ggml.pre
joyai-llm
joyai-llm
tokenizer.ggml.token_type
[3, 3, 3, 1, 1, ...]
[3, 3, 3, 1, 1, ...]
tokenizer.ggml.tokens
[<|begin▁of▁sentence|>, <|end▁of▁sentence|>, <|▁pad▁|>, !, ", ...]
[<|begin▁of▁sentence|>, <|end▁of▁sentence|>, <|▁pad▁|>, !, ", ...]
Tensor
Name
Type
Shape
token_embd.weight
BF16
BF16
[4096, 129280]
blk.0
blk.0.attn_kv.weight
Q8_0
Q8_0
[4096, 512]
blk.0.attn_kv_a_norm.weight
F32
F32
[512]
blk.0.attn_norm.weight
F32
F32
[4096]
blk.0.attn_output_a.weight
Q8_0
Q8_0
[4096, 8192]
blk.0.attn_output_b.weight
Q8_0
Q8_0
[8192, 4096]
blk.0.attn_q_a.weight
Q8_0
Q8_0
[4096, 1024]
blk.0.attn_q_a_norm.weight
F32
F32
[1024]
blk.0.attn_q_b.weight
Q8_0
Q8_0
[1024, 32768]
blk.0.attn_sinks.weight
F32
F32
[64]
blk.0.ffn_down_exps.weight
MXFP4
MXFP4
[2048, 4096, 256]
blk.0.ffn_down_shexp.weight
Q8_0
Q8_0
[2048, 4096]
blk.0.ffn_gate_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.0.ffn_gate_inp.weight
BF16
BF16
[4096, 256]
blk.0.ffn_gate_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.0.ffn_gate_tid2eid.weight
I32
I32
[6, 129280]
blk.0.ffn_norm.weight
F32
F32
[4096]
blk.0.ffn_up_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.0.ffn_up_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.0.hc_attn_base.weight
F32
F32
[24]
blk.0.hc_attn_fn.weight
F32
F32
[16384, 24]
blk.0.hc_attn_scale.weight
F32
F32
[3]
blk.0.hc_ffn_base.weight
F32
F32
[24]
blk.0.hc_ffn_fn.weight
F32
F32
[16384, 24]
blk.0.hc_ffn_scale.weight
F32
F32
[3]
blk.1
blk.1.attn_kv.weight
Q8_0
Q8_0
[4096, 512]
blk.1.attn_kv_a_norm.weight
F32
F32
[512]
blk.1.attn_norm.weight
F32
F32
[4096]
blk.1.attn_output_a.weight
Q8_0
Q8_0
[4096, 8192]
blk.1.attn_output_b.weight
Q8_0
Q8_0
[8192, 4096]
blk.1.attn_q_a.weight
Q8_0
Q8_0
[4096, 1024]
blk.1.attn_q_a_norm.weight
F32
F32
[1024]
blk.1.attn_q_b.weight
Q8_0
Q8_0
[1024, 32768]
blk.1.attn_sinks.weight
F32
F32
[64]
blk.1.ffn_down_exps.weight
MXFP4
MXFP4
[2048, 4096, 256]
blk.1.ffn_down_shexp.weight
Q8_0
Q8_0
[2048, 4096]
blk.1.ffn_gate_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.1.ffn_gate_inp.weight
BF16
BF16
[4096, 256]
blk.1.ffn_gate_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.1.ffn_gate_tid2eid.weight
I32
I32
[6, 129280]
blk.1.ffn_norm.weight
F32
F32
[4096]
blk.1.ffn_up_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.1.ffn_up_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.1.hc_attn_base.weight
F32
F32
[24]
blk.1.hc_attn_fn.weight
F32
F32
[16384, 24]
blk.1.hc_attn_scale.weight
F32
F32
[3]
blk.1.hc_ffn_base.weight
F32
F32
[24]
blk.1.hc_ffn_fn.weight
F32
F32
[16384, 24]
blk.1.hc_ffn_scale.weight
F32
F32
[3]
blk.2
blk.2.attn_compressor_ape.weight
F32
F32
[1024, 4]
blk.2.attn_compressor_gate.weight
BF16
BF16
[4096, 1024]
blk.2.attn_compressor_kv.weight
BF16
BF16
[4096, 1024]
blk.2.attn_compressor_norm.weight
F32
F32
[512]
blk.2.attn_kv.weight
Q8_0
Q8_0
[4096, 512]
blk.2.attn_kv_a_norm.weight
F32
F32
[512]
blk.2.attn_norm.weight
F32
F32
[4096]
blk.2.attn_output_a.weight
Q8_0
Q8_0
[4096, 8192]
blk.2.attn_output_b.weight
Q8_0
Q8_0
[8192, 4096]
blk.2.attn_q_a.weight
Q8_0
Q8_0
[4096, 1024]
blk.2.attn_q_a_norm.weight
F32
F32
[1024]
blk.2.attn_q_b.weight
Q8_0
Q8_0
[1024, 32768]
blk.2.attn_sinks.weight
F32
F32
[64]
blk.2.ffn_down_exps.weight
MXFP4
MXFP4
[2048, 4096, 256]
blk.2.ffn_down_shexp.weight
Q8_0
Q8_0
[2048, 4096]
blk.2.ffn_gate_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.2.ffn_gate_inp.weight
BF16
BF16
[4096, 256]
blk.2.ffn_gate_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.2.ffn_gate_tid2eid.weight
I32
I32
[6, 129280]
blk.2.ffn_norm.weight
F32
F32
[4096]
blk.2.ffn_up_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.2.ffn_up_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.2.hc_attn_base.weight
F32
F32
[24]
blk.2.hc_attn_fn.weight
F32
F32
[16384, 24]
blk.2.hc_attn_scale.weight
F32
F32
[3]
blk.2.hc_ffn_base.weight
F32
F32
[24]
blk.2.hc_ffn_fn.weight
F32
F32
[16384, 24]
blk.2.hc_ffn_scale.weight
F32
F32
[3]
blk.2.indexer.attn_q_b.weight
Q8_0
Q8_0
[1024, 8192]
blk.2.indexer.proj.weight
BF16
BF16
[4096, 64]
blk.2.indexer_compressor_ape.weight
F32
F32
[256, 4]
blk.2.indexer_compressor_gate.weight
BF16
BF16
[4096, 256]
blk.2.indexer_compressor_kv.weight
BF16
BF16
[4096, 256]
blk.2.indexer_compressor_norm.weight
F32
F32
[128]
blk.3
blk.3.attn_compressor_ape.weight
F32
F32
[512, 128]
blk.3.attn_compressor_gate.weight
BF16
BF16
[4096, 512]
blk.3.attn_compressor_kv.weight
BF16
BF16
[4096, 512]
blk.3.attn_compressor_norm.weight
F32
F32
[512]
blk.3.attn_kv.weight
Q8_0
Q8_0
[4096, 512]
blk.3.attn_kv_a_norm.weight
F32
F32
[512]
blk.3.attn_norm.weight
F32
F32
[4096]
blk.3.attn_output_a.weight
Q8_0
Q8_0
[4096, 8192]
blk.3.attn_output_b.weight
Q8_0
Q8_0
[8192, 4096]
blk.3.attn_q_a.weight
Q8_0
Q8_0
[4096, 1024]
blk.3.attn_q_a_norm.weight
F32
F32
[1024]
blk.3.attn_q_b.weight
Q8_0
Q8_0
[1024, 32768]
blk.3.attn_sinks.weight
F32
F32
[64]
blk.3.exp_probs_b.bias
F32
F32
[256]
blk.3.ffn_down_exps.weight
MXFP4
MXFP4
[2048, 4096, 256]
blk.3.ffn_down_shexp.weight
Q8_0
Q8_0
[2048, 4096]
blk.3.ffn_gate_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.3.ffn_gate_inp.weight
BF16
BF16
[4096, 256]
blk.3.ffn_gate_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.3.ffn_norm.weight
F32
F32
[4096]
blk.3.ffn_up_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.3.ffn_up_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.3.hc_attn_base.weight
F32
F32
[24]
blk.3.hc_attn_fn.weight
F32
F32
[16384, 24]
blk.3.hc_attn_scale.weight
F32
F32
[3]
blk.3.hc_ffn_base.weight
F32
F32
[24]
blk.3.hc_ffn_fn.weight
F32
F32
[16384, 24]
blk.3.hc_ffn_scale.weight
F32
F32
[3]
blk.4
blk.4.attn_compressor_ape.weight
F32
F32
[1024, 4]
blk.4.attn_compressor_gate.weight
BF16
BF16
[4096, 1024]
blk.4.attn_compressor_kv.weight
BF16
BF16
[4096, 1024]
blk.4.attn_compressor_norm.weight
F32
F32
[512]
blk.4.attn_kv.weight
Q8_0
Q8_0
[4096, 512]
blk.4.attn_kv_a_norm.weight
F32
F32
[512]
blk.4.attn_norm.weight
F32
F32
[4096]
blk.4.attn_output_a.weight
Q8_0
Q8_0
[4096, 8192]
blk.4.attn_output_b.weight
Q8_0
Q8_0
[8192, 4096]
blk.4.attn_q_a.weight
Q8_0
Q8_0
[4096, 1024]
blk.4.attn_q_a_norm.weight
F32
F32
[1024]
blk.4.attn_q_b.weight
Q8_0
Q8_0
[1024, 32768]
blk.4.attn_sinks.weight
F32
F32
[64]
blk.4.exp_probs_b.bias
F32
F32
[256]
blk.4.ffn_down_exps.weight
MXFP4
MXFP4
[2048, 4096, 256]
blk.4.ffn_down_shexp.weight
Q8_0
Q8_0
[2048, 4096]
blk.4.ffn_gate_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.4.ffn_gate_inp.weight
BF16
BF16
[4096, 256]
blk.4.ffn_gate_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.4.ffn_norm.weight
F32
F32
[4096]
blk.4.ffn_up_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.4.ffn_up_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.4.hc_attn_base.weight
F32
F32
[24]
blk.4.hc_attn_fn.weight
F32
F32
[16384, 24]
blk.4.hc_attn_scale.weight
F32
F32
[3]
blk.4.hc_ffn_base.weight
F32
F32
[24]
blk.4.hc_ffn_fn.weight
F32
F32
[16384, 24]
blk.4.hc_ffn_scale.weight
F32
F32
[3]
blk.4.indexer.attn_q_b.weight
Q8_0
Q8_0
[1024, 8192]
blk.4.indexer.proj.weight
BF16
BF16
[4096, 64]
blk.4.indexer_compressor_ape.weight
F32
F32
[256, 4]
blk.4.indexer_compressor_gate.weight
BF16
BF16
[4096, 256]
blk.4.indexer_compressor_kv.weight
BF16
BF16
[4096, 256]
blk.4.indexer_compressor_norm.weight
F32
F32
[128]
blk.5
blk.5.attn_compressor_ape.weight
F32
F32
[512, 128]
blk.5.attn_compressor_gate.weight
BF16
BF16
[4096, 512]
blk.5.attn_compressor_kv.weight
BF16
BF16
[4096, 512]
blk.5.attn_compressor_norm.weight
F32
F32
[512]
blk.5.attn_kv.weight
Q8_0
Q8_0
[4096, 512]
blk.5.attn_kv_a_norm.weight
F32
F32
[512]
blk.5.attn_norm.weight
F32
F32
[4096]
blk.5.attn_output_a.weight
Q8_0
Q8_0
[4096, 8192]
blk.5.attn_output_b.weight
Q8_0
Q8_0
[8192, 4096]
blk.5.attn_q_a.weight
Q8_0
Q8_0
[4096, 1024]
blk.5.attn_q_a_norm.weight
F32
F32
[1024]
blk.5.attn_q_b.weight
Q8_0
Q8_0
[1024, 32768]
blk.5.attn_sinks.weight
F32
F32
[64]
blk.5.exp_probs_b.bias
F32
F32
[256]
blk.5.ffn_down_exps.weight
MXFP4
MXFP4
[2048, 4096, 256]
blk.5.ffn_down_shexp.weight
Q8_0
Q8_0
[2048, 4096]
blk.5.ffn_gate_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.5.ffn_gate_inp.weight
BF16
BF16
[4096, 256]
