GitHub
183.7K
Download
Models
Docs
Pricing
Sign in
Download
Models
Docs
Pricing
Sign in
maternion
/
LightOnOCR-3
:1b-q8_0
171
Downloads
Updated
6 days ago
Best OCR model. LightOnOCR-3 is our flagship OCR model family 4B, 1B, and 0.8B with layout, grounding, and visual understanding. We recommend the 4B for most OCR tasks.
Best OCR model. LightOnOCR-3 is our flagship OCR model family 4B, 1B, and 0.8B with layout, grounding, and visual understanding. We recommend the 4B for most OCR tasks.
Cancel
vision
tools
thinking
0.8b
1b
4b
LightOnOCR-3:1b-q8_0
...
/
model
2cf9771aaf23 · 639MB
Metadata
general.architecture
qwen3
qwen3
general.file_type
Q8_0
Q8_0
qwen3.attention.head_count
16
16
qwen3.attention.head_count_kv
8
8
qwen3.attention.key_length
128
128
qwen3.attention.layer_norm_rms_epsilon
1e-06
1e-06
qwen3.attention.value_length
128
128
qwen3.block_count
28
28
qwen3.context_length
16384
16384
qwen3.embedding_length
1024
1024
qwen3.feed_forward_length
3072
3072
qwen3.rope.freq_base
1e+06
1e+06
tokenizer.ggml.add_bos_token
false
false
tokenizer.ggml.add_eos_token
false
false
tokenizer.ggml.eos_token_id
151645
151645
tokenizer.ggml.merges
[Ġ Ġ, ĠĠ ĠĠ, i n, Ġ t, ĠĠĠĠ ĠĠĠĠ, ...]
[Ġ Ġ, ĠĠ ĠĠ, i n, Ġ t, ĠĠĠĠ ĠĠĠĠ, ...]
tokenizer.ggml.model
gpt2
gpt2
tokenizer.ggml.padding_token_id
151643
151643
tokenizer.ggml.pre
qwen2
qwen2
tokenizer.ggml.token_type
[1, 1, 1, 1, 1, ...]
[1, 1, 1, 1, 1, ...]
tokenizer.ggml.tokens
[!, ", #, $, %, ...]
[!, ", #, $, %, ...]
Tensor
Name
Type
Shape
token_embd.weight
Q8_0
Q8_0
[1024, 151936]
blk.0
blk.0.attn_k.weight
Q8_0
Q8_0
[1024, 1024]
blk.0.attn_k_norm.weight
F32
F32
[128]
blk.0.attn_norm.weight
F32
F32
[1024]
blk.0.attn_output.weight
Q8_0
Q8_0
[2048, 1024]
blk.0.attn_q.weight
Q8_0
Q8_0
[1024, 2048]
blk.0.attn_q_norm.weight
F32
F32
[128]
blk.0.attn_v.weight
Q8_0
Q8_0
[1024, 1024]
blk.0.ffn_down.weight
Q8_0
Q8_0
[3072, 1024]
blk.0.ffn_gate.weight
Q8_0
Q8_0
[1024, 3072]
blk.0.ffn_norm.weight
F32
F32
[1024]
blk.0.ffn_up.weight
Q8_0
Q8_0
[1024, 3072]
blk.1
blk.1.attn_k.weight
Q8_0
Q8_0
[1024, 1024]
blk.1.attn_k_norm.weight
F32
F32
[128]
blk.1.attn_norm.weight
F32
F32
[1024]
blk.1.attn_output.weight
Q8_0
Q8_0
[2048, 1024]
blk.1.attn_q.weight
Q8_0
Q8_0
[1024, 2048]
blk.1.attn_q_norm.weight
F32
F32
[128]
blk.1.attn_v.weight
Q8_0
Q8_0
[1024, 1024]
blk.1.ffn_down.weight
Q8_0
Q8_0
[3072, 1024]
blk.1.ffn_gate.weight
Q8_0
Q8_0
[1024, 3072]
blk.1.ffn_norm.weight
F32
F32
[1024]
blk.1.ffn_up.weight
Q8_0
Q8_0
[1024, 3072]
blk.2
