Commit History
card: tiered-v3/ — an fp16 vocab path, seven rungs, prefill and verify 15f9e57 verified
tiered-v3/ — an fp16 vocab path, seven rungs, prefill and verify 23761c6 verified
model card: standard structure, measured facts, verified license declarations 2489fa8 verified
tiered: ladder README (staged status, digests, per-rung requirements) 6911b0e verified
tiered: 7-rung capacity ladder (4k-256k), shape-stable windowed KV - staged, unmeasured on device 787fe35 verified
dense2in: add the missing tokenizer/generation_config.json (fix by addition; byte-identical to the file the two dense2in-pf64 folders already ship) 37d64e2 verified
docs: expected shape-specialization memory-growth note (sibling-measured, unverified here) b2265e8 verified
prune: remove superseded/non-functional artifacts 52fd753 verified
docs: refresh measured performance (local retest 2026-08-18) d3fdc88 verified
docs: descriptive model card 117d170 verified
README: dense2in-int8 bundles 1012bbd verified
Ben Graville commited on
Add dense2in-int8/gemma4_12b_qat_decode_int8lin_pf64 0e2210c verified
Ben Graville commited on
README: dense2in-pf64 bundles d2316d5 verified
Ben Graville commited on
Add dense2in-pf64/gemma4_12b_qat_decode_int4linsym_pf64 212deaf verified
Ben Graville commited on
README: dense2in bundles c510bef verified
Ben Graville commited on
Add dense2in/gemma4_12b_qat_decode_int4linsym 6b66256 verified
Ben Graville commited on
Add gemma4_12b_qat_decode_int4linsym_msdpa_g8 d24b588 verified
Ben Graville commited on
Upload README.md with huggingface_hub bc74fda verified
Ben Graville commited on
initial commit d2e5b9a verified
Ben Graville commited on