GGUF v3: 720 descriptors; payloads not read
gemma4.block_count = 42
gemma4.embedding_length = 2560
gemma4.feed_forward_length = 10240
gemma4.embedding_length_per_layer_input = 256
gemma4.attention.shared_kv_layers = 18
gemma4.final_logit_softcapping = 30.0
sliding layers: 35; global layers: 7
global layer indices (zero based): 5, 11, 17, 23, 29, 35, 41
selected matrices:
blk.0.attn_q.weight: IQ4_XS ne=[2560, 2048] bytes=2785280
blk.0.attn_k.weight: IQ4_XS ne=[2560, 512] bytes=696320
blk.0.attn_v.weight: Q5_K ne=[2560, 512] bytes=901120
blk.0.attn_output.weight: IQ4_XS ne=[2048, 2560] bytes=2785280
blk.5.attn_q.weight: IQ4_XS ne=[2560, 4096] bytes=5570560
blk.5.attn_k.weight: IQ4_XS ne=[2560, 1024] bytes=1392640
blk.5.attn_v.weight: Q5_K ne=[2560, 1024] bytes=1802240
blk.5.attn_output.weight: IQ4_XS ne=[4096, 2560] bytes=5570560
per_layer_token_embd.weight: Q5_K ne=[10752, 262144] bytes=1937768448
token_embd.weight: Q4_K ne=[2560, 262144] bytes=377487360
output_norm.weight: F32 ne=[2560] bytes=10240