blk.2.attn_k.weight
Q8_0
Q8_0
[1024, 1024]
blk.2.attn_k_norm.weight
F32
F32
[128]
blk.2.attn_norm.weight
F32
F32
[1024]
blk.2.attn_output.weight
Q8_0
Q8_0
[2048, 1024]
blk.2.attn_q.weight
Q8_0
Q8_0
[1024, 2048]
blk.2.attn_q_norm.weight
F32
F32
[128]
blk.2.attn_v.weight
Q8_0
Q8_0
[1024, 1024]
blk.2.ffn_down.weight
Q8_0
Q8_0
[3072, 1024]
blk.2.ffn_gate.weight
Q8_0
Q8_0
[1024, 3072]
blk.2.ffn_norm.weight
F32
F32
[1024]
blk.2.ffn_up.weight
Q8_0
Q8_0
[1024, 3072]
blk.3
blk.3.attn_k.weight
Q8_0
Q8_0
[1024, 1024]
blk.3.attn_k_norm.weight
F32
F32
[128]
blk.3.attn_norm.weight
F32
F32
[1024]
blk.3.attn_output.weight
Q8_0
Q8_0
[2048, 1024]
blk.3.attn_q.weight
Q8_0
Q8_0
[1024, 2048]
blk.3.attn_q_norm.weight
F32
F32
[128]
blk.3.attn_v.weight
Q8_0
Q8_0
[1024, 1024]
blk.3.ffn_down.weight
Q8_0
Q8_0
[3072, 1024]
blk.3.ffn_gate.weight
Q8_0
Q8_0
[1024, 3072]
blk.3.ffn_norm.weight
F32
F32
[1024]
blk.3.ffn_up.weight
Q8_0
Q8_0
[1024, 3072]
blk.4
blk.4.attn_k.weight
Q8_0
Q8_0
[1024, 1024]
blk.4.attn_k_norm.weight
F32
F32
[128]
blk.4.attn_norm.weight
F32
F32
[1024]
blk.4.attn_output.weight
Q8_0
Q8_0
[2048, 1024]
blk.4.attn_q.weight
Q8_0
Q8_0
[1024, 2048]
blk.4.attn_q_norm.weight
F32
F32
[128]
blk.4.attn_v.weight
Q8_0
Q8_0
[1024, 1024]
blk.4.ffn_down.weight
Q8_0
Q8_0
[3072, 1024]
blk.4.ffn_gate.weight
Q8_0
Q8_0
[1024, 3072]
blk.4.ffn_norm.weight
F32
F32
[1024]
blk.4.ffn_up.weight
Q8_0
Q8_0
[1024, 3072]
blk.5
blk.5.attn_k.weight
Q8_0
Q8_0
[1024, 1024]
blk.5.attn_k_norm.weight
F32
F32
[128]
blk.5.attn_norm.weight
F32
F32
[1024]
blk.5.attn_output.weight
Q8_0
Q8_0
[2048, 1024]
blk.5.attn_q.weight
Q8_0
Q8_0
[1024, 2048]
blk.5.attn_q_norm.weight
F32
F32
[128]
blk.5.attn_v.weight
Q8_0
Q8_0
[1024, 1024]
blk.5.ffn_down.weight
Q8_0
Q8_0
[3072, 1024]
blk.5.ffn_gate.weight
Q8_0
Q8_0
[1024, 3072]
blk.5.ffn_norm.weight
F32
F32
[1024]
blk.5.ffn_up.weight
Q8_0
Q8_0
[1024, 3072]
blk.6
blk.6.attn_k.weight
Q8_0
Q8_0
[1024, 1024]
blk.6.attn_k_norm.weight
F32
F32
[128]
blk.6.attn_norm.weight
F32
F32
[1024]
blk.6.attn_output.weight
Q8_0
Q8_0
[2048, 1024]
blk.6.attn_q.weight
Q8_0
Q8_0
[1024, 2048]
blk.6.attn_q_norm.weight
F32
F32
[128]
blk.6.attn_v.weight
Q8_0
Q8_0
[1024, 1024]
blk.6.ffn_down.weight
Q8_0
Q8_0
[3072, 1024]
blk.6.ffn_gate.weight
Q8_0
Q8_0
[1024, 3072]
blk.6.ffn_norm.weight
F32
F32
[1024]
blk.6.ffn_up.weight
Q8_0
Q8_0
[1024, 3072]
blk.7
blk.7.attn_k.weight
Q8_0
Q8_0
[1024, 1024]
blk.7.attn_k_norm.weight
F32
F32
[128]
blk.7.attn_norm.weight
F32
F32
[1024]
blk.7.attn_output.weight
