Models
GitHub
Discord
Turbo
Sign in
Download
Models
Download
GitHub
Discord
Sign in
aquiffoo
/
aquif-3.5
:7b
44
Downloads
Updated
2 weeks ago
State-of-the-art compact LLMs from aquif, with two MoE models (2.6B-A0.6B and 12B-A4B-Think) and three dense models (3B, 7B and 8B-Think), released on August 30th 2025.
State-of-the-art compact LLMs from aquif, with two MoE models (2.6B-A0.6B and 12B-A4B-Think) and three dense models (3B, 7B and 8B-Think), released on August 30th 2025.
Cancel
3b
7b
8b
aquif-3.5:7b
...
/
model
18a4b8b6039b · 7.7GB
Metadata
general.architecture
qwen2
qwen2
general.file_type
Q8_0
Q8_0
qwen2.attention.head_count
28
28
qwen2.attention.head_count_kv
4
4
qwen2.attention.layer_norm_rms_epsilon
1e-05
1e-05
qwen2.block_count
28
28
qwen2.context_length
16384
16384
qwen2.embedding_length
3584
3584
qwen2.feed_forward_length
18944
18944
qwen2.rope.freq_base
1e+06
1e+06
tokenizer.ggml.add_bos_token
false
false
tokenizer.ggml.bos_token_id
0
0
tokenizer.ggml.eos_token_id
27
27
tokenizer.ggml.mask_token_id
4
4
tokenizer.ggml.merges
[Ġ ì, Ġ Ġ, Ġ ë, ì Ŀ, í ķ, ...]
[Ġ ì, Ġ Ġ, Ġ ë, ì Ŀ, í ķ, ...]
tokenizer.ggml.model
gpt2
gpt2
tokenizer.ggml.padding_token_id
1
1
tokenizer.ggml.pre
a.x-4.0
a.x-4.0
tokenizer.ggml.seperator_token_id
3
3
tokenizer.ggml.token_type
[3, 3, 3, 3, 3, ...]
[3, 3, 3, 3, 3, ...]
tokenizer.ggml.tokens
[<|endoftext|>, <|pad|>, <|unk|>, <|sep|>, <|mask|>, ...]
[<|endoftext|>, <|pad|>, <|unk|>, <|sep|>, <|mask|>, ...]
tokenizer.ggml.unknown_token_id
2
2
mradermacher.convert_type
hf
hf
mradermacher.quantize_version
2
2
mradermacher.quantized_at
2025-08-31T06:10:12+02:00
2025-08-31T06:10:12+02:00
mradermacher.quantized_by
mradermacher
mradermacher
mradermacher.quantized_on
rich1
rich1
Tensor
Name
Type
Shape
token_embd.weight
Q8_0
Q8_0
[3584, 102400]
blk.0
blk.0.attn_k.bias
F32
F32
[512]
blk.0.attn_k.weight
Q8_0
Q8_0
[3584, 512]
blk.0.attn_norm.weight
F32
F32
[3584]
blk.0.attn_output.weight
Q8_0
Q8_0
[3584, 3584]
blk.0.attn_q.bias
F32
F32
[3584]
blk.0.attn_q.weight
Q8_0
Q8_0
[3584, 3584]
blk.0.attn_v.bias
F32
F32
[512]
blk.0.attn_v.weight
Q8_0
Q8_0
[3584, 512]
blk.0.ffn_down.weight
Q8_0
Q8_0
[18944, 3584]
blk.0.ffn_gate.weight
Q8_0
Q8_0
[3584, 18944]
blk.0.ffn_norm.weight
F32
F32
[3584]
blk.0.ffn_up.weight