blk.5.ffn_gate_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.5.ffn_norm.weight
F32
F32
[4096]
blk.5.ffn_up_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.5.ffn_up_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.5.hc_attn_base.weight
F32
F32
[24]
blk.5.hc_attn_fn.weight
F32
F32
[16384, 24]
blk.5.hc_attn_scale.weight
F32
F32
[3]
blk.5.hc_ffn_base.weight
F32
F32
[24]
blk.5.hc_ffn_fn.weight
F32
F32
[16384, 24]
blk.5.hc_ffn_scale.weight
F32
F32
[3]
blk.6
blk.6.attn_compressor_ape.weight
F32
F32
[1024, 4]
blk.6.attn_compressor_gate.weight
BF16
BF16
[4096, 1024]
blk.6.attn_compressor_kv.weight
BF16
BF16
[4096, 1024]
blk.6.attn_compressor_norm.weight
F32
F32
[512]
blk.6.attn_kv.weight
Q8_0
Q8_0
[4096, 512]
blk.6.attn_kv_a_norm.weight
F32
F32
[512]
blk.6.attn_norm.weight
F32
F32
[4096]
blk.6.attn_output_a.weight
Q8_0
Q8_0
[4096, 8192]
blk.6.attn_output_b.weight
Q8_0
Q8_0
[8192, 4096]
blk.6.attn_q_a.weight
Q8_0
Q8_0
[4096, 1024]
blk.6.attn_q_a_norm.weight
F32
F32
[1024]
blk.6.attn_q_b.weight
Q8_0
Q8_0
[1024, 32768]
blk.6.attn_sinks.weight
F32
F32
[64]
blk.6.exp_probs_b.bias
F32
F32
[256]
blk.6.ffn_down_exps.weight
MXFP4
MXFP4
[2048, 4096, 256]
blk.6.ffn_down_shexp.weight
Q8_0
Q8_0
[2048, 4096]
blk.6.ffn_gate_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.6.ffn_gate_inp.weight
BF16
BF16
[4096, 256]
blk.6.ffn_gate_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.6.ffn_norm.weight
F32
F32
[4096]
blk.6.ffn_up_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.6.ffn_up_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.6.hc_attn_base.weight
F32
F32
[24]
blk.6.hc_attn_fn.weight
F32
F32
[16384, 24]
blk.6.hc_attn_scale.weight
F32
F32
[3]
blk.6.hc_ffn_base.weight
F32
F32
[24]
blk.6.hc_ffn_fn.weight
F32
F32
[16384, 24]
blk.6.hc_ffn_scale.weight
F32
F32
[3]
blk.6.indexer.attn_q_b.weight
Q8_0
Q8_0
[1024, 8192]
blk.6.indexer.proj.weight
BF16
BF16
[4096, 64]
blk.6.indexer_compressor_ape.weight
F32
F32
[256, 4]
blk.6.indexer_compressor_gate.weight
BF16
BF16
[4096, 256]
blk.6.indexer_compressor_kv.weight
BF16
BF16
[4096, 256]
blk.6.indexer_compressor_norm.weight
F32
F32
[128]
blk.7
blk.7.attn_compressor_ape.weight
F32
F32
[512, 128]
blk.7.attn_compressor_gate.weight
BF16
BF16
[4096, 512]
blk.7.attn_compressor_kv.weight
BF16
BF16
[4096, 512]
blk.7.attn_compressor_norm.weight
F32
F32
[512]
blk.7.attn_kv.weight
Q8_0
Q8_0
[4096, 512]
blk.7.attn_kv_a_norm.weight
F32
F32
[512]
blk.7.attn_norm.weight
F32
F32
[4096]
blk.7.attn_output_a.weight
Q8_0
Q8_0
[4096, 8192]
blk.7.attn_output_b.weight
Q8_0
Q8_0
[8192, 4096]
blk.7.attn_q_a.weight
Q8_0
Q8_0
[4096, 1024]
blk.7.attn_q_a_norm.weight
F32
F32
[1024]
blk.7.attn_q_b.weight
Q8_0
Q8_0
[1024, 32768]
blk.7.attn_sinks.weight
F32
F32
[64]
blk.7.exp_probs_b.bias
F32
F32
[256]
blk.7.ffn_down_exps.weight
MXFP4
MXFP4
[2048, 4096, 256]
blk.7.ffn_down_shexp.weight
Q8_0
Q8_0
[2048, 4096]
blk.7.ffn_gate_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.7.ffn_gate_inp.weight
BF16
BF16
[4096, 256]
blk.7.ffn_gate_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.7.ffn_norm.weight
F32
F32
[4096]
blk.7.ffn_up_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.7.ffn_up_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.7.hc_attn_base.weight
F32
F32
[24]
blk.7.hc_attn_fn.weight
F32
F32
[16384, 24]
blk.7.hc_attn_scale.weight
F32
F32
[3]
blk.7.hc_ffn_base.weight
F32
F32
[24]
blk.7.hc_ffn_fn.weight
F32
F32
[16384, 24]
blk.7.hc_ffn_scale.weight
F32
F32
[3]
blk.8
blk.8.attn_compressor_ape.weight
F32
F32
[1024, 4]
blk.8.attn_compressor_gate.weight
BF16
BF16
[4096, 1024]
blk.8.attn_compressor_kv.weight
BF16
BF16
[4096, 1024]
blk.8.attn_compressor_norm.weight
F32
F32
[512]
blk.8.attn_kv.weight
Q8_0
Q8_0
[4096, 512]
blk.8.attn_kv_a_norm.weight
F32
F32
[512]
blk.8.attn_norm.weight
F32
F32
[4096]
blk.8.attn_output_a.weight
Q8_0
Q8_0
[4096, 8192]
blk.8.attn_output_b.weight
Q8_0
Q8_0
[8192, 4096]
blk.8.attn_q_a.weight
Q8_0
Q8_0
[4096, 1024]
blk.8.attn_q_a_norm.weight
F32
F32
[1024]
blk.8.attn_q_b.weight
Q8_0
Q8_0
[1024, 32768]
blk.8.attn_sinks.weight
F32
F32
[64]
blk.8.exp_probs_b.bias
F32
F32
[256]
blk.8.ffn_down_exps.weight
MXFP4
MXFP4
[2048, 4096, 256]
blk.8.ffn_down_shexp.weight
Q8_0
Q8_0
[2048, 4096]
blk.8.ffn_gate_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.8.ffn_gate_inp.weight
BF16
BF16
[4096, 256]
blk.8.ffn_gate_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.8.ffn_norm.weight
F32
F32
[4096]
blk.8.ffn_up_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.8.ffn_up_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.8.hc_attn_base.weight
F32
F32
[24]
blk.8.hc_attn_fn.weight
F32
F32
[16384, 24]
blk.8.hc_attn_scale.weight
F32
F32
[3]
blk.8.hc_ffn_base.weight
F32
F32
[24]
blk.8.hc_ffn_fn.weight
F32
F32
[16384, 24]
blk.8.hc_ffn_scale.weight
F32
F32
[3]
blk.8.indexer.attn_q_b.weight
Q8_0
Q8_0
[1024, 8192]
blk.8.indexer.proj.weight
BF16
BF16
[4096, 64]
blk.8.indexer_compressor_ape.weight
F32
F32
[256, 4]
blk.8.indexer_compressor_gate.weight
BF16
BF16
[4096, 256]
blk.8.indexer_compressor_kv.weight
BF16
BF16
[4096, 256]
blk.8.indexer_compressor_norm.weight
F32
F32
[128]
blk.9
blk.9.attn_compressor_ape.weight
F32
F32
[512, 128]
blk.9.attn_compressor_gate.weight
BF16
BF16
[4096, 512]
blk.9.attn_compressor_kv.weight
BF16
BF16
[4096, 512]
blk.9.attn_compressor_norm.weight
F32
F32
[512]
blk.9.attn_kv.weight
Q8_0
Q8_0
[4096, 512]
blk.9.attn_kv_a_norm.weight
F32
F32
[512]
blk.9.attn_norm.weight
F32
F32
[4096]
blk.9.attn_output_a.weight
Q8_0
Q8_0
[4096, 8192]
blk.9.attn_output_b.weight
Q8_0
Q8_0
[8192, 4096]
blk.9.attn_q_a.weight
Q8_0
Q8_0
[4096, 1024]
blk.9.attn_q_a_norm.weight
F32
F32
[1024]
blk.9.attn_q_b.weight
Q8_0
Q8_0
[1024, 32768]
blk.9.attn_sinks.weight
F32
F32
[64]
blk.9.exp_probs_b.bias
F32
F32
[256]
blk.9.ffn_down_exps.weight
MXFP4
MXFP4
[2048, 4096, 256]
blk.9.ffn_down_shexp.weight
Q8_0
Q8_0
[2048, 4096]
blk.9.ffn_gate_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.9.ffn_gate_inp.weight
BF16
BF16
[4096, 256]
blk.9.ffn_gate_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.9.ffn_norm.weight
F32
F32
[4096]
blk.9.ffn_up_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.9.ffn_up_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.9.hc_attn_base.weight
F32
F32
[24]
blk.9.hc_attn_fn.weight
F32
F32
[16384, 24]
blk.9.hc_attn_scale.weight
F32
F32
[3]
blk.9.hc_ffn_base.weight
F32
F32
[24]
blk.9.hc_ffn_fn.weight
F32
F32
[16384, 24]
blk.9.hc_ffn_scale.weight
F32
F32
[3]
blk.10
blk.10.attn_compressor_ape.weight
F32
F32
[1024, 4]
blk.10.attn_compressor_gate.weight
BF16
BF16
[4096, 1024]
blk.10.attn_compressor_kv.weight
BF16
BF16
[4096, 1024]
blk.10.attn_compressor_norm.weight
F32
F32
[512]
blk.10.attn_kv.weight
Q8_0
Q8_0
[4096, 512]
blk.10.attn_kv_a_norm.weight
F32
F32
[512]
blk.10.attn_norm.weight
F32
F32
[4096]
blk.10.attn_output_a.weight
Q8_0
Q8_0
[4096, 8192]
blk.10.attn_output_b.weight
Q8_0
Q8_0
[8192, 4096]
blk.10.attn_q_a.weight
Q8_0
Q8_0
[4096, 1024]
blk.10.attn_q_a_norm.weight
F32
F32
[1024]
blk.10.attn_q_b.weight
Q8_0
Q8_0
[1024, 32768]
blk.10.attn_sinks.weight
F32
F32
[64]
blk.10.exp_probs_b.bias
F32
F32
[256]
blk.10.ffn_down_exps.weight
MXFP4
MXFP4
[2048, 4096, 256]
blk.10.ffn_down_shexp.weight
Q8_0
Q8_0
[2048, 4096]
blk.10.ffn_gate_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.10.ffn_gate_inp.weight
BF16
BF16
[4096, 256]
blk.10.ffn_gate_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.10.ffn_norm.weight
F32
F32
[4096]
blk.10.ffn_up_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.10.ffn_up_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.10.hc_attn_base.weight
F32
F32
[24]
blk.10.hc_attn_fn.weight
F32
F32
[16384, 24]
blk.10.hc_attn_scale.weight
F32
F32
[3]
blk.10.hc_ffn_base.weight
F32
F32
[24]
blk.10.hc_ffn_fn.weight
F32
F32
[16384, 24]
blk.10.hc_ffn_scale.weight
F32
F32
[3]
blk.10.indexer.attn_q_b.weight
Q8_0
Q8_0
[1024, 8192]
blk.10.indexer.proj.weight
BF16
BF16
[4096, 64]
blk.10.indexer_compressor_ape.weight
F32
F32
[256, 4]
blk.10.indexer_compressor_gate.weight
BF16
BF16
[4096, 256]
blk.10.indexer_compressor_kv.weight
BF16
BF16
[4096, 256]
blk.10.indexer_compressor_norm.weight
F32
F32
[128]
blk.11
blk.11.attn_compressor_ape.weight
F32
F32
[512, 128]
