Models
GitHub
Discord
Turbo
Sign in
Download
Models
Download
GitHub
Discord
Sign in
JollyLlama
/
Megrez-3B-Instruct
:Q8_0
81
Downloads
Updated
7 months ago
Megrez-3B-Instruct is a large language model trained by Infinigence AI. Megrez-3B aims to provide a fast inference, compact, and powerful edge-side intelligent solution through software-hardware co-design.
Megrez-3B-Instruct is a large language model trained by Infinigence AI. Megrez-3B aims to provide a fast inference, compact, and powerful edge-side intelligent solution through software-hardware co-design.
Cancel
Megrez-3B-Instruct:Q8_0
...
/
model
cc41d24d1c43 · 3.1GB
Metadata
general.architecture
llama
llama
general.file_type
Q8_0
Q8_0
llama.attention.head_count
40
40
llama.attention.head_count_kv
10
10
llama.attention.layer_norm_rms_epsilon
1e-05
1e-05
llama.block_count
32
32
llama.context_length
32768
32768
llama.embedding_length
2560
2560
llama.feed_forward_length
7168
7168
llama.rope.dimension_count
64
64
llama.rope.freq_base
5e+06
5e+06
llama.vocab_size
122880
122880
tokenizer.ggml.add_bos_token
false
false
tokenizer.ggml.eos_token_id
120005
120005
tokenizer.ggml.merges
[Ġ Ġ, ĠĠ ĠĠ, Ġ t, i n, Ġ a, ...]
[Ġ Ġ, ĠĠ ĠĠ, Ġ t, i n, Ġ a, ...]
tokenizer.ggml.model
gpt2
gpt2
tokenizer.ggml.padding_token_id
120002
120002
tokenizer.ggml.pre
megrez
megrez
tokenizer.ggml.token_type
[1, 1, 1, 1, 1, ...]
[1, 1, 1, 1, 1, ...]
tokenizer.ggml.tokens
[!, ", #, $, %, ...]
[!, ", #, $, %, ...]
quantize.imatrix.chunks_count
133
133
quantize.imatrix.dataset
/training_dir/calibration_datav3.txt
/training_dir/calibration_datav3.txt
quantize.imatrix.entries_count
224
224
quantize.imatrix.file
/models_out/Megrez-3B-Instruct-GGUF/Megrez-3B-Instruct.imatrix
/models_out/Megrez-3B-Instruct-GGUF/Megrez-3B-Instruct.imatrix
Tensor
Name
Type
Shape
token_embd.weight
Q8_0
Q8_0
[2560, 122880]
blk.0
blk.0.attn_k.weight
Q8_0
Q8_0
[2560, 640]
blk.0.attn_norm.weight
F32
F32
[2560]
blk.0.attn_output.weight
Q8_0
Q8_0
[2560, 2560]
blk.0.attn_q.weight
Q8_0
Q8_0
[2560, 2560]
blk.0.attn_v.weight
Q8_0
Q8_0
[2560, 640]
blk.0.ffn_down.weight
Q8_0
Q8_0
[7168, 2560]
blk.0.ffn_gate.weight
Q8_0
Q8_0
[2560, 7168]
blk.0.ffn_norm.weight
F32
F32
[2560]
blk.0.ffn_up.weight
Q8_0
Q8_0
[2560, 7168]
blk.1
blk.1.attn_k.weight
Q8_0
Q8_0
[2560, 640]
blk.1.attn_norm.weight
F32
F32
[2560]
blk.1.attn_output.weight
Q8_0
Q8_0
[2560, 2560]
blk.1.attn_q.weight
Q8_0
Q8_0
[2560, 2560]
blk.1.attn_v.weight