Q8_0
Q8_0
[3584, 18944]
blk.1
blk.1.attn_k.bias
F32
F32
[512]
blk.1.attn_k.weight
Q8_0
Q8_0
[3584, 512]
blk.1.attn_norm.weight
F32
F32
[3584]
blk.1.attn_output.weight
Q8_0
Q8_0
[3584, 3584]
blk.1.attn_q.bias
F32
F32
[3584]
blk.1.attn_q.weight
Q8_0
Q8_0
[3584, 3584]
blk.1.attn_v.bias
F32
F32
[512]
blk.1.attn_v.weight
Q8_0
Q8_0
[3584, 512]
blk.1.ffn_down.weight
Q8_0
Q8_0
[18944, 3584]
blk.1.ffn_gate.weight
Q8_0
Q8_0
[3584, 18944]
blk.1.ffn_norm.weight
F32
F32
[3584]
blk.1.ffn_up.weight
Q8_0
Q8_0
[3584, 18944]
blk.2
blk.2.attn_k.bias
F32
F32
[512]
blk.2.attn_k.weight
Q8_0
Q8_0
[3584, 512]
blk.2.attn_norm.weight
F32
F32
[3584]
blk.2.attn_output.weight
Q8_0
Q8_0
[3584, 3584]
blk.2.attn_q.bias
F32
F32
[3584]
blk.2.attn_q.weight
Q8_0
Q8_0
[3584, 3584]
blk.2.attn_v.bias
F32
F32
[512]
blk.2.attn_v.weight
Q8_0
Q8_0
[3584, 512]
blk.2.ffn_down.weight
Q8_0
Q8_0
[18944, 3584]
blk.2.ffn_gate.weight
Q8_0
Q8_0
[3584, 18944]
blk.2.ffn_norm.weight
F32
F32
[3584]
blk.2.ffn_up.weight
Q8_0
Q8_0
[3584, 18944]
blk.3
blk.3.attn_k.bias
F32
F32
[512]
blk.3.attn_k.weight
Q8_0
Q8_0
[3584, 512]
blk.3.attn_norm.weight
F32
F32
[3584]
blk.3.attn_output.weight
Q8_0
Q8_0
[3584, 3584]
blk.3.attn_q.bias
F32
F32
[3584]
blk.3.attn_q.weight
Q8_0
Q8_0
[3584, 3584]
blk.3.attn_v.bias
F32
F32
[512]
blk.3.attn_v.weight
Q8_0
Q8_0
[3584, 512]
blk.3.ffn_down.weight
Q8_0
Q8_0
[18944, 3584]
blk.3.ffn_gate.weight
Q8_0
Q8_0
[3584, 18944]
blk.3.ffn_norm.weight
F32
F32
[3584]
blk.3.ffn_up.weight
Q8_0
Q8_0
[3584, 18944]
blk.4
blk.4.attn_k.bias
F32
F32
[512]
blk.4.attn_k.weight
Q8_0
Q8_0
[3584, 512]
blk.4.attn_norm.weight
F32
F32
[3584]
blk.4.attn_output.weight
Q8_0
Q8_0
[3584, 3584]
blk.4.attn_q.bias
F32
F32
[3584]
blk.4.attn_q.weight
Q8_0
Q8_0
[3584, 3584]
blk.4.attn_v.bias
F32
F32
[512]
blk.4.attn_v.weight
Q8_0
Q8_0
[3584, 512]
blk.4.ffn_down.weight
Q8_0
Q8_0
[18944, 3584]
blk.4.ffn_gate.weight
Q8_0
Q8_0
[3584, 18944]
blk.4.ffn_norm.weight
F32
F32
[3584]
blk.4.ffn_up.weight
Q8_0
Q8_0
[3584, 18944]
blk.5
blk.5.attn_k.bias
F32
F32
[512]
blk.5.attn_k.weight
Q8_0
Q8_0
[3584, 512]
blk.5.attn_norm.weight
F32
F32
[3584]
blk.5.attn_output.weight
Q8_0
Q8_0
[3584, 3584]
blk.5.attn_q.bias
F32
F32
[3584]
blk.5.attn_q.weight
Q8_0
Q8_0
[3584, 3584]
blk.5.attn_v.bias
F32
F32
