-
general.architecture
gemma2
-
general.file_type
Q3_K_M
-
gemma2.attention.head_count
8
-
gemma2.attention.head_count_kv
4
-
gemma2.attention.key_length
256
-
gemma2.attention.layer_norm_rms_epsilon
1e-06
-
gemma2.attention.sliding_window
4096
-
gemma2.attention.value_length
256
-
gemma2.attn_logit_softcapping
50
-
gemma2.block_count
26
-
gemma2.context_length
8192
-
gemma2.embedding_length
2304
-
gemma2.feed_forward_length
9216
-
gemma2.final_logit_softcapping
30
-
tokenizer.ggml.add_bos_token
true
-
tokenizer.ggml.add_eos_token
false
-
tokenizer.ggml.add_space_prefix
false
-
tokenizer.ggml.bos_token_id
2
-
tokenizer.ggml.eos_token_id
1
-
tokenizer.ggml.model
llama
-
tokenizer.ggml.padding_token_id
0
-
tokenizer.ggml.pre
default
-
tokenizer.ggml.scores
[-1000 -1000 -1000 -1000 -1000 ...]
-
tokenizer.ggml.token_type
[3 3 3 3 3 ...]
-
tokenizer.ggml.tokens
[<pad> <eos> <bos> <unk> <mask> ...]
-
tokenizer.ggml.unknown_token_id
3
-
Name
Type
Shape
-
token_embd.weight
Q6_K
[2304 256000]
-
blk.0.attn_norm.weight
F32
[2304]
-
blk.0.ffn_down.weight
Q5_K
[9216 2304]
-
blk.0.ffn_gate.weight
Q3_K
[2304 9216]
-
blk.0.ffn_up.weight
Q3_K
[2304 9216]
-
blk.0.post_attention_norm.weight
F32
[2304]
-
blk.0.post_ffw_norm.weight
F32
[2304]
-
blk.0.ffn_norm.weight
F32
[2304]
-
blk.0.attn_k.weight
Q3_K
[2304 1024]
-
blk.0.attn_output.weight
Q4_K
[2048 2304]
-
blk.0.attn_q.weight
Q3_K
[2304 2048]
-
blk.0.attn_v.weight
Q5_K
[2304 1024]
-
blk.1.attn_norm.weight
F32
[2304]
-
blk.1.ffn_down.weight
Q4_K
[9216 2304]
-
blk.1.ffn_gate.weight
Q3_K
[2304 9216]
-
blk.1.ffn_up.weight
Q3_K
[2304 9216]
-
blk.1.post_attention_norm.weight
F32
[2304]
-
blk.1.post_ffw_norm.weight
F32
[2304]
-
blk.1.ffn_norm.weight
F32
[2304]
-
blk.1.attn_k.weight
Q3_K
[2304 1024]
-
blk.1.attn_output.weight
Q4_K
[2048 2304]
-
blk.1.attn_q.weight
Q3_K
[2304 2048]
-
blk.1.attn_v.weight
Q5_K
[2304 1024]
-
blk.2.attn_norm.weight
F32
[2304]
-
blk.2.ffn_down.weight
Q4_K
[9216 2304]
-
blk.2.ffn_gate.weight
Q3_K
[2304 9216]
-
blk.2.ffn_up.weight
Q3_K
[2304 9216]
-
blk.2.post_attention_norm.weight
F32
[2304]
-
blk.2.post_ffw_norm.weight
F32
[2304]
-
blk.2.ffn_norm.weight
F32
[2304]
-
blk.2.attn_k.weight
Q3_K
[2304 1024]
-
blk.2.attn_output.weight
Q4_K
[2048 2304]
-
blk.2.attn_q.weight
Q3_K
[2304 2048]
-
blk.2.attn_v.weight
Q4_K
[2304 1024]
-
blk.3.attn_norm.weight
F32
[2304]
-
blk.3.ffn_down.weight
Q4_K
[9216 2304]
-
blk.3.ffn_gate.weight
Q3_K
[2304 9216]
-
blk.3.ffn_up.weight
Q3_K
[2304 9216]
-
blk.3.post_attention_norm.weight
F32
[2304]
-