Q8_0
Q8_0
[2048, 1024]
blk.7.attn_q.weight
Q8_0
Q8_0
[1024, 2048]
blk.7.attn_q_norm.weight
F32
F32
[128]
blk.7.attn_v.weight
Q8_0
Q8_0
[1024, 1024]
blk.7.ffn_down.weight
Q8_0
Q8_0
[3072, 1024]
blk.7.ffn_gate.weight
Q8_0
Q8_0
[1024, 3072]
blk.7.ffn_norm.weight
F32
F32
[1024]
blk.7.ffn_up.weight
Q8_0
Q8_0
[1024, 3072]
blk.8
blk.8.attn_k.weight
Q8_0
Q8_0
[1024, 1024]
blk.8.attn_k_norm.weight
F32
F32
[128]
blk.8.attn_norm.weight
F32
F32
[1024]
blk.8.attn_output.weight
Q8_0
Q8_0
[2048, 1024]
blk.8.attn_q.weight
Q8_0
Q8_0
[1024, 2048]
blk.8.attn_q_norm.weight
F32
F32
[128]
blk.8.attn_v.weight
Q8_0
Q8_0
[1024, 1024]
blk.8.ffn_down.weight
Q8_0
Q8_0
[3072, 1024]
blk.8.ffn_gate.weight
Q8_0
Q8_0
[1024, 3072]
blk.8.ffn_norm.weight
F32
F32
[1024]
blk.8.ffn_up.weight
Q8_0
Q8_0
[1024, 3072]
blk.9
blk.9.attn_k.weight
Q8_0
Q8_0
[1024, 1024]
blk.9.attn_k_norm.weight
F32
F32
[128]
blk.9.attn_norm.weight
F32
F32
[1024]
blk.9.attn_output.weight
Q8_0
Q8_0
[2048, 1024]
blk.9.attn_q.weight
Q8_0
Q8_0
[1024, 2048]
blk.9.attn_q_norm.weight
F32
F32
[128]
blk.9.attn_v.weight
Q8_0
Q8_0
[1024, 1024]
blk.9.ffn_down.weight
Q8_0
Q8_0
[3072, 1024]
blk.9.ffn_gate.weight
Q8_0
Q8_0
[1024, 3072]
blk.9.ffn_norm.weight
F32
F32
[1024]
blk.9.ffn_up.weight
Q8_0
Q8_0
[1024, 3072]
blk.10
blk.10.attn_k.weight
Q8_0
Q8_0
[1024, 1024]
blk.10.attn_k_norm.weight
F32
F32
[128]
blk.10.attn_norm.weight
F32
F32
[1024]
blk.10.attn_output.weight
Q8_0
Q8_0
[2048, 1024]
blk.10.attn_q.weight
Q8_0
Q8_0
[1024, 2048]
blk.10.attn_q_norm.weight
F32
F32
[128]
blk.10.attn_v.weight
Q8_0
Q8_0
[1024, 1024]
blk.10.ffn_down.weight
Q8_0
Q8_0
[3072, 1024]
blk.10.ffn_gate.weight
Q8_0
Q8_0
[1024, 3072]
blk.10.ffn_norm.weight
F32
F32
[1024]
blk.10.ffn_up.weight
Q8_0
Q8_0
[1024, 3072]
blk.11
blk.11.attn_k.weight
Q8_0
Q8_0
[1024, 1024]
blk.11.attn_k_norm.weight
F32
F32
[128]
blk.11.attn_norm.weight
F32
F32
[1024]
blk.11.attn_output.weight
Q8_0
Q8_0
[2048, 1024]
blk.11.attn_q.weight
Q8_0
Q8_0
[1024, 2048]
blk.11.attn_q_norm.weight
F32
F32
[128]
blk.11.attn_v.weight
Q8_0
Q8_0
[1024, 1024]
blk.11.ffn_down.weight
Q8_0
Q8_0
[3072, 1024]
blk.11.ffn_gate.weight
Q8_0
Q8_0
[1024, 3072]
blk.11.ffn_norm.weight
F32
F32
[1024]
blk.11.ffn_up.weight
Q8_0
Q8_0
[1024, 3072]
blk.12
blk.12.attn_k.weight
Q8_0
Q8_0
[1024, 1024]
blk.12.attn_k_norm.weight
F32
F32
[128]
blk.12.attn_norm.weight
F32
F32
[1024]
blk.12.attn_output.weight
Q8_0
Q8_0
[2048, 1024]
blk.12.attn_q.weight
Q8_0
Q8_0
[1024, 2048]
blk.12.attn_q_norm.weight
F32
F32
[128]