Q8_0
Q8_0
[2560, 640]
blk.1.ffn_down.weight
Q8_0
Q8_0
[7168, 2560]
blk.1.ffn_gate.weight
Q8_0
Q8_0
[2560, 7168]
blk.1.ffn_norm.weight
F32
F32
[2560]
blk.1.ffn_up.weight
Q8_0
Q8_0
[2560, 7168]
blk.2
blk.2.attn_k.weight
Q8_0
Q8_0
[2560, 640]
blk.2.attn_norm.weight
F32
F32
[2560]
blk.2.attn_output.weight
Q8_0
Q8_0
[2560, 2560]
blk.2.attn_q.weight
Q8_0
Q8_0
[2560, 2560]
blk.2.attn_v.weight
Q8_0
Q8_0
[2560, 640]
blk.2.ffn_down.weight
Q8_0
Q8_0
[7168, 2560]
blk.2.ffn_gate.weight
Q8_0
Q8_0
[2560, 7168]
blk.2.ffn_norm.weight
F32
F32
[2560]
blk.2.ffn_up.weight
Q8_0
Q8_0
[2560, 7168]
blk.3
blk.3.attn_k.weight
Q8_0
Q8_0
[2560, 640]
blk.3.attn_norm.weight
F32
F32
[2560]
blk.3.attn_output.weight
Q8_0
Q8_0
[2560, 2560]
blk.3.attn_q.weight
Q8_0
Q8_0
[2560, 2560]
blk.3.attn_v.weight
Q8_0
Q8_0
[2560, 640]
blk.3.ffn_down.weight
Q8_0
Q8_0
[7168, 2560]
blk.3.ffn_gate.weight
Q8_0
Q8_0
[2560, 7168]
blk.3.ffn_norm.weight
F32
F32
[2560]
blk.3.ffn_up.weight
Q8_0
Q8_0
[2560, 7168]
blk.4
blk.4.attn_k.weight
Q8_0
Q8_0
[2560, 640]
blk.4.attn_norm.weight
F32
F32
[2560]
blk.4.attn_output.weight
Q8_0
Q8_0
[2560, 2560]
blk.4.attn_q.weight
Q8_0
Q8_0
[2560, 2560]
blk.4.attn_v.weight
Q8_0
Q8_0
[2560, 640]
blk.4.ffn_down.weight
Q8_0
Q8_0
[7168, 2560]
blk.4.ffn_gate.weight
Q8_0
Q8_0
[2560, 7168]
blk.4.ffn_norm.weight
F32
F32
[2560]
blk.4.ffn_up.weight
Q8_0
Q8_0
[2560, 7168]
blk.5
blk.5.attn_k.weight
Q8_0
Q8_0
[2560, 640]
blk.5.attn_norm.weight
F32
F32
[2560]
blk.5.attn_output.weight
Q8_0
Q8_0
[2560, 2560]
blk.5.attn_q.weight
Q8_0
Q8_0
[2560, 2560]
blk.5.attn_v.weight
Q8_0
Q8_0
[2560, 640]
blk.5.ffn_down.weight
Q8_0
Q8_0
[7168, 2560]
blk.5.ffn_gate.weight
Q8_0
Q8_0
[2560, 7168]
blk.5.ffn_norm.weight
F32
F32
[2560]
blk.5.ffn_up.weight
Q8_0
Q8_0
[2560, 7168]
blk.6
blk.6.attn_k.weight
Q8_0
Q8_0
[2560, 640]
blk.6.attn_norm.weight
F32
F32
[2560]
blk.6.attn_output.weight
Q8_0
Q8_0
[2560, 2560]
blk.6.attn_q.weight
Q8_0
Q8_0
[2560, 2560]
blk.6.attn_v.weight
Q8_0
Q8_0
[2560, 640]
blk.6.ffn_down.weight
Q8_0
Q8_0
[7168, 2560]
blk.6.ffn_gate.weight
Q8_0
Q8_0
[2560, 7168]
blk.6.ffn_norm.weight
F32
F32
[2560]
blk.6.ffn_up.weight
Q8_0
Q8_0
[2560, 7168]
blk.7
blk.7.attn_k.weight
Q8_0
Q8_0
[2560, 640]
blk.7.attn_norm.weight
F32
F32
[2560]
blk.7.attn_output.weight
Q8_0
Q8_0
[2560, 2560]
blk.7.attn_q.weight
Q8_0
Q8_0
[2560, 2560]
blk.7.attn_v.weight
Q8_0
Q8_0
[2560, 640]
blk.7.ffn_down.weight
Q8_0
Q8_0
[7168, 2560]
blk.7.ffn_gate.weight
Q8_0