blk.11.attn_compressor_gate.weight
BF16
BF16
[4096, 512]
blk.11.attn_compressor_kv.weight
BF16
BF16
[4096, 512]
blk.11.attn_compressor_norm.weight
F32
F32
[512]
blk.11.attn_kv.weight
Q8_0
Q8_0
[4096, 512]
blk.11.attn_kv_a_norm.weight
F32
F32
[512]
blk.11.attn_norm.weight
F32
F32
[4096]
blk.11.attn_output_a.weight
Q8_0
Q8_0
[4096, 8192]
blk.11.attn_output_b.weight
Q8_0
Q8_0
[8192, 4096]
blk.11.attn_q_a.weight
Q8_0
Q8_0
[4096, 1024]
blk.11.attn_q_a_norm.weight
F32
F32
[1024]
blk.11.attn_q_b.weight
Q8_0
Q8_0
[1024, 32768]
blk.11.attn_sinks.weight
F32
F32
[64]
blk.11.exp_probs_b.bias
F32
F32
[256]
blk.11.ffn_down_exps.weight
MXFP4
MXFP4
[2048, 4096, 256]
blk.11.ffn_down_shexp.weight
Q8_0
Q8_0
[2048, 4096]
blk.11.ffn_gate_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.11.ffn_gate_inp.weight
BF16
BF16
[4096, 256]
blk.11.ffn_gate_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.11.ffn_norm.weight
F32
F32
[4096]
blk.11.ffn_up_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.11.ffn_up_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.11.hc_attn_base.weight
F32
F32
[24]
blk.11.hc_attn_fn.weight
F32
F32
[16384, 24]
blk.11.hc_attn_scale.weight
F32
F32
[3]
blk.11.hc_ffn_base.weight
F32
F32
[24]
blk.11.hc_ffn_fn.weight
F32
F32
[16384, 24]
blk.11.hc_ffn_scale.weight
F32
F32
[3]
blk.12
blk.12.attn_compressor_ape.weight
F32
F32
[1024, 4]
blk.12.attn_compressor_gate.weight
BF16
BF16
[4096, 1024]
blk.12.attn_compressor_kv.weight
BF16
BF16
[4096, 1024]
blk.12.attn_compressor_norm.weight
F32
F32
[512]
blk.12.attn_kv.weight
Q8_0
Q8_0
[4096, 512]
blk.12.attn_kv_a_norm.weight
F32
F32
[512]
blk.12.attn_norm.weight
F32
F32
[4096]
blk.12.attn_output_a.weight
Q8_0
Q8_0
[4096, 8192]
blk.12.attn_output_b.weight
Q8_0
Q8_0
[8192, 4096]
blk.12.attn_q_a.weight
Q8_0
Q8_0
[4096, 1024]
blk.12.attn_q_a_norm.weight
F32
F32
[1024]
blk.12.attn_q_b.weight
Q8_0
Q8_0
[1024, 32768]
blk.12.attn_sinks.weight
F32
F32
[64]
blk.12.exp_probs_b.bias
F32
F32
[256]
blk.12.ffn_down_exps.weight
MXFP4
MXFP4
[2048, 4096, 256]
blk.12.ffn_down_shexp.weight
Q8_0
Q8_0
[2048, 4096]
blk.12.ffn_gate_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.12.ffn_gate_inp.weight
BF16
BF16
[4096, 256]
blk.12.ffn_gate_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.12.ffn_norm.weight
F32
F32
[4096]
blk.12.ffn_up_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.12.ffn_up_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.12.hc_attn_base.weight
F32
F32
[24]
blk.12.hc_attn_fn.weight
F32
F32
[16384, 24]
blk.12.hc_attn_scale.weight
F32
F32
[3]
blk.12.hc_ffn_base.weight
F32
F32
[24]
blk.12.hc_ffn_fn.weight
F32
F32
[16384, 24]
blk.12.hc_ffn_scale.weight
F32
F32
[3]
blk.12.indexer.attn_q_b.weight
Q8_0
Q8_0
[1024, 8192]
blk.12.indexer.proj.weight
BF16
BF16
[4096, 64]
blk.12.indexer_compressor_ape.weight
F32
F32
[256, 4]
blk.12.indexer_compressor_gate.weight
BF16
BF16
[4096, 256]
blk.12.indexer_compressor_kv.weight
BF16
BF16
[4096, 256]
blk.12.indexer_compressor_norm.weight
F32
F32
[128]
blk.13
blk.13.attn_compressor_ape.weight
F32
F32
[512, 128]
blk.13.attn_compressor_gate.weight
BF16
BF16
[4096, 512]
blk.13.attn_compressor_kv.weight
BF16
BF16
[4096, 512]
blk.13.attn_compressor_norm.weight
F32
F32
[512]
blk.13.attn_kv.weight
Q8_0
Q8_0
[4096, 512]
blk.13.attn_kv_a_norm.weight
F32
F32
[512]
blk.13.attn_norm.weight
F32
F32
[4096]
blk.13.attn_output_a.weight
Q8_0
Q8_0
[4096, 8192]
blk.13.attn_output_b.weight
Q8_0
Q8_0
[8192, 4096]
blk.13.attn_q_a.weight
Q8_0
Q8_0
[4096, 1024]
blk.13.attn_q_a_norm.weight
F32
F32
[1024]
blk.13.attn_q_b.weight
Q8_0
Q8_0
[1024, 32768]
blk.13.attn_sinks.weight
F32
F32
[64]
blk.13.exp_probs_b.bias
F32
F32
[256]
blk.13.ffn_down_exps.weight
MXFP4
MXFP4
[2048, 4096, 256]
blk.13.ffn_down_shexp.weight
Q8_0
Q8_0
[2048, 4096]
blk.13.ffn_gate_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.13.ffn_gate_inp.weight
BF16
BF16
[4096, 256]
blk.13.ffn_gate_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.13.ffn_norm.weight
F32
F32
[4096]
blk.13.ffn_up_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.13.ffn_up_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.13.hc_attn_base.weight
F32
F32
[24]
blk.13.hc_attn_fn.weight
F32
F32
[16384, 24]
blk.13.hc_attn_scale.weight
F32
F32
[3]
blk.13.hc_ffn_base.weight
F32
F32
[24]
blk.13.hc_ffn_fn.weight
F32
F32
[16384, 24]
blk.13.hc_ffn_scale.weight
F32
F32
[3]
blk.14
blk.14.attn_compressor_ape.weight
F32
F32
[1024, 4]
blk.14.attn_compressor_gate.weight
BF16
BF16
[4096, 1024]
blk.14.attn_compressor_kv.weight
BF16
BF16
[4096, 1024]
blk.14.attn_compressor_norm.weight
F32
F32
[512]
blk.14.attn_kv.weight
Q8_0
Q8_0
[4096, 512]
blk.14.attn_kv_a_norm.weight
F32
F32
[512]
blk.14.attn_norm.weight
F32
F32
[4096]
blk.14.attn_output_a.weight
Q8_0
Q8_0
[4096, 8192]
blk.14.attn_output_b.weight
Q8_0
Q8_0
[8192, 4096]
blk.14.attn_q_a.weight
Q8_0
Q8_0
[4096, 1024]
blk.14.attn_q_a_norm.weight
F32
F32
[1024]
blk.14.attn_q_b.weight
Q8_0
Q8_0
[1024, 32768]
blk.14.attn_sinks.weight
F32
F32
[64]
blk.14.exp_probs_b.bias
F32
F32
[256]
blk.14.ffn_down_exps.weight
MXFP4
MXFP4
[2048, 4096, 256]
blk.14.ffn_down_shexp.weight
Q8_0
Q8_0
[2048, 4096]
blk.14.ffn_gate_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.14.ffn_gate_inp.weight
BF16
BF16
[4096, 256]
blk.14.ffn_gate_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.14.ffn_norm.weight
F32
F32
[4096]
blk.14.ffn_up_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.14.ffn_up_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.14.hc_attn_base.weight
F32
F32
[24]
blk.14.hc_attn_fn.weight
F32
F32
[16384, 24]
blk.14.hc_attn_scale.weight
F32
F32
[3]
blk.14.hc_ffn_base.weight
F32
F32
[24]
blk.14.hc_ffn_fn.weight
F32
F32
[16384, 24]
blk.14.hc_ffn_scale.weight
F32
F32
[3]
blk.14.indexer.attn_q_b.weight
Q8_0
Q8_0
[1024, 8192]
blk.14.indexer.proj.weight
BF16
BF16
[4096, 64]
blk.14.indexer_compressor_ape.weight
F32
F32
[256, 4]
blk.14.indexer_compressor_gate.weight
BF16
BF16
[4096, 256]
blk.14.indexer_compressor_kv.weight
BF16
BF16
[4096, 256]
blk.14.indexer_compressor_norm.weight
F32
F32
[128]
blk.15
blk.15.attn_compressor_ape.weight
F32
F32
[512, 128]
blk.15.attn_compressor_gate.weight
BF16
BF16
[4096, 512]
blk.15.attn_compressor_kv.weight
BF16
BF16
[4096, 512]
blk.15.attn_compressor_norm.weight
F32
F32
[512]
blk.15.attn_kv.weight
Q8_0
Q8_0
[4096, 512]
blk.15.attn_kv_a_norm.weight
F32
F32
[512]
blk.15.attn_norm.weight
F32
F32
[4096]
blk.15.attn_output_a.weight
Q8_0
Q8_0
[4096, 8192]
blk.15.attn_output_b.weight
Q8_0
Q8_0
[8192, 4096]
blk.15.attn_q_a.weight
Q8_0
Q8_0
[4096, 1024]
blk.15.attn_q_a_norm.weight
F32
F32
[1024]
blk.15.attn_q_b.weight
Q8_0
Q8_0
[1024, 32768]
blk.15.attn_sinks.weight
F32
F32
[64]
blk.15.exp_probs_b.bias
F32
F32
[256]
blk.15.ffn_down_exps.weight
MXFP4
MXFP4
[2048, 4096, 256]
blk.15.ffn_down_shexp.weight
Q8_0
Q8_0
[2048, 4096]
blk.15.ffn_gate_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.15.ffn_gate_inp.weight
BF16
BF16
[4096, 256]
blk.15.ffn_gate_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.15.ffn_norm.weight
F32
F32
[4096]
blk.15.ffn_up_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.15.ffn_up_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.15.hc_attn_base.weight
F32
F32
[24]
blk.15.hc_attn_fn.weight
F32
F32
[16384, 24]
blk.15.hc_attn_scale.weight
F32
F32
[3]
blk.15.hc_ffn_base.weight
F32
F32
[24]
blk.15.hc_ffn_fn.weight
F32
F32
[16384, 24]
blk.15.hc_ffn_scale.weight
F32
F32
[3]
blk.16
blk.16.attn_compressor_ape.weight
F32
F32
[1024, 4]
blk.16.attn_compressor_gate.weight
BF16
BF16
[4096, 1024]
blk.16.attn_compressor_kv.weight
BF16
BF16
[4096, 1024]
blk.16.attn_compressor_norm.weight
F32
F32
[512]
blk.16.attn_kv.weight
Q8_0
Q8_0
[4096, 512]
blk.16.attn_kv_a_norm.weight
F32
F32
[512]
blk.16.attn_norm.weight
F32
F32
[4096]
blk.16.attn_output_a.weight
Q8_0
Q8_0
[4096, 8192]
blk.16.attn_output_b.weight
Q8_0
Q8_0
[8192, 4096]
blk.16.attn_q_a.weight
Q8_0
Q8_0
[4096, 1024]
blk.16.attn_q_a_norm.weight
F32
F32
[1024]
blk.16.attn_q_b.weight
Q8_0
Q8_0
[1024, 32768]
blk.16.attn_sinks.weight
F32
F32
[64]
blk.16.exp_probs_b.bias
F32
F32
[256]
blk.16.ffn_down_exps.weight
MXFP4
MXFP4
[2048, 4096, 256]
blk.16.ffn_down_shexp.weight
