| Loading checkpoint shards: 0%| | 0/3 [00:00<?, ?it/s] Loading checkpoint shards: 33%|ββββ | 1/3 [00:00<00:00, 8.92it/s] Loading checkpoint shards: 67%|βββββββ | 2/3 [00:00<00:00, 8.94it/s] Loading checkpoint shards: 100%|ββββββββββ| 3/3 [00:00<00:00, 9.85it/s] | |
| baseline c4=4.3505 | |
| Loading checkpoint shards: 0%| | 0/3 [00:00<?, ?it/s] Loading checkpoint shards: 33%|ββββ | 1/3 [00:00<00:00, 9.39it/s] Loading checkpoint shards: 67%|βββββββ | 2/3 [00:00<00:00, 9.15it/s] Loading checkpoint shards: 100%|ββββββββββ| 3/3 [00:00<00:00, 10.04it/s] | |
| uniform_keep50 c4=5.1110 [dead 0/1024, kept 524288, params 3.70B] | |
| decision gate (variant NLL - glean NLL; positive = glean better): | |
| saved -> outputs/uniform_control_materialize/results.json | |