Q8_0
[2560, 7168]
blk.7.ffn_norm.weight
F32
F32
[2560]
blk.7.ffn_up.weight
Q8_0
Q8_0
[2560, 7168]
blk.8
blk.8.attn_k.weight
Q8_0
Q8_0
[2560, 640]
blk.8.attn_norm.weight
F32
F32
[2560]
blk.8.attn_output.weight
Q8_0
Q8_0
[2560, 2560]
blk.8.attn_q.weight
Q8_0
Q8_0
[2560, 2560]
blk.8.attn_v.weight
Q8_0
Q8_0
[2560, 640]
blk.8.ffn_down.weight
Q8_0
Q8_0
[7168, 2560]
blk.8.ffn_gate.weight
Q8_0
Q8_0
[2560, 7168]
blk.8.ffn_norm.weight
F32
F32
[2560]
blk.8.ffn_up.weight
Q8_0
Q8_0
[2560, 7168]
blk.9
blk.9.attn_k.weight
Q8_0
Q8_0
[2560, 640]
blk.9.attn_norm.weight
F32
F32
[2560]
blk.9.attn_output.weight
Q8_0
Q8_0
[2560, 2560]
blk.9.attn_q.weight
Q8_0
Q8_0
[2560, 2560]
blk.9.attn_v.weight
Q8_0
Q8_0
[2560, 640]
blk.9.ffn_down.weight
Q8_0
Q8_0
[7168, 2560]
blk.9.ffn_gate.weight
Q8_0
Q8_0
[2560, 7168]
blk.9.ffn_norm.weight
F32
F32
[2560]
blk.9.ffn_up.weight
Q8_0
Q8_0
[2560, 7168]
blk.10
blk.10.attn_k.weight
Q8_0
Q8_0
[2560, 640]
blk.10.attn_norm.weight
F32
F32
[2560]
blk.10.attn_output.weight
Q8_0
Q8_0
[2560, 2560]
blk.10.attn_q.weight
Q8_0
Q8_0
[2560, 2560]
blk.10.attn_v.weight
Q8_0
Q8_0
[2560, 640]
blk.10.ffn_down.weight
Q8_0
Q8_0
[7168, 2560]
blk.10.ffn_gate.weight
Q8_0
Q8_0
[2560, 7168]
blk.10.ffn_norm.weight
F32
F32
[2560]
blk.10.ffn_up.weight
Q8_0
Q8_0
[2560, 7168]
blk.11
blk.11.attn_k.weight
Q8_0
Q8_0
[2560, 640]
blk.11.attn_norm.weight
F32
F32
[2560]
blk.11.attn_output.weight
Q8_0
Q8_0
[2560, 2560]
blk.11.attn_q.weight
Q8_0
Q8_0
[2560, 2560]
blk.11.attn_v.weight
Q8_0
Q8_0
[2560, 640]
blk.11.ffn_down.weight
Q8_0
Q8_0
[7168, 2560]
blk.11.ffn_gate.weight
Q8_0
Q8_0
[2560, 7168]
blk.11.ffn_norm.weight
F32
F32
[2560]
blk.11.ffn_up.weight
Q8_0
Q8_0
[2560, 7168]
blk.12
blk.12.attn_k.weight
Q8_0
Q8_0
[2560, 640]
blk.12.attn_norm.weight
F32
F32
[2560]
blk.12.attn_output.weight
Q8_0
Q8_0
[2560, 2560]
blk.12.attn_q.weight
Q8_0
Q8_0
[2560, 2560]
blk.12.attn_v.weight
Q8_0
Q8_0
[2560, 640]
blk.12.ffn_down.weight
Q8_0
Q8_0
[7168, 2560]
blk.12.ffn_gate.weight
Q8_0
Q8_0
[2560, 7168]
blk.12.ffn_norm.weight
F32
F32
[2560]
blk.12.ffn_up.weight
Q8_0
Q8_0
[2560, 7168]
blk.13
blk.13.attn_k.weight
Q8_0
Q8_0
[2560, 640]
blk.13.attn_norm.weight
F32
F32
[2560]
blk.13.attn_output.weight
Q8_0
Q8_0
[2560, 2560]
blk.13.attn_q.weight
Q8_0
Q8_0
[2560, 2560]
blk.13.attn_v.weight
Q8_0
Q8_0
[2560, 640]
blk.13.ffn_down.weight
Q8_0
Q8_0
[7168, 2560]
blk.13.ffn_gate.weight
Q8_0
Q8_0
[2560, 7168]
blk.13.ffn_norm.weight
F32
F32
[2560]