Q8_0
Q8_0
[2048, 4096]
blk.16.ffn_gate_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.16.ffn_gate_inp.weight
BF16
BF16
[4096, 256]
blk.16.ffn_gate_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.16.ffn_norm.weight
F32
F32
[4096]
blk.16.ffn_up_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.16.ffn_up_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.16.hc_attn_base.weight
F32
F32
[24]
blk.16.hc_attn_fn.weight
F32
F32
[16384, 24]
blk.16.hc_attn_scale.weight
F32
F32
[3]
blk.16.hc_ffn_base.weight
F32
F32
[24]
blk.16.hc_ffn_fn.weight
F32
F32
[16384, 24]
blk.16.hc_ffn_scale.weight
F32
F32
[3]
blk.16.indexer.attn_q_b.weight
Q8_0
Q8_0
[1024, 8192]
blk.16.indexer.proj.weight
BF16
BF16
[4096, 64]
blk.16.indexer_compressor_ape.weight
F32
F32
[256, 4]
blk.16.indexer_compressor_gate.weight
BF16
BF16
[4096, 256]
blk.16.indexer_compressor_kv.weight
BF16
BF16
[4096, 256]
blk.16.indexer_compressor_norm.weight
F32
F32
[128]
blk.17
blk.17.attn_compressor_ape.weight
F32
F32
[512, 128]
blk.17.attn_compressor_gate.weight
BF16
BF16
[4096, 512]
blk.17.attn_compressor_kv.weight
BF16
BF16
[4096, 512]
blk.17.attn_compressor_norm.weight
F32
F32
[512]
blk.17.attn_kv.weight
Q8_0
Q8_0
[4096, 512]
blk.17.attn_kv_a_norm.weight
F32
F32
[512]
blk.17.attn_norm.weight
F32
F32
[4096]
blk.17.attn_output_a.weight
Q8_0
Q8_0
[4096, 8192]
blk.17.attn_output_b.weight
Q8_0
Q8_0
[8192, 4096]
blk.17.attn_q_a.weight
Q8_0
Q8_0
[4096, 1024]
blk.17.attn_q_a_norm.weight
F32
F32
[1024]
blk.17.attn_q_b.weight
Q8_0
Q8_0
[1024, 32768]
blk.17.attn_sinks.weight
F32
F32
[64]
blk.17.exp_probs_b.bias
F32
F32
[256]
blk.17.ffn_down_exps.weight
MXFP4
MXFP4
[2048, 4096, 256]
blk.17.ffn_down_shexp.weight
Q8_0
Q8_0
[2048, 4096]
blk.17.ffn_gate_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.17.ffn_gate_inp.weight
BF16
BF16
[4096, 256]
blk.17.ffn_gate_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.17.ffn_norm.weight
F32
F32
[4096]
blk.17.ffn_up_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.17.ffn_up_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.17.hc_attn_base.weight
F32
F32
[24]
blk.17.hc_attn_fn.weight
F32
F32
[16384, 24]
blk.17.hc_attn_scale.weight
F32
F32
[3]
blk.17.hc_ffn_base.weight
F32
F32
[24]
blk.17.hc_ffn_fn.weight
F32
F32
[16384, 24]
blk.17.hc_ffn_scale.weight
F32
F32
[3]
blk.18
blk.18.attn_compressor_ape.weight
F32
F32
[1024, 4]
blk.18.attn_compressor_gate.weight
BF16
BF16
[4096, 1024]
blk.18.attn_compressor_kv.weight
BF16
BF16
[4096, 1024]
blk.18.attn_compressor_norm.weight
F32
F32
[512]
blk.18.attn_kv.weight
Q8_0
Q8_0
[4096, 512]
blk.18.attn_kv_a_norm.weight
F32
F32
[512]
blk.18.attn_norm.weight
F32
F32
[4096]
blk.18.attn_output_a.weight
Q8_0
Q8_0
[4096, 8192]
blk.18.attn_output_b.weight
Q8_0
Q8_0
[8192, 4096]
blk.18.attn_q_a.weight
Q8_0
Q8_0
[4096, 1024]
blk.18.attn_q_a_norm.weight
F32
F32
[1024]
blk.18.attn_q_b.weight
Q8_0
Q8_0
[1024, 32768]
blk.18.attn_sinks.weight
F32
F32
[64]
blk.18.exp_probs_b.bias
F32
F32
[256]
blk.18.ffn_down_exps.weight
MXFP4
MXFP4
[2048, 4096, 256]
blk.18.ffn_down_shexp.weight
Q8_0
Q8_0
[2048, 4096]
blk.18.ffn_gate_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.18.ffn_gate_inp.weight
BF16
BF16
[4096, 256]
blk.18.ffn_gate_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.18.ffn_norm.weight
F32
F32
[4096]
blk.18.ffn_up_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.18.ffn_up_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.18.hc_attn_base.weight
F32
F32
[24]
blk.18.hc_attn_fn.weight
F32
F32
[16384, 24]
blk.18.hc_attn_scale.weight
F32
F32
[3]
blk.18.hc_ffn_base.weight
F32
F32
[24]
blk.18.hc_ffn_fn.weight
F32
F32
[16384, 24]
blk.18.hc_ffn_scale.weight
F32
F32
[3]
blk.18.indexer.attn_q_b.weight
Q8_0
Q8_0
[1024, 8192]
blk.18.indexer.proj.weight
BF16
BF16
[4096, 64]
blk.18.indexer_compressor_ape.weight
F32
F32
[256, 4]
blk.18.indexer_compressor_gate.weight
BF16
BF16
[4096, 256]
blk.18.indexer_compressor_kv.weight
BF16
BF16
[4096, 256]
blk.18.indexer_compressor_norm.weight
F32
F32
[128]
blk.19
blk.19.attn_compressor_ape.weight
F32
F32
[512, 128]
blk.19.attn_compressor_gate.weight
BF16
BF16
[4096, 512]
blk.19.attn_compressor_kv.weight
BF16
BF16
[4096, 512]
blk.19.attn_compressor_norm.weight
F32
F32
[512]
blk.19.attn_kv.weight
Q8_0
Q8_0
[4096, 512]
blk.19.attn_kv_a_norm.weight
F32
F32
[512]
blk.19.attn_norm.weight
F32
F32
[4096]
blk.19.attn_output_a.weight
Q8_0
Q8_0
[4096, 8192]
blk.19.attn_output_b.weight
Q8_0
Q8_0
[8192, 4096]
blk.19.attn_q_a.weight
Q8_0
Q8_0
[4096, 1024]
blk.19.attn_q_a_norm.weight
F32
F32
[1024]
blk.19.attn_q_b.weight
Q8_0
Q8_0
[1024, 32768]
blk.19.attn_sinks.weight
F32
F32
[64]
blk.19.exp_probs_b.bias
F32
F32
[256]
blk.19.ffn_down_exps.weight
MXFP4
MXFP4
[2048, 4096, 256]
blk.19.ffn_down_shexp.weight
Q8_0
Q8_0
[2048, 4096]
blk.19.ffn_gate_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.19.ffn_gate_inp.weight
BF16
BF16
[4096, 256]
blk.19.ffn_gate_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.19.ffn_norm.weight
F32
F32
[4096]
blk.19.ffn_up_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.19.ffn_up_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.19.hc_attn_base.weight
F32
F32
[24]
blk.19.hc_attn_fn.weight
F32
F32
[16384, 24]
blk.19.hc_attn_scale.weight
F32
F32
[3]
blk.19.hc_ffn_base.weight
F32
F32
[24]
blk.19.hc_ffn_fn.weight
F32
F32
[16384, 24]
blk.19.hc_ffn_scale.weight
F32
F32
[3]
blk.20
blk.20.attn_compressor_ape.weight
F32
F32
[1024, 4]
blk.20.attn_compressor_gate.weight
BF16
BF16
[4096, 1024]
blk.20.attn_compressor_kv.weight
BF16
BF16
[4096, 1024]
blk.20.attn_compressor_norm.weight
F32
F32
[512]
blk.20.attn_kv.weight
Q8_0
Q8_0
[4096, 512]
blk.20.attn_kv_a_norm.weight
F32
F32
[512]
blk.20.attn_norm.weight
F32
F32
[4096]
blk.20.attn_output_a.weight
Q8_0
Q8_0
[4096, 8192]
blk.20.attn_output_b.weight
Q8_0
Q8_0
[8192, 4096]
blk.20.attn_q_a.weight
Q8_0
Q8_0
[4096, 1024]
blk.20.attn_q_a_norm.weight
F32
F32
[1024]
blk.20.attn_q_b.weight
Q8_0
Q8_0
[1024, 32768]
blk.20.attn_sinks.weight
F32
F32
[64]
blk.20.exp_probs_b.bias
F32
F32
[256]
blk.20.ffn_down_exps.weight
MXFP4
MXFP4
[2048, 4096, 256]
blk.20.ffn_down_shexp.weight
Q8_0
Q8_0
[2048, 4096]
blk.20.ffn_gate_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.20.ffn_gate_inp.weight
BF16
BF16
[4096, 256]
blk.20.ffn_gate_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.20.ffn_norm.weight
F32
F32
[4096]
blk.20.ffn_up_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.20.ffn_up_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.20.hc_attn_base.weight
F32
F32
[24]
blk.20.hc_attn_fn.weight
F32
F32
[16384, 24]
blk.20.hc_attn_scale.weight
F32
F32
[3]
blk.20.hc_ffn_base.weight
F32
F32
[24]
blk.20.hc_ffn_fn.weight
F32
F32
[16384, 24]
blk.20.hc_ffn_scale.weight
F32
F32
[3]
blk.20.indexer.attn_q_b.weight
Q8_0
Q8_0
[1024, 8192]
blk.20.indexer.proj.weight
BF16
BF16
[4096, 64]
blk.20.indexer_compressor_ape.weight
F32
F32
[256, 4]
blk.20.indexer_compressor_gate.weight
BF16
BF16
[4096, 256]
blk.20.indexer_compressor_kv.weight
BF16
BF16
[4096, 256]
blk.20.indexer_compressor_norm.weight
F32
F32
[128]
blk.21
blk.21.attn_compressor_ape.weight
F32
F32
[512, 128]
blk.21.attn_compressor_gate.weight
BF16
BF16
[4096, 512]
blk.21.attn_compressor_kv.weight
BF16
BF16
[4096, 512]
blk.21.attn_compressor_norm.weight
F32
F32
[512]
blk.21.attn_kv.weight
Q8_0
Q8_0
[4096, 512]
blk.21.attn_kv_a_norm.weight
F32
F32
[512]
blk.21.attn_norm.weight
F32
F32
[4096]
blk.21.attn_output_a.weight
Q8_0
Q8_0
[4096, 8192]
blk.21.attn_output_b.weight
Q8_0
Q8_0
[8192, 4096]
blk.21.attn_q_a.weight
Q8_0
Q8_0
[4096, 1024]
blk.21.attn_q_a_norm.weight
F32
F32
[1024]
blk.21.attn_q_b.weight
Q8_0
Q8_0
[1024, 32768]
blk.21.attn_sinks.weight
F32
F32
[64]
blk.21.exp_probs_b.bias
F32
F32
[256]
blk.21.ffn_down_exps.weight
MXFP4
MXFP4
[2048, 4096, 256]
blk.21.ffn_down_shexp.weight
Q8_0
Q8_0
[2048, 4096]
blk.21.ffn_gate_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.21.ffn_gate_inp.weight
BF16
BF16
[4096, 256]
blk.21.ffn_gate_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.21.ffn_norm.weight
F32
F32
[4096]
blk.21.ffn_up_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.21.ffn_up_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.21.hc_attn_base.weight
