dankclaw: MTP wins the spec bench; raise ubatch to 4096

MTP (draft-mtp, n_max 3, a4lg MTP-ONLY-Q8_0) benches 15.3-19.5 t/s vs
ngram-mod 6.7-8.7 t/s and passes the strict LCP-reuse parallel test
(sim 0.996 / f_keep 0.994 launch while another slot generates - the
pattern that killed DFlash). ubatch 4096 to use the ~33 GiB VRAM
headroom; watch peak during load.
This commit is contained in:
Frank
2026-08-13 22:02:40 -07:00
parent 5665e1ad6f
commit f203f1bc40
+1 -1
View File
@@ -39,7 +39,7 @@ CACHE_TYPE_V=q8_0
# ubatch is the compute-graph driver; larger ubatch = more VRAM, faster prefill.
# Re-verified 2026-08-14 during the spec swap (see benchmark doc).
BATCH_SIZE=8192
UBATCH_SIZE=2048
UBATCH_SIZE=4096
# Sampling settings — per HauhauCS Qwen3.6-27B model card.
# Thinking-general: temp=1.0, top_p=0.95, top_k=20, min_p=0.
# Clients may override per-request; these are the server-side defaults.