blk.12.attn_v.weight
Q8_0
Q8_0
[1024, 1024]
blk.12.ffn_down.weight
Q8_0
Q8_0
[3072, 1024]
blk.12.ffn_gate.weight
Q8_0
Q8_0
[1024, 3072]
blk.12.ffn_norm.weight
F32
F32
[1024]
blk.12.ffn_up.weight
Q8_0
Q8_0
[1024, 3072]
blk.13
blk.13.attn_k.weight
Q8_0
Q8_0
[1024, 1024]
blk.13.attn_k_norm.weight
F32
F32
[128]
blk.13.attn_norm.weight
F32
F32
[1024]
blk.13.attn_output.weight
Q8_0
Q8_0
[2048, 1024]
blk.13.attn_q.weight
Q8_0
Q8_0
[1024, 2048]
blk.13.attn_q_norm.weight
F32
F32
[128]
blk.13.attn_v.weight
Q8_0
Q8_0
[1024, 1024]
blk.13.ffn_down.weight
Q8_0
Q8_0
[3072, 1024]
blk.13.ffn_gate.weight
Q8_0
Q8_0
[1024, 3072]
blk.13.ffn_norm.weight
F32
F32
[1024]
blk.13.ffn_up.weight
Q8_0
Q8_0
[1024, 3072]
blk.14
blk.14.attn_k.weight
Q8_0
Q8_0
[1024, 1024]
blk.14.attn_k_norm.weight
F32
F32
[128]
blk.14.attn_norm.weight
F32
F32
[1024]
blk.14.attn_output.weight
Q8_0
Q8_0
[2048, 1024]
blk.14.attn_q.weight
Q8_0
Q8_0
[1024, 2048]
blk.14.attn_q_norm.weight
F32
F32
[128]
blk.14.attn_v.weight
Q8_0
Q8_0
[1024, 1024]
blk.14.ffn_down.weight
Q8_0
Q8_0
[3072, 1024]
blk.14.ffn_gate.weight
Q8_0
Q8_0
[1024, 3072]
blk.14.ffn_norm.weight
F32
F32
[1024]
blk.14.ffn_up.weight
Q8_0
Q8_0
[1024, 3072]
blk.15
blk.15.attn_k.weight
Q8_0
Q8_0
[1024, 1024]
blk.15.attn_k_norm.weight
F32
F32
[128]
blk.15.attn_norm.weight
F32
F32
[1024]
blk.15.attn_output.weight
Q8_0
Q8_0
[2048, 1024]
blk.15.attn_q.weight
Q8_0
Q8_0
[1024, 2048]
blk.15.attn_q_norm.weight
F32
F32
[128]
blk.15.attn_v.weight
Q8_0
Q8_0
[1024, 1024]
blk.15.ffn_down.weight
Q8_0
Q8_0
[3072, 1024]
blk.15.ffn_gate.weight
Q8_0
Q8_0
[1024, 3072]
blk.15.ffn_norm.weight
F32
F32
[1024]
blk.15.ffn_up.weight
Q8_0
Q8_0
[1024, 3072]
blk.16
blk.16.attn_k.weight
Q8_0
Q8_0
[1024, 1024]
blk.16.attn_k_norm.weight
F32
F32
[128]
blk.16.attn_norm.weight
F32
F32
[1024]
blk.16.attn_output.weight
Q8_0
Q8_0
[2048, 1024]
blk.16.attn_q.weight
Q8_0
Q8_0
[1024, 2048]
blk.16.attn_q_norm.weight
F32
F32
[128]
blk.16.attn_v.weight
Q8_0
Q8_0
[1024, 1024]
blk.16.ffn_down.weight
Q8_0
Q8_0
[3072, 1024]
blk.16.ffn_gate.weight
Q8_0
Q8_0
[1024, 3072]
blk.16.ffn_norm.weight
F32
F32
[1024]
blk.16.ffn_up.weight
Q8_0
Q8_0
[1024, 3072]
blk.17
blk.17.attn_k.weight
Q8_0
Q8_0
[1024, 1024]
blk.17.attn_k_norm.weight
F32
F32
[128]
blk.17.attn_norm.weight
F32
F32
[1024]
blk.17.attn_output.weight
Q8_0
Q8_0
[2048, 1024]
blk.17.attn_q.weight
Q8_0
Q8_0
[1024, 2048]
blk.17.attn_q_norm.weight
F32
F32
[128]
blk.17.attn_v.weight
Q8_0
Q8_0
[1024, 1024]
blk.17.ffn_down.weight
Q8_0
Q8_0
[3072, 1024]