F32
F32
[24]
blk.21.hc_attn_fn.weight
F32
F32
[16384, 24]
blk.21.hc_attn_scale.weight
F32
F32
[3]
blk.21.hc_ffn_base.weight
F32
F32
[24]
blk.21.hc_ffn_fn.weight
F32
F32
[16384, 24]
blk.21.hc_ffn_scale.weight
F32
F32
[3]
blk.22
blk.22.attn_compressor_ape.weight
F32
F32
[1024, 4]
blk.22.attn_compressor_gate.weight
BF16
BF16
[4096, 1024]
blk.22.attn_compressor_kv.weight
BF16
BF16
[4096, 1024]
blk.22.attn_compressor_norm.weight
F32
F32
[512]
blk.22.attn_kv.weight
Q8_0
Q8_0
[4096, 512]
blk.22.attn_kv_a_norm.weight
F32
F32
[512]
blk.22.attn_norm.weight
F32
F32
[4096]
blk.22.attn_output_a.weight
Q8_0
Q8_0
[4096, 8192]
blk.22.attn_output_b.weight
Q8_0
Q8_0
[8192, 4096]
blk.22.attn_q_a.weight
Q8_0
Q8_0
[4096, 1024]
blk.22.attn_q_a_norm.weight
F32
F32
[1024]
blk.22.attn_q_b.weight
Q8_0
Q8_0
[1024, 32768]
blk.22.attn_sinks.weight
F32
F32
[64]
blk.22.exp_probs_b.bias
F32
F32
[256]
blk.22.ffn_down_exps.weight
MXFP4
MXFP4
[2048, 4096, 256]
blk.22.ffn_down_shexp.weight
Q8_0
Q8_0
[2048, 4096]
blk.22.ffn_gate_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.22.ffn_gate_inp.weight
BF16
BF16
[4096, 256]
blk.22.ffn_gate_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.22.ffn_norm.weight
F32
F32
[4096]
blk.22.ffn_up_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.22.ffn_up_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.22.hc_attn_base.weight
F32
F32
[24]
blk.22.hc_attn_fn.weight
F32
F32
[16384, 24]
blk.22.hc_attn_scale.weight
F32
F32
[3]
blk.22.hc_ffn_base.weight
F32
F32
[24]
blk.22.hc_ffn_fn.weight
F32
F32
[16384, 24]
blk.22.hc_ffn_scale.weight
F32
F32
[3]
blk.22.indexer.attn_q_b.weight
Q8_0
Q8_0
[1024, 8192]
blk.22.indexer.proj.weight
BF16
BF16
[4096, 64]
blk.22.indexer_compressor_ape.weight
F32
F32
[256, 4]
blk.22.indexer_compressor_gate.weight
BF16
BF16
[4096, 256]
blk.22.indexer_compressor_kv.weight
BF16
BF16
[4096, 256]
blk.22.indexer_compressor_norm.weight
F32
F32
[128]
blk.23
blk.23.attn_compressor_ape.weight
F32
F32
[512, 128]
blk.23.attn_compressor_gate.weight
BF16
BF16
[4096, 512]
blk.23.attn_compressor_kv.weight
BF16
BF16
[4096, 512]
blk.23.attn_compressor_norm.weight
F32
F32
[512]
blk.23.attn_kv.weight
Q8_0
Q8_0
[4096, 512]
blk.23.attn_kv_a_norm.weight
F32
F32
[512]
blk.23.attn_norm.weight
F32
F32
[4096]
blk.23.attn_output_a.weight
Q8_0
Q8_0
[4096, 8192]
blk.23.attn_output_b.weight
Q8_0
Q8_0
[8192, 4096]
blk.23.attn_q_a.weight
Q8_0
Q8_0
[4096, 1024]
blk.23.attn_q_a_norm.weight
F32
F32
[1024]
blk.23.attn_q_b.weight
Q8_0
Q8_0
[1024, 32768]
blk.23.attn_sinks.weight
F32
F32
[64]
blk.23.exp_probs_b.bias
F32
F32
[256]
blk.23.ffn_down_exps.weight
MXFP4
MXFP4
[2048, 4096, 256]
blk.23.ffn_down_shexp.weight
Q8_0
Q8_0
[2048, 4096]
blk.23.ffn_gate_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.23.ffn_gate_inp.weight
BF16
BF16
[4096, 256]
blk.23.ffn_gate_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.23.ffn_norm.weight
F32
F32
[4096]
blk.23.ffn_up_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.23.ffn_up_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.23.hc_attn_base.weight
F32
F32
[24]
blk.23.hc_attn_fn.weight
F32
F32
[16384, 24]
blk.23.hc_attn_scale.weight
F32
F32
[3]
blk.23.hc_ffn_base.weight
F32
F32
[24]
blk.23.hc_ffn_fn.weight
F32
F32
[16384, 24]
blk.23.hc_ffn_scale.weight
F32
F32
[3]
blk.24
blk.24.attn_compressor_ape.weight
F32
F32
[1024, 4]
blk.24.attn_compressor_gate.weight
BF16
BF16
[4096, 1024]
blk.24.attn_compressor_kv.weight
BF16
BF16
[4096, 1024]
blk.24.attn_compressor_norm.weight
F32
F32
[512]
blk.24.attn_kv.weight
Q8_0
Q8_0
[4096, 512]
blk.24.attn_kv_a_norm.weight
F32
F32
[512]
blk.24.attn_norm.weight
F32
F32
[4096]
blk.24.attn_output_a.weight
Q8_0
Q8_0
[4096, 8192]
blk.24.attn_output_b.weight
Q8_0
Q8_0
[8192, 4096]
blk.24.attn_q_a.weight
Q8_0
Q8_0
[4096, 1024]
blk.24.attn_q_a_norm.weight
F32
F32
[1024]
blk.24.attn_q_b.weight
Q8_0
Q8_0
[1024, 32768]
blk.24.attn_sinks.weight
F32
F32
[64]
blk.24.exp_probs_b.bias
F32
F32
[256]
blk.24.ffn_down_exps.weight
MXFP4
MXFP4
[2048, 4096, 256]
blk.24.ffn_down_shexp.weight
Q8_0
Q8_0
[2048, 4096]
blk.24.ffn_gate_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.24.ffn_gate_inp.weight
BF16
BF16
[4096, 256]
blk.24.ffn_gate_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.24.ffn_norm.weight
F32
F32
[4096]
blk.24.ffn_up_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.24.ffn_up_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.24.hc_attn_base.weight
F32
F32
[24]
blk.24.hc_attn_fn.weight
F32
F32
[16384, 24]
blk.24.hc_attn_scale.weight
F32
F32
[3]
blk.24.hc_ffn_base.weight
F32
F32
[24]
blk.24.hc_ffn_fn.weight
F32
F32
[16384, 24]
blk.24.hc_ffn_scale.weight
F32
F32
[3]
blk.24.indexer.attn_q_b.weight
Q8_0
Q8_0
[1024, 8192]
blk.24.indexer.proj.weight
BF16
BF16
[4096, 64]
blk.24.indexer_compressor_ape.weight
F32
F32
[256, 4]
blk.24.indexer_compressor_gate.weight
BF16
BF16
[4096, 256]
blk.24.indexer_compressor_kv.weight
BF16
BF16
[4096, 256]
blk.24.indexer_compressor_norm.weight
F32
F32
[128]
blk.25
blk.25.attn_compressor_ape.weight
F32
F32
[512, 128]
blk.25.attn_compressor_gate.weight
BF16
BF16
[4096, 512]
blk.25.attn_compressor_kv.weight
BF16
BF16
[4096, 512]
blk.25.attn_compressor_norm.weight
F32
F32
[512]
blk.25.attn_kv.weight
Q8_0
Q8_0
[4096, 512]
blk.25.attn_kv_a_norm.weight
F32
F32
[512]
blk.25.attn_norm.weight
F32
F32
[4096]
blk.25.attn_output_a.weight
Q8_0
Q8_0
[4096, 8192]
blk.25.attn_output_b.weight
Q8_0
Q8_0
[8192, 4096]
blk.25.attn_q_a.weight
Q8_0
Q8_0
[4096, 1024]
blk.25.attn_q_a_norm.weight
F32
F32
[1024]
blk.25.attn_q_b.weight
Q8_0
Q8_0
[1024, 32768]
blk.25.attn_sinks.weight
F32
F32
[64]
blk.25.exp_probs_b.bias
F32
F32
[256]
blk.25.ffn_down_exps.weight
MXFP4
MXFP4
[2048, 4096, 256]
blk.25.ffn_down_shexp.weight
Q8_0
Q8_0
[2048, 4096]
blk.25.ffn_gate_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.25.ffn_gate_inp.weight
BF16
BF16
[4096, 256]
blk.25.ffn_gate_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.25.ffn_norm.weight
F32
F32
[4096]
blk.25.ffn_up_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.25.ffn_up_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.25.hc_attn_base.weight
F32
F32
[24]
blk.25.hc_attn_fn.weight
F32
F32
[16384, 24]
blk.25.hc_attn_scale.weight
F32
F32
[3]
blk.25.hc_ffn_base.weight
F32
F32
[24]
blk.25.hc_ffn_fn.weight
F32
F32
[16384, 24]
blk.25.hc_ffn_scale.weight
F32
F32
[3]
blk.26
blk.26.attn_compressor_ape.weight
F32
F32
[1024, 4]
blk.26.attn_compressor_gate.weight
BF16
BF16
[4096, 1024]
blk.26.attn_compressor_kv.weight
BF16
BF16
[4096, 1024]
blk.26.attn_compressor_norm.weight
F32
F32
[512]
blk.26.attn_kv.weight
Q8_0
Q8_0
[4096, 512]
blk.26.attn_kv_a_norm.weight
F32
F32
[512]
blk.26.attn_norm.weight
F32
F32
[4096]
blk.26.attn_output_a.weight
Q8_0
Q8_0
[4096, 8192]
blk.26.attn_output_b.weight
Q8_0
Q8_0
[8192, 4096]
blk.26.attn_q_a.weight
Q8_0
Q8_0
[4096, 1024]
blk.26.attn_q_a_norm.weight
F32
F32
[1024]
blk.26.attn_q_b.weight
Q8_0
Q8_0
[1024, 32768]
blk.26.attn_sinks.weight
F32
F32
[64]
blk.26.exp_probs_b.bias
F32
F32
[256]
blk.26.ffn_down_exps.weight
MXFP4
MXFP4
[2048, 4096, 256]
blk.26.ffn_down_shexp.weight
Q8_0
Q8_0
[2048, 4096]
blk.26.ffn_gate_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.26.ffn_gate_inp.weight
BF16
BF16
[4096, 256]
blk.26.ffn_gate_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.26.ffn_norm.weight
F32
F32
[4096]
blk.26.ffn_up_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.26.ffn_up_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.26.hc_attn_base.weight
F32
F32
[24]
blk.26.hc_attn_fn.weight
F32
F32
[16384, 24]
blk.26.hc_attn_scale.weight
F32
F32
[3]
blk.26.hc_ffn_base.weight
F32
F32
[24]
blk.26.hc_ffn_fn.weight
F32
F32
[16384, 24]
blk.26.hc_ffn_scale.weight
F32
F32
[3]
blk.26.indexer.attn_q_b.weight
Q8_0
Q8_0
[1024, 8192]
blk.26.indexer.proj.weight
BF16
BF16
[4096, 64]
blk.26.indexer_compressor_ape.weight
F32
F32
[256, 4]
blk.26.indexer_compressor_gate.weight
BF16
BF16
[4096, 256]
blk.26.indexer_compressor_kv.weight
BF16
BF16
[4096, 256]
blk.26.indexer_compressor_norm.weight
F32
F32
[128]
blk.27
blk.27.attn_compressor_ape.weight
F32
F32
[512, 128]
blk.27.attn_compressor_gate.weight
BF16
BF16
