2026-08-24

  • 18:40 submitted 20260824-183902-qwen3_4b_gsm8k_fixed_batch_reward_signal_jean_zay-39c4380 (job 1339698 on jean-zay, commit 39c4380)
  • 18:40 submitted fixed-batch Qwen3 GSM8K reward-signal diagnostic 20260824-183902-qwen3_4b_gsm8k_fixed_batch_reward_signal_jean_zay-39c4380 on Jean Zay as job 1339698 after fixing bootstrap quoting; this tests whether exact-answer reward has usable signal under lr=0, lr=+1, and lr=-1 controls