blk.17.ffn_gate.weight
Q8_0
Q8_0
[1024, 3072]
blk.17.ffn_norm.weight
F32
F32
[1024]
blk.17.ffn_up.weight
Q8_0
Q8_0
[1024, 3072]
blk.18
blk.18.attn_k.weight
Q8_0
Q8_0
[1024, 1024]
blk.18.attn_k_norm.weight
F32
F32
[128]
blk.18.attn_norm.weight
F32
F32
[1024]
blk.18.attn_output.weight
Q8_0
Q8_0
[2048, 1024]
blk.18.attn_q.weight
Q8_0
Q8_0
[1024, 2048]
blk.18.attn_q_norm.weight
F32
F32
[128]
blk.18.attn_v.weight
Q8_0
Q8_0
[1024, 1024]
blk.18.ffn_down.weight
Q8_0
Q8_0
[3072, 1024]
blk.18.ffn_gate.weight
Q8_0
Q8_0
[1024, 3072]
blk.18.ffn_norm.weight
F32
F32
[1024]
blk.18.ffn_up.weight
Q8_0
Q8_0
[1024, 3072]
blk.19
blk.19.attn_k.weight
Q8_0
Q8_0
[1024, 1024]
blk.19.attn_k_norm.weight
F32
F32
[128]
blk.19.attn_norm.weight
F32
F32
[1024]
blk.19.attn_output.weight
Q8_0
Q8_0
[2048, 1024]
blk.19.attn_q.weight
Q8_0
Q8_0
[1024, 2048]
blk.19.attn_q_norm.weight
F32
F32
[128]
blk.19.attn_v.weight
Q8_0
Q8_0
[1024, 1024]
blk.19.ffn_down.weight
Q8_0
Q8_0
[3072, 1024]
blk.19.ffn_gate.weight
Q8_0
Q8_0
[1024, 3072]
blk.19.ffn_norm.weight
F32
F32
[1024]
blk.19.ffn_up.weight
Q8_0
Q8_0
[1024, 3072]
blk.20
blk.20.attn_k.weight
Q8_0
Q8_0
[1024, 1024]
blk.20.attn_k_norm.weight
F32
F32
[128]
blk.20.attn_norm.weight
F32
F32
[1024]
blk.20.attn_output.weight
Q8_0
Q8_0
[2048, 1024]
blk.20.attn_q.weight
Q8_0
Q8_0
[1024, 2048]
blk.20.attn_q_norm.weight
F32
F32
[128]
blk.20.attn_v.weight
Q8_0
Q8_0
[1024, 1024]
blk.20.ffn_down.weight
Q8_0
Q8_0
[3072, 1024]
blk.20.ffn_gate.weight
Q8_0
Q8_0
[1024, 3072]
blk.20.ffn_norm.weight
F32
F32
[1024]
blk.20.ffn_up.weight
Q8_0
Q8_0
[1024, 3072]
blk.21
blk.21.attn_k.weight
Q8_0
Q8_0
[1024, 1024]
blk.21.attn_k_norm.weight
F32
F32
[128]
blk.21.attn_norm.weight
F32
F32
[1024]
blk.21.attn_output.weight
Q8_0
Q8_0
[2048, 1024]
blk.21.attn_q.weight
Q8_0
Q8_0
[1024, 2048]
blk.21.attn_q_norm.weight
F32
F32
[128]
blk.21.attn_v.weight
Q8_0
Q8_0
[1024, 1024]
blk.21.ffn_down.weight
Q8_0
Q8_0
[3072, 1024]
blk.21.ffn_gate.weight
Q8_0
Q8_0
[1024, 3072]
blk.21.ffn_norm.weight
F32
F32
[1024]
blk.21.ffn_up.weight
Q8_0
Q8_0
[1024, 3072]
blk.22
blk.22.attn_k.weight
Q8_0
Q8_0
[1024, 1024]
blk.22.attn_k_norm.weight
F32
F32
[128]
blk.22.attn_norm.weight
F32
F32
[1024]
blk.22.attn_output.weight
Q8_0
Q8_0
[2048, 1024]
blk.22.attn_q.weight
Q8_0
Q8_0
[1024, 2048]
blk.22.attn_q_norm.weight
F32
F32
[128]
blk.22.attn_v.weight
Q8_0
Q8_0
[1024, 1024]
blk.22.ffn_down.weight
Q8_0
Q8_0
[3072, 1024]
blk.22.ffn_gate.weight
Q8_0
Q8_0
[1024, 3072]
blk.22.ffn_norm.weight
F32
F32