[4096, 512]
blk.27.attn_compressor_kv.weight
BF16
BF16
[4096, 512]
blk.27.attn_compressor_norm.weight
F32
F32
[512]
blk.27.attn_kv.weight
Q8_0
Q8_0
[4096, 512]
blk.27.attn_kv_a_norm.weight
F32
F32
[512]
blk.27.attn_norm.weight
F32
F32
[4096]
blk.27.attn_output_a.weight
Q8_0
Q8_0
[4096, 8192]
blk.27.attn_output_b.weight
Q8_0
Q8_0
[8192, 4096]
blk.27.attn_q_a.weight
Q8_0
Q8_0
[4096, 1024]
blk.27.attn_q_a_norm.weight
F32
F32
[1024]
blk.27.attn_q_b.weight
Q8_0
Q8_0
[1024, 32768]
blk.27.attn_sinks.weight
F32
F32
[64]
blk.27.exp_probs_b.bias
F32
F32
[256]
blk.27.ffn_down_exps.weight
MXFP4
MXFP4
[2048, 4096, 256]
blk.27.ffn_down_shexp.weight
Q8_0
Q8_0
[2048, 4096]
blk.27.ffn_gate_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.27.ffn_gate_inp.weight
BF16
BF16
[4096, 256]
blk.27.ffn_gate_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.27.ffn_norm.weight
F32
F32
[4096]
blk.27.ffn_up_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.27.ffn_up_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.27.hc_attn_base.weight
F32
F32
[24]
blk.27.hc_attn_fn.weight
F32
F32
[16384, 24]
blk.27.hc_attn_scale.weight
F32
F32
[3]
blk.27.hc_ffn_base.weight
F32
F32
[24]
blk.27.hc_ffn_fn.weight
F32
F32
[16384, 24]
blk.27.hc_ffn_scale.weight
F32
F32
[3]
blk.28
blk.28.attn_compressor_ape.weight
F32
F32
[1024, 4]
blk.28.attn_compressor_gate.weight
BF16
BF16
[4096, 1024]
blk.28.attn_compressor_kv.weight
BF16
BF16
[4096, 1024]
blk.28.attn_compressor_norm.weight
F32
F32
[512]
blk.28.attn_kv.weight
Q8_0
Q8_0
[4096, 512]
blk.28.attn_kv_a_norm.weight
F32
F32
[512]
blk.28.attn_norm.weight
F32
F32
[4096]
blk.28.attn_output_a.weight
Q8_0
Q8_0
[4096, 8192]
blk.28.attn_output_b.weight
Q8_0
Q8_0
[8192, 4096]
blk.28.attn_q_a.weight
Q8_0
Q8_0
[4096, 1024]
blk.28.attn_q_a_norm.weight
F32
F32
[1024]
blk.28.attn_q_b.weight
Q8_0
Q8_0
[1024, 32768]
blk.28.attn_sinks.weight
F32
F32
[64]
blk.28.exp_probs_b.bias
F32
F32
[256]
blk.28.ffn_down_exps.weight
MXFP4
MXFP4
[2048, 4096, 256]
blk.28.ffn_down_shexp.weight
Q8_0
Q8_0
[2048, 4096]
blk.28.ffn_gate_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.28.ffn_gate_inp.weight
BF16
BF16
[4096, 256]
blk.28.ffn_gate_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.28.ffn_norm.weight
F32
F32
[4096]
blk.28.ffn_up_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.28.ffn_up_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.28.hc_attn_base.weight
F32
F32
[24]
blk.28.hc_attn_fn.weight
F32
F32
[16384, 24]
blk.28.hc_attn_scale.weight
F32
F32
[3]
blk.28.hc_ffn_base.weight
F32
F32
[24]
blk.28.hc_ffn_fn.weight
F32
F32
[16384, 24]
blk.28.hc_ffn_scale.weight
F32
F32
[3]
blk.28.indexer.attn_q_b.weight
Q8_0
Q8_0
[1024, 8192]
blk.28.indexer.proj.weight
BF16
BF16
[4096, 64]
blk.28.indexer_compressor_ape.weight
F32
F32
[256, 4]
blk.28.indexer_compressor_gate.weight
BF16
BF16
[4096, 256]
blk.28.indexer_compressor_kv.weight
BF16
BF16
[4096, 256]
blk.28.indexer_compressor_norm.weight
F32
F32
[128]
blk.29
blk.29.attn_compressor_ape.weight
F32
F32
[512, 128]
blk.29.attn_compressor_gate.weight
BF16
BF16
[4096, 512]
blk.29.attn_compressor_kv.weight
BF16
BF16
[4096, 512]
blk.29.attn_compressor_norm.weight
F32
F32
[512]
blk.29.attn_kv.weight
Q8_0
Q8_0
[4096, 512]
blk.29.attn_kv_a_norm.weight
F32
F32
[512]
blk.29.attn_norm.weight
F32
F32
[4096]
blk.29.attn_output_a.weight
Q8_0
Q8_0
[4096, 8192]
blk.29.attn_output_b.weight
Q8_0
Q8_0
[8192, 4096]
blk.29.attn_q_a.weight
Q8_0
Q8_0
[4096, 1024]
blk.29.attn_q_a_norm.weight
F32
F32
[1024]
blk.29.attn_q_b.weight
Q8_0
Q8_0
[1024, 32768]
blk.29.attn_sinks.weight
F32
F32
[64]
blk.29.exp_probs_b.bias
F32
F32
[256]
blk.29.ffn_down_exps.weight
MXFP4
MXFP4
[2048, 4096, 256]
blk.29.ffn_down_shexp.weight
Q8_0
Q8_0
[2048, 4096]
blk.29.ffn_gate_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.29.ffn_gate_inp.weight
BF16
BF16
[4096, 256]
blk.29.ffn_gate_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.29.ffn_norm.weight
F32
F32
[4096]
blk.29.ffn_up_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.29.ffn_up_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.29.hc_attn_base.weight
F32
F32
[24]
blk.29.hc_attn_fn.weight
F32
F32
[16384, 24]
blk.29.hc_attn_scale.weight
F32
F32
[3]
blk.29.hc_ffn_base.weight
F32
F32
[24]
blk.29.hc_ffn_fn.weight
F32
F32
[16384, 24]
blk.29.hc_ffn_scale.weight
F32
F32
[3]
blk.30
blk.30.attn_compressor_ape.weight
F32
F32
[1024, 4]
blk.30.attn_compressor_gate.weight
BF16
BF16
[4096, 1024]
blk.30.attn_compressor_kv.weight
BF16
BF16
[4096, 1024]
blk.30.attn_compressor_norm.weight
F32
F32
[512]
blk.30.attn_kv.weight
Q8_0
Q8_0
[4096, 512]
blk.30.attn_kv_a_norm.weight
F32
F32
[512]
blk.30.attn_norm.weight
F32
F32
[4096]
blk.30.attn_output_a.weight
Q8_0
Q8_0
[4096, 8192]
blk.30.attn_output_b.weight
Q8_0
Q8_0
[8192, 4096]
blk.30.attn_q_a.weight
Q8_0
Q8_0
[4096, 1024]
blk.30.attn_q_a_norm.weight
F32
F32
[1024]
blk.30.attn_q_b.weight
Q8_0
Q8_0
[1024, 32768]
blk.30.attn_sinks.weight
F32
F32
[64]
blk.30.exp_probs_b.bias
F32
F32
[256]
blk.30.ffn_down_exps.weight
MXFP4
MXFP4
[2048, 4096, 256]
blk.30.ffn_down_shexp.weight
Q8_0
Q8_0
[2048, 4096]
blk.30.ffn_gate_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.30.ffn_gate_inp.weight
BF16
BF16
[4096, 256]
blk.30.ffn_gate_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.30.ffn_norm.weight
F32
F32
[4096]
blk.30.ffn_up_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.30.ffn_up_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.30.hc_attn_base.weight
F32
F32
[24]
blk.30.hc_attn_fn.weight
F32
F32
[16384, 24]
blk.30.hc_attn_scale.weight
F32
F32
[3]
blk.30.hc_ffn_base.weight
F32
F32
[24]
blk.30.hc_ffn_fn.weight
F32
F32
[16384, 24]
blk.30.hc_ffn_scale.weight
F32
F32
[3]
blk.30.indexer.attn_q_b.weight
Q8_0
Q8_0
[1024, 8192]
blk.30.indexer.proj.weight
BF16
BF16
[4096, 64]
blk.30.indexer_compressor_ape.weight
F32
F32
[256, 4]
blk.30.indexer_compressor_gate.weight
BF16
BF16
[4096, 256]
blk.30.indexer_compressor_kv.weight
BF16
BF16
[4096, 256]
blk.30.indexer_compressor_norm.weight
F32
F32
[128]
blk.31
blk.31.attn_compressor_ape.weight
F32
F32
[512, 128]
blk.31.attn_compressor_gate.weight
BF16
BF16
[4096, 512]
blk.31.attn_compressor_kv.weight
BF16
BF16
[4096, 512]
blk.31.attn_compressor_norm.weight
F32
F32
[512]
blk.31.attn_kv.weight
Q8_0
Q8_0
[4096, 512]
blk.31.attn_kv_a_norm.weight
F32
F32
[512]
blk.31.attn_norm.weight
F32
F32
[4096]
blk.31.attn_output_a.weight
Q8_0
Q8_0
[4096, 8192]
blk.31.attn_output_b.weight
Q8_0
Q8_0
[8192, 4096]
blk.31.attn_q_a.weight
Q8_0
Q8_0
[4096, 1024]
blk.31.attn_q_a_norm.weight
F32
F32
[1024]
blk.31.attn_q_b.weight
Q8_0
Q8_0
[1024, 32768]
blk.31.attn_sinks.weight
F32
F32
[64]
blk.31.exp_probs_b.bias
F32
F32
[256]
blk.31.ffn_down_exps.weight
MXFP4
MXFP4
[2048, 4096, 256]
blk.31.ffn_down_shexp.weight
Q8_0
Q8_0
[2048, 4096]
blk.31.ffn_gate_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.31.ffn_gate_inp.weight
BF16
BF16
[4096, 256]
blk.31.ffn_gate_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.31.ffn_norm.weight
F32
F32
[4096]
blk.31.ffn_up_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.31.ffn_up_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.31.hc_attn_base.weight
F32
F32
[24]
blk.31.hc_attn_fn.weight
F32
F32
[16384, 24]
blk.31.hc_attn_scale.weight
F32
F32
[3]
blk.31.hc_ffn_base.weight
F32
F32
[24]
blk.31.hc_ffn_fn.weight
F32
F32
[16384, 24]
blk.31.hc_ffn_scale.weight
F32
F32
[3]
blk.32
blk.32.attn_compressor_ape.weight
F32
F32
[1024, 4]
blk.32.attn_compressor_gate.weight
BF16
BF16
[4096, 1024]
blk.32.attn_compressor_kv.weight
BF16
BF16
[4096, 1024]
blk.32.attn_compressor_norm.weight
F32
F32
[512]
blk.32.attn_kv.weight
Q8_0
Q8_0
[4096, 512]
blk.32.attn_kv_a_norm.weight
F32
F32
[512]
blk.32.attn_norm.weight
F32
F32
[4096]
blk.32.attn_output_a.weight
Q8_0
Q8_0
[4096, 8192]
blk.32.attn_output_b.weight
Q8_0
Q8_0
[8192, 4096]
blk.32.attn_q_a.weight
Q8_0
Q8_0
[4096, 1024]
blk.32.attn_q_a_norm.weight
F32
F32
[1024]
blk.32.attn_q_b.weight
Q8_0
Q8_0
[1024, 32768]
blk.32.attn_sinks.weight
F32
F32
[64]
blk.32.exp_probs_b.bias
F32
F32
[256]
blk.32.ffn_down_exps.weight
MXFP4
MXFP4
[2048, 4096, 256]
blk.32.ffn_down_shexp.weight
Q8_0
Q8_0
[2048, 4096]
