[08-07 13:59] ===== 1. rebuild the SimPO anchors (rounds 1-6, two seeds in parallel) =====
[08-07 13:59] sp44 r1 simpo training on gpu 1
[08-07 13:59] p43 r1 simpo training on gpu 0
[08-07 14:37] sp44 r2 simpo training on gpu 1
[08-07 14:38] p43 r2 simpo training on gpu 0
[08-07 15:15] sp44 r3 simpo training on gpu 1
[08-07 15:15] p43 r3 simpo training on gpu 0
[08-07 15:52] sp44 r4 simpo training on gpu 1
[08-07 15:53] p43 r4 simpo training on gpu 0
[08-07 16:30] sp44 r5 simpo training on gpu 1
[08-07 16:31] p43 r5 simpo training on gpu 0
[08-07 17:07] sp44 r6 simpo training on gpu 1
[08-07 17:08] p43 r6 simpo training on gpu 0
[08-07 17:45] sp44 simpo chain rebuilt through round 6
[08-07 17:45] p43 simpo chain rebuilt through round 6
[08-07 17:45] ===== 2. INPO and DPO at round 7 (four jobs, four GPUs) =====
[08-07 18:22] round-7 INPO and DPO trained at both seeds
[08-07 18:22] ===== 3. evaluate (Qwen3-32B) =====
[08-07 18:36] qwen evaluation done
[08-07 18:36] ===== 4. cross-judge (Llama-3.3-70B, Phi-4) =====
[08-07 19:02] QUEUE_ID3_DONE