blk.13.ffn_up.weight
Q8_0
Q8_0
[2560, 7168]
blk.14
blk.14.attn_k.weight
Q8_0
Q8_0
[2560, 640]
blk.14.attn_norm.weight
F32
F32
[2560]
blk.14.attn_output.weight
Q8_0
Q8_0
[2560, 2560]
blk.14.attn_q.weight
Q8_0
Q8_0
[2560, 2560]
blk.14.attn_v.weight
Q8_0
Q8_0
[2560, 640]
blk.14.ffn_down.weight
Q8_0
Q8_0
[7168, 2560]
blk.14.ffn_gate.weight
Q8_0
Q8_0
[2560, 7168]
blk.14.ffn_norm.weight
F32
F32
[2560]
blk.14.ffn_up.weight
Q8_0
Q8_0
[2560, 7168]
blk.15
blk.15.attn_k.weight
Q8_0
Q8_0
[2560, 640]
blk.15.attn_norm.weight
F32
F32
[2560]
blk.15.attn_output.weight
Q8_0
Q8_0
[2560, 2560]
blk.15.attn_q.weight
Q8_0
Q8_0
[2560, 2560]
blk.15.attn_v.weight
Q8_0
Q8_0
[2560, 640]
blk.15.ffn_down.weight
Q8_0
Q8_0
[7168, 2560]
blk.15.ffn_gate.weight
Q8_0
Q8_0
[2560, 7168]
blk.15.ffn_norm.weight
F32
F32
[2560]
blk.15.ffn_up.weight
Q8_0
Q8_0
[2560, 7168]
blk.16
blk.16.attn_k.weight
Q8_0
Q8_0
[2560, 640]
blk.16.attn_norm.weight
F32
F32
[2560]
blk.16.attn_output.weight
Q8_0
Q8_0
[2560, 2560]
blk.16.attn_q.weight
Q8_0
Q8_0
[2560, 2560]
blk.16.attn_v.weight
Q8_0
Q8_0
[2560, 640]
blk.16.ffn_down.weight
Q8_0
Q8_0
[7168, 2560]
blk.16.ffn_gate.weight
Q8_0
Q8_0
[2560, 7168]
blk.16.ffn_norm.weight
F32
F32
[2560]
blk.16.ffn_up.weight
Q8_0
Q8_0
[2560, 7168]
blk.17
blk.17.attn_k.weight
Q8_0
Q8_0
[2560, 640]
blk.17.attn_norm.weight
F32
F32
[2560]
blk.17.attn_output.weight
Q8_0
Q8_0
[2560, 2560]
blk.17.attn_q.weight
Q8_0
Q8_0
[2560, 2560]
blk.17.attn_v.weight
Q8_0
Q8_0
[2560, 640]
blk.17.ffn_down.weight
Q8_0
Q8_0
[7168, 2560]
blk.17.ffn_gate.weight
Q8_0
Q8_0
[2560, 7168]
blk.17.ffn_norm.weight
F32
F32
[2560]
blk.17.ffn_up.weight
Q8_0
Q8_0
[2560, 7168]
blk.18
blk.18.attn_k.weight
Q8_0
Q8_0
[2560, 640]
blk.18.attn_norm.weight
F32
F32
[2560]
blk.18.attn_output.weight
Q8_0
Q8_0
[2560, 2560]
blk.18.attn_q.weight
Q8_0
Q8_0
[2560, 2560]
blk.18.attn_v.weight
Q8_0
Q8_0
[2560, 640]
blk.18.ffn_down.weight
Q8_0
Q8_0
[7168, 2560]
blk.18.ffn_gate.weight
Q8_0
Q8_0
[2560, 7168]
blk.18.ffn_norm.weight
F32
F32
[2560]
blk.18.ffn_up.weight
Q8_0
Q8_0
[2560, 7168]
blk.19
blk.19.attn_k.weight
Q8_0
Q8_0
[2560, 640]
blk.19.attn_norm.weight
F32
F32
[2560]
blk.19.attn_output.weight
Q8_0
Q8_0
[2560, 2560]
blk.19.attn_q.weight
Q8_0
Q8_0
[2560, 2560]
blk.19.attn_v.weight
Q8_0
Q8_0
[2560, 640]
blk.19.ffn_down.weight
Q8_0
Q8_0
[7168, 2560]
blk.19.ffn_gate.weight
Q8_0
Q8_0
[2560, 7168]
blk.19.ffn_norm.weight
F32
F32
[2560]
blk.19.ffn_up.weight
Q8_0
Q8_0
[2560, 7168]