blk.32.ffn_gate_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.32.ffn_gate_inp.weight
BF16
BF16
[4096, 256]
blk.32.ffn_gate_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.32.ffn_norm.weight
F32
F32
[4096]
blk.32.ffn_up_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.32.ffn_up_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.32.hc_attn_base.weight
F32
F32
[24]
blk.32.hc_attn_fn.weight
F32
F32
[16384, 24]
blk.32.hc_attn_scale.weight
F32
F32
[3]
blk.32.hc_ffn_base.weight
F32
F32
[24]
blk.32.hc_ffn_fn.weight
F32
F32
[16384, 24]
blk.32.hc_ffn_scale.weight
F32
F32
[3]
blk.32.indexer.attn_q_b.weight
Q8_0
Q8_0
[1024, 8192]
blk.32.indexer.proj.weight
BF16
BF16
[4096, 64]
blk.32.indexer_compressor_ape.weight
F32
F32
[256, 4]
blk.32.indexer_compressor_gate.weight
BF16
BF16
[4096, 256]
blk.32.indexer_compressor_kv.weight
BF16
BF16
[4096, 256]
blk.32.indexer_compressor_norm.weight
F32
F32
[128]
blk.33
blk.33.attn_compressor_ape.weight
F32
F32
[512, 128]
blk.33.attn_compressor_gate.weight
BF16
BF16
[4096, 512]
blk.33.attn_compressor_kv.weight
BF16
BF16
[4096, 512]
blk.33.attn_compressor_norm.weight
F32
F32
[512]
blk.33.attn_kv.weight
Q8_0
Q8_0
[4096, 512]
blk.33.attn_kv_a_norm.weight
F32
F32
[512]
blk.33.attn_norm.weight
F32
F32
[4096]
blk.33.attn_output_a.weight
Q8_0
Q8_0
[4096, 8192]
blk.33.attn_output_b.weight
Q8_0
Q8_0
[8192, 4096]
blk.33.attn_q_a.weight
Q8_0
Q8_0
[4096, 1024]
blk.33.attn_q_a_norm.weight
F32
F32
[1024]
blk.33.attn_q_b.weight
Q8_0
Q8_0
[1024, 32768]
blk.33.attn_sinks.weight
F32
F32
[64]
blk.33.exp_probs_b.bias
F32
F32
[256]
blk.33.ffn_down_exps.weight
MXFP4
MXFP4
[2048, 4096, 256]
blk.33.ffn_down_shexp.weight
Q8_0
Q8_0
[2048, 4096]
blk.33.ffn_gate_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.33.ffn_gate_inp.weight
BF16
BF16
[4096, 256]
blk.33.ffn_gate_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.33.ffn_norm.weight
F32
F32
[4096]
blk.33.ffn_up_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.33.ffn_up_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.33.hc_attn_base.weight
F32
F32
[24]
blk.33.hc_attn_fn.weight
F32
F32
[16384, 24]
blk.33.hc_attn_scale.weight
F32
F32
[3]
blk.33.hc_ffn_base.weight
F32
F32
[24]
blk.33.hc_ffn_fn.weight
F32
F32
[16384, 24]
blk.33.hc_ffn_scale.weight
F32
F32
[3]
blk.34
blk.34.attn_compressor_ape.weight
F32
F32
[1024, 4]
blk.34.attn_compressor_gate.weight
BF16
BF16
[4096, 1024]
blk.34.attn_compressor_kv.weight
BF16
BF16
[4096, 1024]
blk.34.attn_compressor_norm.weight
F32
F32
[512]
blk.34.attn_kv.weight
Q8_0
Q8_0
[4096, 512]
blk.34.attn_kv_a_norm.weight
F32
F32
[512]
blk.34.attn_norm.weight
F32
F32
[4096]
blk.34.attn_output_a.weight
Q8_0
Q8_0
[4096, 8192]
blk.34.attn_output_b.weight
Q8_0
Q8_0
[8192, 4096]
blk.34.attn_q_a.weight
Q8_0
Q8_0
[4096, 1024]
blk.34.attn_q_a_norm.weight
F32
F32
[1024]
blk.34.attn_q_b.weight
Q8_0
Q8_0
[1024, 32768]
blk.34.attn_sinks.weight
F32
F32
[64]
blk.34.exp_probs_b.bias
F32
F32
[256]
blk.34.ffn_down_exps.weight
MXFP4
MXFP4
[2048, 4096, 256]
blk.34.ffn_down_shexp.weight
Q8_0
Q8_0
[2048, 4096]
blk.34.ffn_gate_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.34.ffn_gate_inp.weight
BF16
BF16
[4096, 256]
blk.34.ffn_gate_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.34.ffn_norm.weight
F32
F32
[4096]
blk.34.ffn_up_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.34.ffn_up_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.34.hc_attn_base.weight
F32
F32
[24]
blk.34.hc_attn_fn.weight
F32
F32
[16384, 24]
blk.34.hc_attn_scale.weight
F32
F32
[3]
blk.34.hc_ffn_base.weight
F32
F32
[24]
blk.34.hc_ffn_fn.weight
F32
F32
[16384, 24]
blk.34.hc_ffn_scale.weight
F32
F32
[3]
blk.34.indexer.attn_q_b.weight
Q8_0
Q8_0
[1024, 8192]
blk.34.indexer.proj.weight
BF16
BF16
[4096, 64]
blk.34.indexer_compressor_ape.weight
F32
F32
[256, 4]
blk.34.indexer_compressor_gate.weight
BF16
BF16
[4096, 256]
blk.34.indexer_compressor_kv.weight
BF16
BF16
[4096, 256]
blk.34.indexer_compressor_norm.weight
F32
F32
[128]
blk.35
blk.35.attn_compressor_ape.weight
F32
F32
[512, 128]
blk.35.attn_compressor_gate.weight
BF16
BF16
[4096, 512]
blk.35.attn_compressor_kv.weight
BF16
BF16
[4096, 512]
blk.35.attn_compressor_norm.weight
F32
F32
[512]
blk.35.attn_kv.weight
Q8_0
Q8_0
[4096, 512]
blk.35.attn_kv_a_norm.weight
F32
F32
[512]
blk.35.attn_norm.weight
F32
F32
[4096]
blk.35.attn_output_a.weight
Q8_0
Q8_0
[4096, 8192]
blk.35.attn_output_b.weight
Q8_0
Q8_0
[8192, 4096]
blk.35.attn_q_a.weight
Q8_0
Q8_0
[4096, 1024]
blk.35.attn_q_a_norm.weight
F32
F32
[1024]
blk.35.attn_q_b.weight
Q8_0
Q8_0
[1024, 32768]
blk.35.attn_sinks.weight
F32
F32
[64]
blk.35.exp_probs_b.bias
F32
F32
[256]
blk.35.ffn_down_exps.weight
MXFP4
MXFP4
[2048, 4096, 256]
blk.35.ffn_down_shexp.weight
Q8_0
Q8_0
[2048, 4096]
blk.35.ffn_gate_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.35.ffn_gate_inp.weight
BF16
BF16
[4096, 256]
blk.35.ffn_gate_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.35.ffn_norm.weight
F32
F32
[4096]
blk.35.ffn_up_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.35.ffn_up_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.35.hc_attn_base.weight
F32
F32
[24]
blk.35.hc_attn_fn.weight
F32
F32
[16384, 24]
blk.35.hc_attn_scale.weight
F32
F32
[3]
blk.35.hc_ffn_base.weight
F32
F32
[24]
blk.35.hc_ffn_fn.weight
F32
F32
[16384, 24]
blk.35.hc_ffn_scale.weight
F32
F32
[3]
blk.36
blk.36.attn_compressor_ape.weight
F32
F32
[1024, 4]
blk.36.attn_compressor_gate.weight
BF16
BF16
[4096, 1024]
blk.36.attn_compressor_kv.weight
BF16
BF16
[4096, 1024]
blk.36.attn_compressor_norm.weight
F32
F32
[512]
blk.36.attn_kv.weight
Q8_0
Q8_0
[4096, 512]
blk.36.attn_kv_a_norm.weight
F32
F32
[512]
blk.36.attn_norm.weight
F32
F32
[4096]
blk.36.attn_output_a.weight
Q8_0
Q8_0
[4096, 8192]
blk.36.attn_output_b.weight
Q8_0
Q8_0
[8192, 4096]
blk.36.attn_q_a.weight
Q8_0
Q8_0
[4096, 1024]
blk.36.attn_q_a_norm.weight
F32
F32
[1024]
blk.36.attn_q_b.weight
Q8_0
Q8_0
[1024, 32768]
blk.36.attn_sinks.weight
F32
F32
[64]
blk.36.exp_probs_b.bias
F32
F32
[256]
blk.36.ffn_down_exps.weight
MXFP4
MXFP4
[2048, 4096, 256]
blk.36.ffn_down_shexp.weight
Q8_0
Q8_0
[2048, 4096]
blk.36.ffn_gate_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.36.ffn_gate_inp.weight
BF16
BF16
[4096, 256]
blk.36.ffn_gate_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.36.ffn_norm.weight
F32
F32
[4096]
blk.36.ffn_up_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.36.ffn_up_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.36.hc_attn_base.weight
F32
F32
[24]
blk.36.hc_attn_fn.weight
F32
F32
[16384, 24]
blk.36.hc_attn_scale.weight
F32
F32
[3]
blk.36.hc_ffn_base.weight
F32
F32
[24]
blk.36.hc_ffn_fn.weight
F32
F32
[16384, 24]
blk.36.hc_ffn_scale.weight
F32
F32
[3]
blk.36.indexer.attn_q_b.weight
Q8_0
Q8_0
[1024, 8192]
blk.36.indexer.proj.weight
BF16
BF16
[4096, 64]
blk.36.indexer_compressor_ape.weight
F32
F32
[256, 4]
blk.36.indexer_compressor_gate.weight
BF16
BF16
[4096, 256]
blk.36.indexer_compressor_kv.weight
BF16
BF16
[4096, 256]
blk.36.indexer_compressor_norm.weight
F32
F32
[128]
blk.37
blk.37.attn_compressor_ape.weight
F32
F32
[512, 128]
blk.37.attn_compressor_gate.weight
BF16
BF16
[4096, 512]
blk.37.attn_compressor_kv.weight
BF16
BF16
[4096, 512]
blk.37.attn_compressor_norm.weight
F32
F32
[512]
blk.37.attn_kv.weight
Q8_0
Q8_0
[4096, 512]
blk.37.attn_kv_a_norm.weight
F32
F32
[512]
blk.37.attn_norm.weight
F32
F32
[4096]
blk.37.attn_output_a.weight
Q8_0
Q8_0
[4096, 8192]
blk.37.attn_output_b.weight
Q8_0
Q8_0
[8192, 4096]
blk.37.attn_q_a.weight
Q8_0
Q8_0
[4096, 1024]
blk.37.attn_q_a_norm.weight
F32
F32
[1024]
blk.37.attn_q_b.weight
Q8_0
Q8_0
[1024, 32768]
blk.37.attn_sinks.weight
F32
F32
[64]
blk.37.exp_probs_b.bias
F32
F32
[256]
blk.37.ffn_down_exps.weight
MXFP4
MXFP4
[2048, 4096, 256]
blk.37.ffn_down_shexp.weight
Q8_0
Q8_0
[2048, 4096]
blk.37.ffn_gate_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.37.ffn_gate_inp.weight
BF16
BF16
[4096, 256]
blk.37.ffn_gate_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.37.ffn_norm.weight
F32
F32
[4096]
blk.37.ffn_up_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.37.ffn_up_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.37.hc_attn_base.weight
F32
F32
[24]