[512]
blk.5.attn_v.weight
Q8_0
Q8_0
[3584, 512]
blk.5.ffn_down.weight
Q8_0
Q8_0
[18944, 3584]
blk.5.ffn_gate.weight
Q8_0
Q8_0
[3584, 18944]
blk.5.ffn_norm.weight
F32
F32
[3584]
blk.5.ffn_up.weight
Q8_0
Q8_0
[3584, 18944]
blk.6
blk.6.attn_k.bias
F32
F32
[512]
blk.6.attn_k.weight
Q8_0
Q8_0
[3584, 512]
blk.6.attn_norm.weight
F32
F32
[3584]
blk.6.attn_output.weight
Q8_0
Q8_0
[3584, 3584]
blk.6.attn_q.bias
F32
F32
[3584]
blk.6.attn_q.weight
Q8_0
Q8_0
[3584, 3584]
blk.6.attn_v.bias
F32
F32
[512]
blk.6.attn_v.weight
Q8_0
Q8_0
[3584, 512]
blk.6.ffn_down.weight
Q8_0
Q8_0
[18944, 3584]
blk.6.ffn_gate.weight
Q8_0
Q8_0
[3584, 18944]
blk.6.ffn_norm.weight
F32
F32
[3584]
blk.6.ffn_up.weight
Q8_0
Q8_0
[3584, 18944]
blk.7
blk.7.attn_k.bias
F32
F32
[512]
blk.7.attn_k.weight
Q8_0
Q8_0
[3584, 512]
blk.7.attn_norm.weight
F32
F32
[3584]
blk.7.attn_output.weight
Q8_0
Q8_0
[3584, 3584]
blk.7.attn_q.bias
F32
F32
[3584]
blk.7.attn_q.weight
Q8_0
Q8_0
[3584, 3584]
blk.7.attn_v.bias
F32
F32
[512]
blk.7.attn_v.weight
Q8_0
Q8_0
[3584, 512]
blk.7.ffn_down.weight
Q8_0
Q8_0
[18944, 3584]
blk.7.ffn_gate.weight
Q8_0
Q8_0
[3584, 18944]
blk.7.ffn_norm.weight
F32
F32
[3584]
blk.7.ffn_up.weight
Q8_0
Q8_0
[3584, 18944]
blk.8
blk.8.attn_k.bias
F32
F32
[512]
blk.8.attn_k.weight
Q8_0
Q8_0
[3584, 512]
blk.8.attn_norm.weight
F32
F32
[3584]
blk.8.attn_output.weight
Q8_0
Q8_0
[3584, 3584]
blk.8.attn_q.bias
F32
F32
[3584]
blk.8.attn_q.weight
Q8_0
Q8_0
[3584, 3584]
blk.8.attn_v.bias
F32
F32
[512]
blk.8.attn_v.weight
Q8_0
Q8_0
[3584, 512]
blk.8.ffn_down.weight
Q8_0
Q8_0
[18944, 3584]
blk.8.ffn_gate.weight
Q8_0
Q8_0
[3584, 18944]
blk.8.ffn_norm.weight
F32
F32
[3584]
blk.8.ffn_up.weight
Q8_0
Q8_0
[3584, 18944]
blk.9
blk.9.attn_k.bias
F32
F32
[512]
blk.9.attn_k.weight
Q8_0
Q8_0
[3584, 512]
blk.9.attn_norm.weight
F32
F32
[3584]
blk.9.attn_output.weight
Q8_0
Q8_0
[3584, 3584]
blk.9.attn_q.bias
F32
F32
[3584]
blk.9.attn_q.weight
Q8_0
Q8_0
[3584, 3584]
blk.9.attn_v.bias
F32
F32
[512]
blk.9.attn_v.weight
Q8_0
Q8_0
[3584, 512]
blk.9.ffn_down.weight
Q8_0
Q8_0
[18944, 3584]
blk.9.ffn_gate.weight
Q8_0
Q8_0
[3584, 18944]
blk.9.ffn_norm.weight
F32
F32
[3584]
blk.9.ffn_up.weight
Q8_0
Q8_0
[3584, 18944]
blk.10
blk.10.attn_k.bias
F32
F32
[512]
blk.10.attn_k.weight
Q8_0
Q8_0
[3584, 512]