blk.3.post_ffw_norm.weight
F32
[2304]
-
blk.3.ffn_norm.weight
F32
[2304]
-
blk.3.attn_k.weight
Q3_K
[2304 1024]
-
blk.3.attn_output.weight
Q4_K
[2048 2304]
-
blk.3.attn_q.weight
Q3_K
[2304 2048]
-
blk.3.attn_v.weight
Q4_K
[2304 1024]
-
blk.4.attn_norm.weight
F32
[2304]
-
blk.4.ffn_down.weight
Q4_K
[9216 2304]
-
blk.4.ffn_gate.weight
Q3_K
[2304 9216]
-
blk.4.ffn_up.weight
Q3_K
[2304 9216]
-
blk.4.post_attention_norm.weight
F32
[2304]
-
blk.4.post_ffw_norm.weight
F32
[2304]
-
blk.4.ffn_norm.weight
F32
[2304]
-
blk.4.attn_k.weight
Q3_K
[2304 1024]
-
blk.4.attn_output.weight
Q4_K
[2048 2304]
-
blk.4.attn_q.weight
Q3_K
[2304 2048]
-
blk.4.attn_v.weight
Q4_K
[2304 1024]
-
blk.5.attn_norm.weight
F32
[2304]
-
blk.5.ffn_down.weight
Q4_K
[9216 2304]
-
blk.5.ffn_gate.weight
Q3_K
[2304 9216]
-
blk.5.ffn_up.weight
Q3_K
[2304 9216]
-
blk.5.post_attention_norm.weight
F32
[2304]
-
blk.5.post_ffw_norm.weight
F32
[2304]
-
blk.5.ffn_norm.weight
F32
[2304]
-
blk.5.attn_k.weight
Q3_K
[2304 1024]
-
blk.5.attn_output.weight
Q4_K
[2048 2304]
-
blk.5.attn_q.weight
Q3_K
[2304 2048]
-
blk.5.attn_v.weight
Q4_K
[2304 1024]
-
blk.6.attn_norm.weight
F32
[2304]
-
blk.6.ffn_down.weight
Q4_K
[9216 2304]
-
blk.6.ffn_gate.weight
Q3_K
[2304 9216]
-
blk.6.ffn_up.weight
Q3_K
[2304 9216]
-
blk.6.post_attention_norm.weight
F32
[2304]
-
blk.6.post_ffw_norm.weight
F32
[2304]
-
blk.6.ffn_norm.weight
F32
[2304]
-
blk.6.attn_k.weight
Q3_K
[2304 1024]
-
blk.6.attn_output.weight
Q4_K
[2048 2304]
-
blk.6.attn_q.weight
Q3_K
[2304 2048]
-
blk.6.attn_v.weight
Q4_K
[2304 1024]
-
blk.7.attn_norm.weight
F32
[2304]
-
blk.7.ffn_down.weight
Q4_K
[9216 2304]
-
blk.7.ffn_gate.weight
Q3_K
[2304 9216]
-
blk.7.ffn_up.weight
Q3_K
[2304 9216]
-
blk.7.post_attention_norm.weight
F32
[2304]
-
blk.7.post_ffw_norm.weight
F32
[2304]
-
blk.7.ffn_norm.weight
F32
[2304]
-
blk.7.attn_k.weight
Q3_K
[2304 1024]
-
blk.7.attn_output.weight
Q4_K
[2048 2304]
-
blk.7.attn_q.weight
Q3_K
[2304 2048]
-
blk.7.attn_v.weight
Q4_K
[2304 1024]
-
blk.8.attn_norm.weight
F32
[2304]
-
blk.8.ffn_down.weight
Q4_K
[9216 2304]
-
blk.8.ffn_gate.weight
Q3_K
[2304 9216]
-
blk.8.ffn_up.weight
Q3_K
[2304 9216]
-
blk.8.post_attention_norm.weight
F32
[2304]
-
blk.8.post_ffw_norm.weight
F32
[2304]
-
blk.8.ffn_norm.weight
F32
[2304]
-
blk.8.attn_k.weight
Q3_K
[2304 1024]
-
blk.8.attn_output.weight
Q4_K
[2048 2304]
-
blk.8.attn_q.weight
Q3_K
[2304 2048]
-
blk.8.attn_v.weight
Q4_K
[2304 1024]
-
blk.9.attn_norm.weight
F32
[2304]
-
blk.9.ffn_down.weight
Q4_K
[9216 2304]
-
blk.9.ffn_gate.weight
