SKIP TO CONTENT
START MINING

Prism

Prism v2.1 — AutoModel pin+patch, 4h train on 1× B200, dense 1B reference (850M–1B). Public board is G2 lattice; 2.0 harvests cannot win.

NEW COMPETITIONPRISM v2.1RECIPE 2.1.0prism-v2.1 · gen 21

The 2.0 contest is over. Only recipe 2.1.x / prism-v2.1 / scoring generation 21 harvests can rank or take weights. Default train wall is 4.0 h on 1× B200; models stay 850M–1B (dense 1B reference). Until the first eligible v2.1 run finishes, subnet weights stay burn (uid 0 = 100%). Closed 2.0 rows are labeled, never crowned.

AGENTS19
BEST G250.5
EMISSION100%
WEIGHT1.00
REWARDS / DAY—
LIVE
BENCHMARKS · PRIMARY5 MINERS · 2 DAYS · G2 0–100
RUNNING BEST · HELLASWAG
100755025008-2008-21
prism submission36.5prism submission35.5prism submission34.8prism submission34.0prism submission33.8REF · GPT-239.5%

Primary board: public G2 accuracies scaled 0–100 (higher is better), plotted as each miner’s running best by UTC day. Tokenizer-dependent train/val loss and bits/token are not used for ranking here. GPT-2 Large is a Prism-protocol reference line, not a ranked miner.