blk.37.hc_attn_fn.weight
F32
F32
[16384, 24]
blk.37.hc_attn_scale.weight
F32
F32
[3]
blk.37.hc_ffn_base.weight
F32
F32
[24]
blk.37.hc_ffn_fn.weight
F32
F32
[16384, 24]
blk.37.hc_ffn_scale.weight
F32
F32
[3]
blk.38
blk.38.attn_compressor_ape.weight
F32
F32
[1024, 4]
blk.38.attn_compressor_gate.weight
BF16
BF16
[4096, 1024]
blk.38.attn_compressor_kv.weight
BF16
BF16
[4096, 1024]
blk.38.attn_compressor_norm.weight
F32
F32
[512]
blk.38.attn_kv.weight
Q8_0
Q8_0
[4096, 512]
blk.38.attn_kv_a_norm.weight
F32
F32
[512]
blk.38.attn_norm.weight
F32
F32
[4096]
blk.38.attn_output_a.weight
Q8_0
Q8_0
[4096, 8192]
blk.38.attn_output_b.weight
Q8_0
Q8_0
[8192, 4096]
blk.38.attn_q_a.weight
Q8_0
Q8_0
[4096, 1024]
blk.38.attn_q_a_norm.weight
F32
F32
[1024]
blk.38.attn_q_b.weight
Q8_0
Q8_0
[1024, 32768]
blk.38.attn_sinks.weight
F32
F32
[64]
blk.38.exp_probs_b.bias
F32
F32
[256]
blk.38.ffn_down_exps.weight
MXFP4
MXFP4
[2048, 4096, 256]
blk.38.ffn_down_shexp.weight
Q8_0
Q8_0
[2048, 4096]
blk.38.ffn_gate_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.38.ffn_gate_inp.weight
BF16
BF16
[4096, 256]
blk.38.ffn_gate_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.38.ffn_norm.weight
F32
F32
[4096]
blk.38.ffn_up_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.38.ffn_up_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.38.hc_attn_base.weight
F32
F32
[24]
blk.38.hc_attn_fn.weight
F32
F32
[16384, 24]
blk.38.hc_attn_scale.weight
F32
F32
[3]
blk.38.hc_ffn_base.weight
F32
F32
[24]
blk.38.hc_ffn_fn.weight
F32
F32
[16384, 24]
blk.38.hc_ffn_scale.weight
F32
F32
[3]
blk.38.indexer.attn_q_b.weight
Q8_0
Q8_0
[1024, 8192]
blk.38.indexer.proj.weight
BF16
BF16
[4096, 64]
blk.38.indexer_compressor_ape.weight
F32
F32
[256, 4]
blk.38.indexer_compressor_gate.weight
BF16
BF16
[4096, 256]
blk.38.indexer_compressor_kv.weight
BF16
BF16
[4096, 256]
blk.38.indexer_compressor_norm.weight
F32
F32
[128]
blk.39
blk.39.attn_compressor_ape.weight
F32
F32
[512, 128]
blk.39.attn_compressor_gate.weight
BF16
BF16
[4096, 512]
blk.39.attn_compressor_kv.weight
BF16
BF16
[4096, 512]
blk.39.attn_compressor_norm.weight
F32
F32
[512]
blk.39.attn_kv.weight
Q8_0
Q8_0
[4096, 512]
blk.39.attn_kv_a_norm.weight
F32
F32
[512]
blk.39.attn_norm.weight
F32
F32
[4096]
blk.39.attn_output_a.weight
Q8_0
Q8_0
[4096, 8192]
blk.39.attn_output_b.weight
Q8_0
Q8_0
[8192, 4096]
blk.39.attn_q_a.weight
Q8_0
Q8_0
[4096, 1024]
blk.39.attn_q_a_norm.weight
F32
F32
[1024]
blk.39.attn_q_b.weight
Q8_0
Q8_0
[1024, 32768]
blk.39.attn_sinks.weight
F32
F32
[64]
blk.39.exp_probs_b.bias
F32
F32
[256]
blk.39.ffn_down_exps.weight
MXFP4
MXFP4
[2048, 4096, 256]
blk.39.ffn_down_shexp.weight
Q8_0
Q8_0
[2048, 4096]
blk.39.ffn_gate_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.39.ffn_gate_inp.weight
BF16
BF16
[4096, 256]
blk.39.ffn_gate_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.39.ffn_norm.weight
F32
F32
[4096]
blk.39.ffn_up_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.39.ffn_up_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.39.hc_attn_base.weight
F32
F32
[24]
blk.39.hc_attn_fn.weight
F32
F32
[16384, 24]
blk.39.hc_attn_scale.weight
F32
F32
[3]
blk.39.hc_ffn_base.weight
F32
F32
[24]
blk.39.hc_ffn_fn.weight
F32
F32
[16384, 24]
blk.39.hc_ffn_scale.weight
F32
F32
[3]
blk.40
blk.40.attn_compressor_ape.weight
F32
F32
[1024, 4]
blk.40.attn_compressor_gate.weight
BF16
BF16
[4096, 1024]
blk.40.attn_compressor_kv.weight
BF16
BF16
[4096, 1024]
blk.40.attn_compressor_norm.weight
F32
F32
[512]
blk.40.attn_kv.weight
Q8_0
Q8_0
[4096, 512]
blk.40.attn_kv_a_norm.weight
F32
F32
[512]
blk.40.attn_norm.weight
F32
F32
[4096]
blk.40.attn_output_a.weight
Q8_0
Q8_0
[4096, 8192]
blk.40.attn_output_b.weight
Q8_0
Q8_0
[8192, 4096]
blk.40.attn_q_a.weight
Q8_0
Q8_0
[4096, 1024]
blk.40.attn_q_a_norm.weight
F32
F32
[1024]
blk.40.attn_q_b.weight
Q8_0
Q8_0
[1024, 32768]
blk.40.attn_sinks.weight
F32
F32
[64]
blk.40.exp_probs_b.bias
F32
F32
[256]
blk.40.ffn_down_exps.weight
MXFP4
MXFP4
[2048, 4096, 256]
blk.40.ffn_down_shexp.weight
Q8_0
Q8_0
[2048, 4096]
blk.40.ffn_gate_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.40.ffn_gate_inp.weight
BF16
BF16
[4096, 256]
blk.40.ffn_gate_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.40.ffn_norm.weight
F32
F32
[4096]
blk.40.ffn_up_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.40.ffn_up_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.40.hc_attn_base.weight
F32
F32
[24]
blk.40.hc_attn_fn.weight
F32
F32
[16384, 24]
blk.40.hc_attn_scale.weight
F32
F32
[3]
blk.40.hc_ffn_base.weight
F32
F32
[24]
blk.40.hc_ffn_fn.weight
F32
F32
[16384, 24]
blk.40.hc_ffn_scale.weight
F32
F32
[3]
blk.40.indexer.attn_q_b.weight
Q8_0
Q8_0
[1024, 8192]
blk.40.indexer.proj.weight
BF16
BF16
[4096, 64]
blk.40.indexer_compressor_ape.weight
F32
F32
[256, 4]
blk.40.indexer_compressor_gate.weight
BF16
BF16
[4096, 256]
blk.40.indexer_compressor_kv.weight
BF16
BF16
[4096, 256]
blk.40.indexer_compressor_norm.weight
F32
F32
[128]
blk.41
blk.41.attn_compressor_ape.weight
F32
F32
[512, 128]
blk.41.attn_compressor_gate.weight
BF16
BF16
[4096, 512]
blk.41.attn_compressor_kv.weight
BF16
BF16
[4096, 512]
blk.41.attn_compressor_norm.weight
F32
F32
[512]
blk.41.attn_kv.weight
Q8_0
Q8_0
[4096, 512]
blk.41.attn_kv_a_norm.weight
F32
F32
[512]
blk.41.attn_norm.weight
F32
F32
[4096]
blk.41.attn_output_a.weight
Q8_0
Q8_0
[4096, 8192]
blk.41.attn_output_b.weight
Q8_0
Q8_0
[8192, 4096]
blk.41.attn_q_a.weight
Q8_0
Q8_0
[4096, 1024]
blk.41.attn_q_a_norm.weight
F32
F32
[1024]
blk.41.attn_q_b.weight
Q8_0
Q8_0
[1024, 32768]
blk.41.attn_sinks.weight
F32
F32
[64]
blk.41.exp_probs_b.bias
F32
F32
[256]
blk.41.ffn_down_exps.weight
MXFP4
MXFP4
[2048, 4096, 256]
blk.41.ffn_down_shexp.weight
Q8_0
Q8_0
[2048, 4096]
blk.41.ffn_gate_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.41.ffn_gate_inp.weight
BF16
BF16
[4096, 256]
blk.41.ffn_gate_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.41.ffn_norm.weight
F32
F32
[4096]
blk.41.ffn_up_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.41.ffn_up_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.41.hc_attn_base.weight
F32
F32
[24]
blk.41.hc_attn_fn.weight
F32
F32
[16384, 24]
blk.41.hc_attn_scale.weight
F32
F32
[3]
blk.41.hc_ffn_base.weight
F32
F32
[24]
blk.41.hc_ffn_fn.weight
F32
F32
[16384, 24]
blk.41.hc_ffn_scale.weight
F32
F32
[3]
blk.42
blk.42.attn_compressor_ape.weight
F32
F32
[1024, 4]
blk.42.attn_compressor_gate.weight
BF16
BF16
[4096, 1024]
blk.42.attn_compressor_kv.weight
BF16
BF16
[4096, 1024]
blk.42.attn_compressor_norm.weight
F32
F32
[512]
blk.42.attn_kv.weight
Q8_0
Q8_0
[4096, 512]
blk.42.attn_kv_a_norm.weight
F32
F32
[512]
blk.42.attn_norm.weight
F32
F32
[4096]
blk.42.attn_output_a.weight
Q8_0
Q8_0
[4096, 8192]
blk.42.attn_output_b.weight
Q8_0
Q8_0
[8192, 4096]
blk.42.attn_q_a.weight
Q8_0
Q8_0
[4096, 1024]
blk.42.attn_q_a_norm.weight
F32
F32
[1024]
blk.42.attn_q_b.weight
Q8_0
Q8_0
[1024, 32768]
blk.42.attn_sinks.weight
F32
F32
[64]
blk.42.exp_probs_b.bias
F32
F32
[256]
blk.42.ffn_down_exps.weight
MXFP4
MXFP4
[2048, 4096, 256]
blk.42.ffn_down_shexp.weight
Q8_0
Q8_0
[2048, 4096]
blk.42.ffn_gate_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.42.ffn_gate_inp.weight
BF16
BF16
[4096, 256]
blk.42.ffn_gate_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.42.ffn_norm.weight
F32
F32
[4096]
blk.42.ffn_up_exps.weight
MXFP4
MXFP4
[4096, 2048, 256]
blk.42.ffn_up_shexp.weight
Q8_0
Q8_0
[4096, 2048]
blk.42.hc_attn_base.weight
F32
F32
[24]
blk.42.hc_attn_fn.weight
F32
F32
[16384, 24]
blk.42.hc_attn_scale.weight
F32
F32
[3]
blk.42.hc_ffn_base.weight
F32
F32
[24]
blk.42.hc_ffn_fn.weight
F32
F32
[16384, 24]
blk.42.hc_ffn_scale.weight
F32
F32
[3]
blk.42.indexer.attn_q_b.weight
Q8_0
Q8_0
[1024, 8192]
blk.42.indexer.proj.weight
BF16
BF16
[4096, 64]
blk.42.indexer_compressor_ape.weight
F32
F32
[256, 4]
blk.42.indexer_compressor_gate.weight
BF16
BF16
[4096, 256]
blk.42.indexer_compressor_kv.weight
BF16
BF16
[4096, 256]
blk.42.indexer_compressor_norm.weight
F32
F32
[128]
output.weight
BF16
BF16
[4096, 129280]
output_hc_base.weight
F32
F32
[4]
output_hc_fn.weight
F32
F32
[16384, 4]
output_hc_scale.weight
F32
F32
[1]
output_norm.weight
F32
F32
[4096]