Q3_K
[2304 9216]
-
blk.9.ffn_up.weight
Q3_K
[2304 9216]
-
blk.9.post_attention_norm.weight
F32
[2304]
-
blk.9.post_ffw_norm.weight
F32
[2304]
-
blk.9.ffn_norm.weight
F32
[2304]
-
blk.9.attn_k.weight
Q3_K
[2304 1024]
-
blk.9.attn_output.weight
Q4_K
[2048 2304]
-
blk.9.attn_q.weight
Q3_K
[2304 2048]
-
blk.9.attn_v.weight
Q4_K
[2304 1024]
-
blk.10.attn_norm.weight
F32
[2304]
-
blk.10.ffn_down.weight
Q4_K
[9216 2304]
-
blk.10.ffn_gate.weight
Q3_K
[2304 9216]
-
blk.10.ffn_up.weight
Q3_K
[2304 9216]
-
blk.10.post_attention_norm.weight
F32
[2304]
-
blk.10.post_ffw_norm.weight
F32
[2304]
-
blk.10.ffn_norm.weight
F32
[2304]
-
blk.10.attn_k.weight
Q3_K
[2304 1024]
-
blk.10.attn_output.weight
Q4_K
[2048 2304]
-
blk.10.attn_q.weight
Q3_K
[2304 2048]
-
blk.10.attn_v.weight
Q4_K
[2304 1024]
-
blk.11.attn_norm.weight
F32
[2304]
-
blk.11.ffn_down.weight
Q4_K
[9216 2304]
-
blk.11.ffn_gate.weight
Q3_K
[2304 9216]
-
blk.11.ffn_up.weight
Q3_K
[2304 9216]
-
blk.11.post_attention_norm.weight
F32
[2304]
-
blk.11.post_ffw_norm.weight
F32
[2304]
-
blk.11.ffn_norm.weight
F32
[2304]
-
blk.11.attn_k.weight
Q3_K
[2304 1024]
-
blk.11.attn_output.weight
Q4_K
[2048 2304]
-
blk.11.attn_q.weight
Q3_K
[2304 2048]
-
blk.11.attn_v.weight
Q4_K
[2304 1024]
-
blk.12.attn_norm.weight
F32
[2304]
-
blk.12.ffn_down.weight
Q4_K
[9216 2304]
-
blk.12.ffn_gate.weight
Q3_K
[2304 9216]
-
blk.12.ffn_up.weight
Q3_K
[2304 9216]
-
blk.12.post_attention_norm.weight
F32
[2304]
-
blk.12.post_ffw_norm.weight
F32
[2304]
-
blk.12.ffn_norm.weight
F32
[2304]
-
blk.12.attn_k.weight
Q3_K
[2304 1024]
-
blk.12.attn_output.weight
Q4_K
[2048 2304]
-
blk.12.attn_q.weight
Q3_K
[2304 2048]
-
blk.12.attn_v.weight
Q4_K
[2304 1024]
-
blk.13.attn_norm.weight
F32
[2304]
-
blk.13.ffn_down.weight
Q4_K
[9216 2304]
-
blk.13.ffn_gate.weight
Q3_K
[2304 9216]
-
blk.13.ffn_up.weight
Q3_K
[2304 9216]
-
blk.13.post_attention_norm.weight
F32
[2304]
-
blk.13.post_ffw_norm.weight
F32
[2304]
-
blk.13.ffn_norm.weight
F32
[2304]
-
blk.13.attn_k.weight
Q3_K
[2304 1024]
-
blk.13.attn_output.weight
Q4_K
[2048 2304]
-
blk.13.attn_q.weight
Q3_K
[2304 2048]
-
blk.13.attn_v.weight
Q4_K
[2304 1024]
-
blk.14.attn_norm.weight
F32
[2304]
-
blk.14.ffn_down.weight
Q4_K
[9216 2304]
-
blk.14.ffn_gate.weight
Q3_K
[2304 9216]
-
blk.14.ffn_up.weight
Q3_K
[2304 9216]
-
blk.14.post_attention_norm.weight
F32
[2304]
-
blk.14.post_ffw_norm.weight
F32
[2304]
-
blk.14.ffn_norm.weight
F32
[2304]
-
blk.14.attn_k.weight
Q3_K
[2304 1024]
-
blk.14.attn_output.weight
Q4_K
[2048 2304]
-
blk.14.attn_q.weight
