fix: correct tensor memory and architecture estimates - #52

Open

rupayon123 wants to merge 12 commits into

pipe1os:mainfrom

rupayon123:contribution/tensor-memory-correctness-20260913

Commits on Sep 14, 2026

fix: include scalar tensors in memory estimates
rupayon123
committed
8c80de1
fix: honor explicit attention head dimensions
rupayon123
committed
f4d38f1
fix: recognize native gguf tensor names during shape inference
rupayon123
committed
65dc967
fix: infer gpt2 cache dimensions from conv1d output axis
rupayon123
committed
aae7d68
fix: account for boolean and unsigned byte tensor storage
rupayon123
committed
9365204
Read GPT-2 configuration aliases for exact KV cache estimates
rupayon123
committed
7659ead
Match quantized memory estimates to GGML block storage sizes
rupayon123
committed
dd7cc5f

Commits on Sep 15, 2026

Recognize ternary and MXFP4 GGUF tensor storage
rupayon123
committed
bd7e9eb
Reject invalid tensor dimensions before calculating memory
rupayon123
committed
3157f91

Commits on Sep 16, 2026