blk.20
blk.20.attn_k.weight
Q8_0
Q8_0
[2560, 640]
blk.20.attn_norm.weight
F32
F32
[2560]
blk.20.attn_output.weight
Q8_0
Q8_0
[2560, 2560]
blk.20.attn_q.weight
Q8_0
Q8_0
[2560, 2560]
blk.20.attn_v.weight
Q8_0
Q8_0
[2560, 640]
blk.20.ffn_down.weight
Q8_0
Q8_0
[7168, 2560]
blk.20.ffn_gate.weight
Q8_0
Q8_0
[2560, 7168]
blk.20.ffn_norm.weight
F32
F32
[2560]
blk.20.ffn_up.weight
Q8_0
Q8_0
[2560, 7168]
blk.21
blk.21.attn_k.weight
Q8_0
Q8_0
[2560, 640]
blk.21.attn_norm.weight
F32
F32
[2560]
blk.21.attn_output.weight
Q8_0
Q8_0
[2560, 2560]
blk.21.attn_q.weight
Q8_0
Q8_0
[2560, 2560]
blk.21.attn_v.weight
Q8_0
Q8_0
[2560, 640]
blk.21.ffn_down.weight
Q8_0
Q8_0
[7168, 2560]
blk.21.ffn_gate.weight
Q8_0
Q8_0
[2560, 7168]
blk.21.ffn_norm.weight
F32
F32
[2560]
blk.21.ffn_up.weight
Q8_0
Q8_0
[2560, 7168]
blk.22
blk.22.attn_k.weight
Q8_0
Q8_0
[2560, 640]
blk.22.attn_norm.weight
F32
F32
[2560]
blk.22.attn_output.weight
Q8_0
Q8_0
[2560, 2560]
blk.22.attn_q.weight
Q8_0
Q8_0
[2560, 2560]
blk.22.attn_v.weight
Q8_0
Q8_0
[2560, 640]
blk.22.ffn_down.weight
Q8_0
Q8_0
[7168, 2560]
blk.22.ffn_gate.weight
Q8_0
Q8_0
[2560, 7168]
blk.22.ffn_norm.weight
F32
F32
[2560]
blk.22.ffn_up.weight
Q8_0
Q8_0
[2560, 7168]
blk.23
blk.23.attn_k.weight
Q8_0
Q8_0
[2560, 640]
blk.23.attn_norm.weight
F32
F32
[2560]
blk.23.attn_output.weight
Q8_0
Q8_0
[2560, 2560]
blk.23.attn_q.weight
Q8_0
Q8_0
[2560, 2560]
blk.23.attn_v.weight
Q8_0
Q8_0
[2560, 640]
blk.23.ffn_down.weight
Q8_0
Q8_0
[7168, 2560]
blk.23.ffn_gate.weight
Q8_0
Q8_0
[2560, 7168]
blk.23.ffn_norm.weight
F32
F32
[2560]
blk.23.ffn_up.weight
Q8_0
Q8_0
[2560, 7168]
blk.24
blk.24.attn_k.weight
Q8_0
Q8_0
[2560, 640]
blk.24.attn_norm.weight
F32
F32
[2560]
blk.24.attn_output.weight
Q8_0
Q8_0
[2560, 2560]
blk.24.attn_q.weight
Q8_0
Q8_0
[2560, 2560]
blk.24.attn_v.weight
Q8_0
Q8_0
[2560, 640]
blk.24.ffn_down.weight
Q8_0
Q8_0
[7168, 2560]
blk.24.ffn_gate.weight
Q8_0
Q8_0
[2560, 7168]
blk.24.ffn_norm.weight
F32
F32
[2560]
blk.24.ffn_up.weight
Q8_0
Q8_0
[2560, 7168]
blk.25
blk.25.attn_k.weight
Q8_0
Q8_0
[2560, 640]
blk.25.attn_norm.weight
F32
F32
[2560]
blk.25.attn_output.weight
Q8_0
Q8_0
[2560, 2560]
blk.25.attn_q.weight
Q8_0
Q8_0
[2560, 2560]
blk.25.attn_v.weight
Q8_0
Q8_0
[2560, 640]
blk.25.ffn_down.weight
Q8_0
Q8_0
[7168, 2560]
blk.25.ffn_gate.weight
Q8_0
Q8_0
[2560, 7168]
blk.25.ffn_norm.weight
F32
F32
[2560]
blk.25.ffn_up.weight
Q8_0
Q8_0
[2560, 7168]
blk.26
blk.26.attn_k.weight
Q8_0