blk.10.attn_norm.weight
F32
F32
[3584]
blk.10.attn_output.weight
Q8_0
Q8_0
[3584, 3584]
blk.10.attn_q.bias
F32
F32
[3584]
blk.10.attn_q.weight
Q8_0
Q8_0
[3584, 3584]
blk.10.attn_v.bias
F32
F32
[512]
blk.10.attn_v.weight
Q8_0
Q8_0
[3584, 512]
blk.10.ffn_down.weight
Q8_0
Q8_0
[18944, 3584]
blk.10.ffn_gate.weight
Q8_0
Q8_0
[3584, 18944]
blk.10.ffn_norm.weight
F32
F32
[3584]
blk.10.ffn_up.weight
Q8_0
Q8_0
[3584, 18944]
blk.11
blk.11.attn_k.bias
F32
F32
[512]
blk.11.attn_k.weight
Q8_0
Q8_0
[3584, 512]
blk.11.attn_norm.weight
F32
F32
[3584]
blk.11.attn_output.weight
Q8_0
Q8_0
[3584, 3584]
blk.11.attn_q.bias
F32
F32
[3584]
blk.11.attn_q.weight
Q8_0
Q8_0
[3584, 3584]
blk.11.attn_v.bias
F32
F32
[512]
blk.11.attn_v.weight
Q8_0
Q8_0
[3584, 512]
blk.11.ffn_down.weight
Q8_0
Q8_0
[18944, 3584]
blk.11.ffn_gate.weight
Q8_0
Q8_0
[3584, 18944]
blk.11.ffn_norm.weight
F32
F32
[3584]
blk.11.ffn_up.weight
Q8_0
Q8_0
[3584, 18944]
blk.12
blk.12.attn_k.bias
F32
F32
[512]
blk.12.attn_k.weight
Q8_0
Q8_0
[3584, 512]
blk.12.attn_norm.weight
F32
F32
[3584]
blk.12.attn_output.weight
Q8_0
Q8_0
[3584, 3584]
blk.12.attn_q.bias
F32
F32
[3584]
blk.12.attn_q.weight
Q8_0
Q8_0
[3584, 3584]
blk.12.attn_v.bias
F32
F32
[512]
blk.12.attn_v.weight
Q8_0
Q8_0
[3584, 512]
blk.12.ffn_down.weight
Q8_0
Q8_0
[18944, 3584]
blk.12.ffn_gate.weight
Q8_0
Q8_0
[3584, 18944]
blk.12.ffn_norm.weight
F32
F32
[3584]
blk.12.ffn_up.weight
Q8_0
Q8_0
[3584, 18944]
blk.13
blk.13.attn_k.bias
F32
F32
[512]
blk.13.attn_k.weight
Q8_0
Q8_0
[3584, 512]
blk.13.attn_norm.weight
F32
F32
[3584]
blk.13.attn_output.weight
Q8_0
Q8_0
[3584, 3584]
blk.13.attn_q.bias
F32
F32
[3584]
blk.13.attn_q.weight
Q8_0
Q8_0
[3584, 3584]
blk.13.attn_v.bias
F32
F32
[512]
blk.13.attn_v.weight
Q8_0
Q8_0
[3584, 512]
blk.13.ffn_down.weight
Q8_0
Q8_0
[18944, 3584]
blk.13.ffn_gate.weight
Q8_0
Q8_0
[3584, 18944]
blk.13.ffn_norm.weight
F32
F32
[3584]
blk.13.ffn_up.weight
Q8_0
Q8_0
[3584, 18944]
blk.14
blk.14.attn_k.bias
F32
F32
[512]
blk.14.attn_k.weight
Q8_0
Q8_0
[3584, 512]
blk.14.attn_norm.weight
F32
F32
[3584]
blk.14.attn_output.weight
Q8_0
Q8_0
[3584, 3584]
blk.14.attn_q.bias
F32
F32
[3584]
blk.14.attn_q.weight
Q8_0
Q8_0
[3584, 3584]
blk.14.attn_v.bias
F32
F32
[512]
blk.14.attn_v.weight
Q8_0
Q8_0
[3584, 512]
