W1 GGUF header and tensor-directory inspection
Date: 2026-10-22
Machine architecture: arm64
Command: python3 code/gguf_probe.py ~/Documents/claude/gemma4-e4b-chat/gemma-4-E4B-it-IQ4_XS.gguf
GGUF v3: 720 tensors, 56 metadata entries
file size: 4,715,416,704 bytes; directory ends at 15,826,723; data starts at 15,826,752; alignment: 32
general.architecture = gemma4
general.name = Gemma-4-E4B-It
gemma4.block_count = 42
gemma4.embedding_length = 2560
gemma4.attention.shared_kv_layers = 18
gemma4.embedding_length_per_layer_input = 256
tensor types: BF16=1 (55,050,240 declared bytes), F32=423 (222,464,168 declared bytes), IQ4_XS=234 (2,047,180,800 declared bytes), Q4_K=19 (392,970,240 declared bytes), Q5_K=43 (1,981,923,328 declared bytes)
per_layer_token_embd.weight: Q5_K shape=[10752, 262144] offset=55,061,504 bytes=1,937,768,448
blk.0.attn_q.weight: IQ4_XS shape=[2560, 2048] offset=2,373,811,200 bytes=2,785,280
payload bytes read: 0 (header, metadata, and tensor directory only)