Q8_0
[2560, 640]
blk.26.attn_norm.weight
F32
F32
[2560]
blk.26.attn_output.weight
Q8_0
Q8_0
[2560, 2560]
blk.26.attn_q.weight
Q8_0
Q8_0
[2560, 2560]
blk.26.attn_v.weight
Q8_0
Q8_0
[2560, 640]
blk.26.ffn_down.weight
Q8_0
Q8_0
[7168, 2560]
blk.26.ffn_gate.weight
Q8_0
Q8_0
[2560, 7168]
blk.26.ffn_norm.weight
F32
F32
[2560]
blk.26.ffn_up.weight
Q8_0
Q8_0
[2560, 7168]
blk.27
blk.27.attn_k.weight
Q8_0
Q8_0
[2560, 640]
blk.27.attn_norm.weight
F32
F32
[2560]
blk.27.attn_output.weight
Q8_0
Q8_0
[2560, 2560]
blk.27.attn_q.weight
Q8_0
Q8_0
[2560, 2560]
blk.27.attn_v.weight
Q8_0
Q8_0
[2560, 640]
blk.27.ffn_down.weight
Q8_0
Q8_0
[7168, 2560]
blk.27.ffn_gate.weight
Q8_0
Q8_0
[2560, 7168]
blk.27.ffn_norm.weight
F32
F32
[2560]
blk.27.ffn_up.weight
Q8_0
Q8_0
[2560, 7168]
blk.28
blk.28.attn_k.weight
Q8_0
Q8_0
[2560, 640]
blk.28.attn_norm.weight
F32
F32
[2560]
blk.28.attn_output.weight
Q8_0
Q8_0
[2560, 2560]
blk.28.attn_q.weight
Q8_0
Q8_0
[2560, 2560]
blk.28.attn_v.weight
Q8_0
Q8_0
[2560, 640]
blk.28.ffn_down.weight
Q8_0
Q8_0
[7168, 2560]
blk.28.ffn_gate.weight
Q8_0
Q8_0
[2560, 7168]
blk.28.ffn_norm.weight
F32
F32
[2560]
blk.28.ffn_up.weight
Q8_0
Q8_0
[2560, 7168]
blk.29
blk.29.attn_k.weight
Q8_0
Q8_0
[2560, 640]
blk.29.attn_norm.weight
F32
F32
[2560]
blk.29.attn_output.weight
Q8_0
Q8_0
[2560, 2560]
blk.29.attn_q.weight
Q8_0
Q8_0
[2560, 2560]
blk.29.attn_v.weight
Q8_0
Q8_0
[2560, 640]
blk.29.ffn_down.weight
Q8_0
Q8_0
[7168, 2560]
blk.29.ffn_gate.weight
Q8_0
Q8_0
[2560, 7168]
blk.29.ffn_norm.weight
F32
F32
[2560]
blk.29.ffn_up.weight
Q8_0
Q8_0
[2560, 7168]
blk.30
blk.30.attn_k.weight
Q8_0
Q8_0
[2560, 640]
blk.30.attn_norm.weight
F32
F32
[2560]
blk.30.attn_output.weight
Q8_0
Q8_0
[2560, 2560]
blk.30.attn_q.weight
Q8_0
Q8_0
[2560, 2560]
blk.30.attn_v.weight
Q8_0
Q8_0
[2560, 640]
blk.30.ffn_down.weight
Q8_0
Q8_0
[7168, 2560]
blk.30.ffn_gate.weight
Q8_0
Q8_0
[2560, 7168]
blk.30.ffn_norm.weight
F32
F32
[2560]
blk.30.ffn_up.weight
Q8_0
Q8_0
[2560, 7168]
blk.31
blk.31.attn_k.weight
Q8_0
Q8_0
[2560, 640]
blk.31.attn_norm.weight
F32
F32
[2560]
blk.31.attn_output.weight
Q8_0
Q8_0
[2560, 2560]
blk.31.attn_q.weight
Q8_0
Q8_0
[2560, 2560]
blk.31.attn_v.weight
Q8_0
Q8_0
[2560, 640]
blk.31.ffn_down.weight
Q8_0
Q8_0
[7168, 2560]
blk.31.ffn_gate.weight
Q8_0
Q8_0
[2560, 7168]
blk.31.ffn_norm.weight
F32
F32
[2560]
blk.31.ffn_up.weight
Q8_0
Q8_0
[2560, 7168]
output.weight
Q8_0
Q8_0
[2560, 122880]
output_norm.weight
F32
F32
[2560]