Q3_K
[2304 2048]
-
blk.14.attn_v.weight
Q4_K
[2304 1024]
-
blk.15.attn_norm.weight
F32
[2304]
-
blk.15.ffn_down.weight
Q4_K
[9216 2304]
-
blk.15.ffn_gate.weight
Q3_K
[2304 9216]
-
blk.15.ffn_up.weight
Q3_K
[2304 9216]
-
blk.15.post_attention_norm.weight
F32
[2304]
-
blk.15.post_ffw_norm.weight
F32
[2304]
-
blk.15.ffn_norm.weight
F32
[2304]
-
blk.15.attn_k.weight
Q3_K
[2304 1024]
-
blk.15.attn_output.weight
Q4_K
[2048 2304]
-
blk.15.attn_q.weight
Q3_K
[2304 2048]
-
blk.15.attn_v.weight
Q4_K
[2304 1024]
-
blk.16.attn_norm.weight
F32
[2304]
-
blk.16.ffn_down.weight
Q4_K
[9216 2304]
-
blk.16.ffn_gate.weight
Q3_K
[2304 9216]
-
blk.16.ffn_up.weight
Q3_K
[2304 9216]
-
blk.16.post_attention_norm.weight
F32
[2304]
-
blk.16.post_ffw_norm.weight
F32
[2304]
-
blk.16.ffn_norm.weight
F32
[2304]
-
blk.16.attn_k.weight
Q3_K
[2304 1024]
-
blk.16.attn_output.weight
Q4_K
[2048 2304]
-
blk.16.attn_q.weight
Q3_K
[2304 2048]
-
blk.16.attn_v.weight
Q4_K
[2304 1024]
-
blk.17.attn_norm.weight
F32
[2304]
-
blk.17.ffn_down.weight
Q4_K
[9216 2304]
-
blk.17.ffn_gate.weight
Q3_K
[2304 9216]
-
blk.17.ffn_up.weight
Q3_K
[2304 9216]
-
blk.17.post_attention_norm.weight
F32
[2304]
-
blk.17.post_ffw_norm.weight
F32
[2304]
-
blk.17.ffn_norm.weight
F32
[2304]
-
blk.17.attn_k.weight
Q3_K
[2304 1024]
-
blk.17.attn_output.weight
Q4_K
[2048 2304]
-
blk.17.attn_q.weight
Q3_K
[2304 2048]
-
blk.17.attn_v.weight
Q4_K
[2304 1024]
-
blk.18.attn_norm.weight
F32
[2304]
-
blk.18.ffn_down.weight
Q4_K
[9216 2304]
-
blk.18.ffn_gate.weight
Q3_K
[2304 9216]
-
blk.18.ffn_up.weight
Q3_K
[2304 9216]
-
blk.18.post_attention_norm.weight
F32
[2304]
-
blk.18.post_ffw_norm.weight
F32
[2304]
-
blk.18.ffn_norm.weight
F32
[2304]
-
blk.18.attn_k.weight
Q3_K
[2304 1024]
-
blk.18.attn_output.weight
Q4_K
[2048 2304]
-
blk.18.attn_q.weight
Q3_K
[2304 2048]
-
blk.18.attn_v.weight
Q4_K
[2304 1024]
-
blk.19.attn_norm.weight
F32
[2304]
-
blk.19.ffn_down.weight
Q4_K
[9216 2304]
-
blk.19.ffn_gate.weight
Q3_K
[2304 9216]
-
blk.19.ffn_up.weight
Q3_K
[2304 9216]
-
blk.19.post_attention_norm.weight
F32
[2304]
-
blk.19.post_ffw_norm.weight
F32
[2304]
-
blk.19.ffn_norm.weight
F32
[2304]
-
blk.19.attn_k.weight
Q3_K
[2304 1024]
-
blk.19.attn_output.weight
Q4_K
[2048 2304]
-
blk.19.attn_q.weight
Q3_K
[2304 2048]
-
blk.19.attn_v.weight
Q4_K
[2304 1024]
-
blk.20.attn_norm.weight
F32
[2304]
-
blk.20.ffn_down.weight
Q4_K
[9216 2304]
-
blk.20.ffn_gate.weight
Q3_K
[2304 9216]
-
blk.20.ffn_up.weight
Q3_K
[2304 9216]
-
blk.20.post_attention_norm.weight
F32
[2304]