blk.14.ffn_down.weight
Q8_0
Q8_0
[18944, 3584]
blk.14.ffn_gate.weight
Q8_0
Q8_0
[3584, 18944]
blk.14.ffn_norm.weight
F32
F32
[3584]
blk.14.ffn_up.weight
Q8_0
Q8_0
[3584, 18944]
blk.15
blk.15.attn_k.bias
F32
F32
[512]
blk.15.attn_k.weight
Q8_0
Q8_0
[3584, 512]
blk.15.attn_norm.weight
F32
F32
[3584]
blk.15.attn_output.weight
Q8_0
Q8_0
[3584, 3584]
blk.15.attn_q.bias
F32
F32
[3584]
blk.15.attn_q.weight
Q8_0
Q8_0
[3584, 3584]
blk.15.attn_v.bias
F32
F32
[512]
blk.15.attn_v.weight
Q8_0
Q8_0
[3584, 512]
blk.15.ffn_down.weight
Q8_0
Q8_0
[18944, 3584]
blk.15.ffn_gate.weight
Q8_0
Q8_0
[3584, 18944]
blk.15.ffn_norm.weight
F32
F32
[3584]
blk.15.ffn_up.weight
Q8_0
Q8_0
[3584, 18944]
blk.16
blk.16.attn_k.bias
F32
F32
[512]
blk.16.attn_k.weight
Q8_0
Q8_0
[3584, 512]
blk.16.attn_norm.weight
F32
F32
[3584]
blk.16.attn_output.weight
Q8_0
Q8_0
[3584, 3584]
blk.16.attn_q.bias
F32
F32
[3584]
blk.16.attn_q.weight
Q8_0
Q8_0
[3584, 3584]
blk.16.attn_v.bias
F32
F32
[512]
blk.16.attn_v.weight
Q8_0
Q8_0
[3584, 512]
blk.16.ffn_down.weight
Q8_0
Q8_0
[18944, 3584]
blk.16.ffn_gate.weight
Q8_0
Q8_0
[3584, 18944]
blk.16.ffn_norm.weight
F32
F32
[3584]
blk.16.ffn_up.weight
Q8_0
Q8_0
[3584, 18944]
blk.17
blk.17.attn_k.bias
F32
F32
[512]
blk.17.attn_k.weight
Q8_0
Q8_0
[3584, 512]
blk.17.attn_norm.weight
F32
F32
[3584]
blk.17.attn_output.weight
Q8_0
Q8_0
[3584, 3584]
blk.17.attn_q.bias
F32
F32
[3584]
blk.17.attn_q.weight
Q8_0
Q8_0
[3584, 3584]
blk.17.attn_v.bias
F32
F32
[512]
blk.17.attn_v.weight
Q8_0
Q8_0
[3584, 512]
blk.17.ffn_down.weight
Q8_0
Q8_0
[18944, 3584]
blk.17.ffn_gate.weight
Q8_0
Q8_0
[3584, 18944]
blk.17.ffn_norm.weight
F32
F32
[3584]
blk.17.ffn_up.weight
Q8_0
Q8_0
[3584, 18944]
blk.18
blk.18.attn_k.bias
F32
F32
[512]
blk.18.attn_k.weight
Q8_0
Q8_0
[3584, 512]
blk.18.attn_norm.weight
F32
F32
[3584]
blk.18.attn_output.weight
Q8_0
Q8_0
[3584, 3584]
blk.18.attn_q.bias
F32
F32
[3584]
blk.18.attn_q.weight
Q8_0
Q8_0
[3584, 3584]
blk.18.attn_v.bias
F32
F32
[512]
blk.18.attn_v.weight
Q8_0
Q8_0
[3584, 512]
blk.18.ffn_down.weight
Q8_0
Q8_0
[18944, 3584]
blk.18.ffn_gate.weight
Q8_0
Q8_0
[3584, 18944]
blk.18.ffn_norm.weight
F32
F32
[3584]
blk.18.ffn_up.weight
Q8_0
Q8_0
[3584, 18944]
blk.19
blk.19.attn_k.bias
F32
F32
[512]
blk.19.attn_k.weight
Q8_0
Q8_0
[3584, 512]
