diff --git a/scripts/frank-default-large.env b/scripts/frank-default-large.env index 00ed74a..5142a2a 100644 --- a/scripts/frank-default-large.env +++ b/scripts/frank-default-large.env @@ -12,23 +12,15 @@ PORT=7770 # ~70GB headroom, so 2x262k quantized KV fits. Verify at startup. CTX_SIZE=524288 PARALLEL=2 -# Spec: draft-mtp uses the MTP drafter BUNDLED inside the target GGUF. -# llama.cpp auto-detects it (b10411+). DRAFT_FILE stays EMPTY. -# -# Winner = draft-mtp, n_max=5, p_min off. -# Realistic-prompt tuning (2026-08-14, temp 0.8, production settings): -# n4 12.5 | n5 13.3 (PEAK) | n6 10.6 | n7 10.1 | n10 8.6 | baseline none 7.2. -# The 54-config matrix (repetitive "fleet" prompt) reported n_max=10 ~30 t/s, -# but that prompt MTP drafts almost perfectly; long drafts collapse to ~18% -# acceptance on real writing, so n_max=10 is actually ~8.6 t/s there. -# Layered draft-mtp,ngram-mod is faster on paper (matrix median 133 t/s) but -# non-deterministic (10-148 t/s) and showed a degenerate instant-EOS -> the -# same premature-stop class that retired DFlash. Rejected as production-unsafe. -# p_min off: --spec-draft-p-min 0.8 was neutral to slightly worse for draft-mtp. -# No premature-stop on realistic text at ubatch 256 (see batch notes below); -# the earlier 'premature EOS' on the repetitive bench prompt was sampling -# variance, not a corruption bug. -SPEC_TYPE=draft-mtp +# Spec: 3-layer chain, chosen by the user 2026-08-14 (aesthetic/short-term +# experiment): draft-mtp first, then ngram-mod, then ngram-map-k4v as +# fallbacks. On real prose this benches ~equal to draft-mtp alone +# (10.4 vs ~10 t/s @16k) - the ngram layers only pay off on self-similar text. +# Bundled MTP drafter; llama.cpp auto-detects it (b10411+). DRAFT_FILE empty. +# n_max=5 is the realistic-text peak (n4 12.5, n6 10.6, n10 8.6). +# Occasional slow rounds possible when fallbacks misfire; revert to +# draft-mtp (alone) if the assistant reports stalls. +SPEC_TYPE=draft-mtp,ngram-mod,ngram-map-k4v DRAFT_FILE= SPEC_DRAFT_N_MAX=5 SPEC_DRAFT_P_MIN=