-
blk.20.post_ffw_norm.weight
F32
[2304]
-
blk.20.ffn_norm.weight
F32
[2304]
-
blk.20.attn_k.weight
Q3_K
[2304 1024]
-
blk.20.attn_output.weight
Q4_K
[2048 2304]
-
blk.20.attn_q.weight
Q3_K
[2304 2048]
-
blk.20.attn_v.weight
Q4_K
[2304 1024]
-
blk.21.attn_norm.weight
F32
[2304]
-
blk.21.ffn_down.weight
Q4_K
[9216 2304]
-
blk.21.ffn_gate.weight
Q3_K
[2304 9216]
-
blk.21.ffn_up.weight
Q3_K
[2304 9216]
-
blk.21.post_attention_norm.weight
F32
[2304]
-
blk.21.post_ffw_norm.weight
F32
[2304]
-
blk.21.ffn_norm.weight
F32
[2304]
-
blk.21.attn_k.weight
Q3_K
[2304 1024]
-
blk.21.attn_output.weight
Q4_K
[2048 2304]
-
blk.21.attn_q.weight
Q3_K
[2304 2048]
-
blk.21.attn_v.weight
Q4_K
[2304 1024]
-
blk.22.attn_norm.weight
F32
[2304]
-
blk.22.ffn_down.weight
Q4_K
[9216 2304]
-
blk.22.ffn_gate.weight
Q3_K
[2304 9216]
-
blk.22.ffn_up.weight
Q3_K
[2304 9216]
-
blk.22.post_attention_norm.weight
F32
[2304]
-
blk.22.post_ffw_norm.weight
F32
[2304]
-
blk.22.ffn_norm.weight
F32
[2304]
-
blk.22.attn_k.weight
Q3_K
[2304 1024]
-
blk.22.attn_output.weight
Q4_K
[2048 2304]
-
blk.22.attn_q.weight
Q3_K
[2304 2048]
-
blk.22.attn_v.weight
Q4_K
[2304 1024]
-
blk.23.attn_norm.weight
F32
[2304]
-
blk.23.ffn_down.weight
Q4_K
[9216 2304]
-
blk.23.ffn_gate.weight
Q3_K
[2304 9216]
-
blk.23.ffn_up.weight
Q3_K
[2304 9216]
-
blk.23.post_attention_norm.weight
F32
[2304]
-
blk.23.post_ffw_norm.weight
F32
[2304]
-
blk.23.ffn_norm.weight
F32
[2304]
-
blk.23.attn_k.weight
Q3_K
[2304 1024]
-
blk.23.attn_output.weight
Q4_K
[2048 2304]
-
blk.23.attn_q.weight
Q3_K
[2304 2048]
-
blk.23.attn_v.weight
Q4_K
[2304 1024]
-
blk.24.ffn_gate.weight
Q3_K
[2304 9216]
-
blk.24.attn_k.weight
Q3_K
[2304 1024]
-
blk.24.attn_output.weight
Q4_K
[2048 2304]
-
blk.24.attn_q.weight
Q3_K
[2304 2048]
-
blk.24.attn_v.weight
Q4_K
[2304 1024]
-
blk.24.attn_norm.weight
F32
[2304]
-
blk.24.ffn_down.weight
Q4_K
[9216 2304]
-
blk.24.ffn_up.weight
Q3_K
[2304 9216]
-
blk.24.post_attention_norm.weight
F32
[2304]
-
blk.24.post_ffw_norm.weight
F32
[2304]
-
blk.24.ffn_norm.weight
F32
[2304]
-
blk.25.attn_norm.weight
F32
[2304]
-
blk.25.ffn_down.weight
Q4_K
[9216 2304]
-
blk.25.ffn_gate.weight
Q3_K
[2304 9216]
-
blk.25.ffn_up.weight
Q3_K
[2304 9216]
-
blk.25.post_attention_norm.weight
F32
[2304]
-
blk.25.post_ffw_norm.weight
F32
[2304]
-
blk.25.ffn_norm.weight
F32
[2304]
-
blk.25.attn_k.weight
Q3_K
[2304 1024]
-
blk.25.attn_output.weight
Q4_K
[2048 2304]
-
blk.25.attn_q.weight
Q3_K
[2304 2048]
-
blk.25.attn_v.weight
Q4_K
[2304 1024]
-
output_norm.weight
F32
[2304]