Depth-124 radix trie
Stub README (2026-09-28): this directory predates the README convention. The summary below was reconstructed from its files; see them for detail.
Question. Can the CUDAX v2 trainer handle a full depth-124 Shakespeare radix trie (the depth at which raw contexts become unique), and if not, what is the blocker?
Answer. It is not obviously impossible. The compact K/V cache is about 3.46 GB, only about 14% more slots than d16. The real costs are 129.6M query positions per epoch (about 14x d16) and about 1.5 GiB of full edge-char arrays in the loader. The note recommends chunk_queries=10000 and sampling of mass-1 cap tails.
Sources. Directory holds only the built trie shake_d124_radix; the answer comes from notes/seq-len-extension/d124-radix-feasibility.md ('Actual Blockers' and 'Current Read'), which names this path.
Caveats. The dir is untracked (.gitignore excludes rnd/radix-depth*/**/*.The follow-up training runs are in rnd/cudax-d124-probe.