blk.19.attn_norm.weight
F32
F32
[3584]
blk.19.attn_output.weight
Q8_0
Q8_0
[3584, 3584]
blk.19.attn_q.bias
F32
F32
[3584]
blk.19.attn_q.weight
Q8_0
Q8_0
[3584, 3584]
blk.19.attn_v.bias
F32
F32
[512]
blk.19.attn_v.weight
Q8_0
Q8_0
[3584, 512]
blk.19.ffn_down.weight
Q8_0
Q8_0
[18944, 3584]
blk.19.ffn_gate.weight
Q8_0
Q8_0
[3584, 18944]
blk.19.ffn_norm.weight
F32
F32
[3584]
blk.19.ffn_up.weight
Q8_0
Q8_0
[3584, 18944]
blk.20
blk.20.attn_k.bias
F32
F32
[512]
blk.20.attn_k.weight
Q8_0
Q8_0
[3584, 512]
blk.20.attn_norm.weight
F32
F32
[3584]
blk.20.attn_output.weight
Q8_0
Q8_0
[3584, 3584]
blk.20.attn_q.bias
F32
F32
[3584]
blk.20.attn_q.weight
Q8_0
Q8_0
[3584, 3584]
blk.20.attn_v.bias
F32
F32
[512]
blk.20.attn_v.weight
Q8_0
Q8_0
[3584, 512]
blk.20.ffn_down.weight
Q8_0
Q8_0
[18944, 3584]
blk.20.ffn_gate.weight
Q8_0
Q8_0
[3584, 18944]
blk.20.ffn_norm.weight
F32
F32
[3584]
blk.20.ffn_up.weight
Q8_0
Q8_0
[3584, 18944]
blk.21
blk.21.attn_k.bias
F32
F32
[512]
blk.21.attn_k.weight
Q8_0
Q8_0
[3584, 512]
blk.21.attn_norm.weight
F32
F32
[3584]
blk.21.attn_output.weight
Q8_0
Q8_0
[3584, 3584]
blk.21.attn_q.bias
F32
F32
[3584]
blk.21.attn_q.weight
Q8_0
Q8_0
[3584, 3584]
blk.21.attn_v.bias
F32
F32
[512]
blk.21.attn_v.weight
Q8_0
Q8_0
[3584, 512]
blk.21.ffn_down.weight
Q8_0
Q8_0
[18944, 3584]
blk.21.ffn_gate.weight
Q8_0
Q8_0
[3584, 18944]
blk.21.ffn_norm.weight
F32
F32
[3584]
blk.21.ffn_up.weight
Q8_0
Q8_0
[3584, 18944]
blk.22
blk.22.attn_k.bias
F32
F32
[512]
blk.22.attn_k.weight
Q8_0
Q8_0
[3584, 512]
blk.22.attn_norm.weight
F32
F32
[3584]
blk.22.attn_output.weight
Q8_0
Q8_0
[3584, 3584]
blk.22.attn_q.bias
F32
F32
[3584]
blk.22.attn_q.weight
Q8_0
Q8_0
[3584, 3584]
blk.22.attn_v.bias
F32
F32
[512]
blk.22.attn_v.weight
Q8_0
Q8_0
[3584, 512]
blk.22.ffn_down.weight
Q8_0
Q8_0
[18944, 3584]
blk.22.ffn_gate.weight
Q8_0
Q8_0
[3584, 18944]
blk.22.ffn_norm.weight
F32
F32
[3584]
blk.22.ffn_up.weight
Q8_0
Q8_0
[3584, 18944]
blk.23
blk.23.attn_k.bias
F32
F32
[512]
blk.23.attn_k.weight
Q8_0
Q8_0
[3584, 512]
blk.23.attn_norm.weight
F32
F32
[3584]
blk.23.attn_output.weight
Q8_0
Q8_0
[3584, 3584]
blk.23.attn_q.bias
F32
F32
[3584]
blk.23.attn_q.weight
Q8_0
Q8_0
[3584, 3584]
blk.23.attn_v.bias
F32
F32
[512]
blk.23.attn_v.weight
Q8_0
Q8_0
[3584, 512]