[1024]
blk.22.ffn_up.weight
Q8_0
Q8_0
[1024, 3072]
blk.23
blk.23.attn_k.weight
Q8_0
Q8_0
[1024, 1024]
blk.23.attn_k_norm.weight
F32
F32
[128]
blk.23.attn_norm.weight
F32
F32
[1024]
blk.23.attn_output.weight
Q8_0
Q8_0
[2048, 1024]
blk.23.attn_q.weight
Q8_0
Q8_0
[1024, 2048]
blk.23.attn_q_norm.weight
F32
F32
[128]
blk.23.attn_v.weight
Q8_0
Q8_0
[1024, 1024]
blk.23.ffn_down.weight
Q8_0
Q8_0
[3072, 1024]
blk.23.ffn_gate.weight
Q8_0
Q8_0
[1024, 3072]
blk.23.ffn_norm.weight
F32
F32
[1024]
blk.23.ffn_up.weight
Q8_0
Q8_0
[1024, 3072]
blk.24
blk.24.attn_k.weight
Q8_0
Q8_0
[1024, 1024]
blk.24.attn_k_norm.weight
F32
F32
[128]
blk.24.attn_norm.weight
F32
F32
[1024]
blk.24.attn_output.weight
Q8_0
Q8_0
[2048, 1024]
blk.24.attn_q.weight
Q8_0
Q8_0
[1024, 2048]
blk.24.attn_q_norm.weight
F32
F32
[128]
blk.24.attn_v.weight
Q8_0
Q8_0
[1024, 1024]
blk.24.ffn_down.weight
Q8_0
Q8_0
[3072, 1024]
blk.24.ffn_gate.weight
Q8_0
Q8_0
[1024, 3072]
blk.24.ffn_norm.weight
F32
F32
[1024]
blk.24.ffn_up.weight
Q8_0
Q8_0
[1024, 3072]
blk.25
blk.25.attn_k.weight
Q8_0
Q8_0
[1024, 1024]
blk.25.attn_k_norm.weight
F32
F32
[128]
blk.25.attn_norm.weight
F32
F32
[1024]
blk.25.attn_output.weight
Q8_0
Q8_0
[2048, 1024]
blk.25.attn_q.weight
Q8_0
Q8_0
[1024, 2048]
blk.25.attn_q_norm.weight
F32
F32
[128]
blk.25.attn_v.weight
Q8_0
Q8_0
[1024, 1024]
blk.25.ffn_down.weight
Q8_0
Q8_0
[3072, 1024]
blk.25.ffn_gate.weight
Q8_0
Q8_0
[1024, 3072]
blk.25.ffn_norm.weight
F32
F32
[1024]
blk.25.ffn_up.weight
Q8_0
Q8_0
[1024, 3072]
blk.26
blk.26.attn_k.weight
Q8_0
Q8_0
[1024, 1024]
blk.26.attn_k_norm.weight
F32
F32
[128]
blk.26.attn_norm.weight
F32
F32
[1024]
blk.26.attn_output.weight
Q8_0
Q8_0
[2048, 1024]
blk.26.attn_q.weight
Q8_0
Q8_0
[1024, 2048]
blk.26.attn_q_norm.weight
F32
F32
[128]
blk.26.attn_v.weight
Q8_0
Q8_0
[1024, 1024]
blk.26.ffn_down.weight
Q8_0
Q8_0
[3072, 1024]
blk.26.ffn_gate.weight
Q8_0
Q8_0
[1024, 3072]
blk.26.ffn_norm.weight
F32
F32
[1024]
blk.26.ffn_up.weight
Q8_0
Q8_0
[1024, 3072]
blk.27
blk.27.attn_k.weight
Q8_0
Q8_0
[1024, 1024]
blk.27.attn_k_norm.weight
F32
F32
[128]
blk.27.attn_norm.weight
F32
F32
[1024]
blk.27.attn_output.weight
Q8_0
Q8_0
[2048, 1024]
blk.27.attn_q.weight
Q8_0
Q8_0
[1024, 2048]
blk.27.attn_q_norm.weight
F32
F32
[128]
blk.27.attn_v.weight
Q8_0
Q8_0
[1024, 1024]
blk.27.ffn_down.weight
Q8_0
Q8_0
[3072, 1024]
blk.27.ffn_gate.weight
Q8_0
Q8_0
[1024, 3072]
blk.27.ffn_norm.weight
F32
F32
[1024]
blk.27.ffn_up.weight
Q8_0
Q8_0
[1024, 3072]
output_norm.weight
F32
F32
[1024]