blk.23.ffn_down.weight
Q8_0
Q8_0
[18944, 3584]
blk.23.ffn_gate.weight
Q8_0
Q8_0
[3584, 18944]
blk.23.ffn_norm.weight
F32
F32
[3584]
blk.23.ffn_up.weight
Q8_0
Q8_0
[3584, 18944]
blk.24
blk.24.attn_k.bias
F32
F32
[512]
blk.24.attn_k.weight
Q8_0
Q8_0
[3584, 512]
blk.24.attn_norm.weight
F32
F32
[3584]
blk.24.attn_output.weight
Q8_0
Q8_0
[3584, 3584]
blk.24.attn_q.bias
F32
F32
[3584]
blk.24.attn_q.weight
Q8_0
Q8_0
[3584, 3584]
blk.24.attn_v.bias
F32
F32
[512]
blk.24.attn_v.weight
Q8_0
Q8_0
[3584, 512]
blk.24.ffn_down.weight
Q8_0
Q8_0
[18944, 3584]
blk.24.ffn_gate.weight
Q8_0
Q8_0
[3584, 18944]
blk.24.ffn_norm.weight
F32
F32
[3584]
blk.24.ffn_up.weight
Q8_0
Q8_0
[3584, 18944]
blk.25
blk.25.attn_k.bias
F32
F32
[512]
blk.25.attn_k.weight
Q8_0
Q8_0
[3584, 512]
blk.25.attn_norm.weight
F32
F32
[3584]
blk.25.attn_output.weight
Q8_0
Q8_0
[3584, 3584]
blk.25.attn_q.bias
F32
F32
[3584]
blk.25.attn_q.weight
Q8_0
Q8_0
[3584, 3584]
blk.25.attn_v.bias
F32
F32
[512]
blk.25.attn_v.weight
Q8_0
Q8_0
[3584, 512]
blk.25.ffn_down.weight
Q8_0
Q8_0
[18944, 3584]
blk.25.ffn_gate.weight
Q8_0
Q8_0
[3584, 18944]
blk.25.ffn_norm.weight
F32
F32
[3584]
blk.25.ffn_up.weight
Q8_0
Q8_0
[3584, 18944]
blk.26
blk.26.attn_k.bias
F32
F32
[512]
blk.26.attn_k.weight
Q8_0
Q8_0
[3584, 512]
blk.26.attn_norm.weight
F32
F32
[3584]
blk.26.attn_output.weight
Q8_0
Q8_0
[3584, 3584]
blk.26.attn_q.bias
F32
F32
[3584]
blk.26.attn_q.weight
Q8_0
Q8_0
[3584, 3584]
blk.26.attn_v.bias
F32
F32
[512]
blk.26.attn_v.weight
Q8_0
Q8_0
[3584, 512]
blk.26.ffn_down.weight
Q8_0
Q8_0
[18944, 3584]
blk.26.ffn_gate.weight
Q8_0
Q8_0
[3584, 18944]
blk.26.ffn_norm.weight
F32
F32
[3584]
blk.26.ffn_up.weight
Q8_0
Q8_0
[3584, 18944]
blk.27
blk.27.attn_k.bias
F32
F32
[512]
blk.27.attn_k.weight
Q8_0
Q8_0
[3584, 512]
blk.27.attn_norm.weight
F32
F32
[3584]
blk.27.attn_output.weight
Q8_0
Q8_0
[3584, 3584]
blk.27.attn_q.bias
F32
F32
[3584]
blk.27.attn_q.weight
Q8_0
Q8_0
[3584, 3584]
blk.27.attn_v.bias
F32
F32
[512]
blk.27.attn_v.weight
Q8_0
Q8_0
[3584, 512]
blk.27.ffn_down.weight
Q8_0
Q8_0
[18944, 3584]
blk.27.ffn_gate.weight
Q8_0
Q8_0
[3584, 18944]
blk.27.ffn_norm.weight
F32
F32
[3584]
blk.27.ffn_up.weight
Q8_0
Q8_0
[3584, 18944]
output.weight
Q8_0
Q8_0
[3584, 102400]
output_norm.weight
F32
F32
[3584]