/Users/simon-pierreboucher/Desktop/Recherche-UQO/ia/prisme-slm/training/train_controller.py:131: UserWarning: Converting a tensor with requires_grad=True to a scalar may lead to unexpected behavior.
Consider using tensor.detach() first. (Triggered internally at /Users/runner/work/pytorch/pytorch/torch/csrc/autograd/generated/python_variable_methods.cpp:821.)
  run.log(step=step, loss=float(loss), gain=float(L["gain"]), harm=float(L["harm"]), nll=float(L["nll"]))
step 0 loss 0.72030 gain 0.04784 harm 1.2803
step 100 loss 0.11040 gain 0.00198 harm 0.6636
step 200 loss 0.05812 gain 0.00234 harm 0.5654
  calib: {'huber': 0.0017, 'sign_acc': 0.5849, 'harm_brier': 0.1899, 'harm_ece': 0.0398, 'n_labels': 26916, 'oracle_opportunity_per_state': 0.0136, 'regret_per_state': 0.0154, 'captured_fraction': -0.1302, 'top1_match': 0.0748, 'n_audit_states': 1645}
step 300 loss 0.04144 gain 0.00179 harm 0.5736
step 400 loss -0.00426 gain 0.00227 harm 0.4718
  calib: {'huber': 0.0021, 'sign_acc': 0.4788, 'harm_brier': 0.1847, 'harm_ece': 0.0453, 'n_labels': 26916, 'oracle_opportunity_per_state': 0.0136, 'regret_per_state': 0.0169, 'captured_fraction': -0.239, 'top1_match': 0.0693, 'n_audit_states': 1645}
step 500 loss 0.04916 gain 0.00226 harm 0.5915
step 600 loss -0.00805 gain 0.00153 harm 0.4968
step 700 loss -0.02066 gain 0.00187 harm 0.4985
  calib: {'huber': 0.0015, 'sign_acc': 0.4807, 'harm_brier': 0.1758, 'harm_ece': 0.0474, 'n_labels': 26916, 'oracle_opportunity_per_state': 0.0136, 'regret_per_state': 0.0173, 'captured_fraction': -0.2749, 'top1_match': 0.0802, 'n_audit_states': 1645}
step 800 loss 0.03044 gain 0.00134 harm 0.5327
step 900 loss -0.03801 gain 0.00105 harm 0.4983
  calib: {'huber': 0.0013, 'sign_acc': 0.6034, 'harm_brier': 0.1621, 'harm_ece': 0.0166, 'n_labels': 26916, 'oracle_opportunity_per_state': 0.0136, 'regret_per_state': 0.0181, 'captured_fraction': -0.3316, 'top1_match': 0.0784, 'n_audit_states': 1645}
step 1000 loss -0.02688 gain 0.00095 harm 0.5342
step 1100 loss 0.03442 gain 0.00123 harm 0.4801
step 1200 loss -0.11175 gain 0.00145 harm 0.4062
  calib: {'huber': 0.0012, 'sign_acc': 0.5993, 'harm_brier': 0.1567, 'harm_ece': 0.0241, 'n_labels': 26916, 'oracle_opportunity_per_state': 0.0136, 'regret_per_state': 0.0176, 'captured_fraction': -0.2913, 'top1_match': 0.0748, 'n_audit_states': 1645}
step 1300 loss -0.09109 gain 0.00083 harm 0.4610
step 1400 loss -0.11961 gain 0.00143 harm 0.4281
  calib: {'huber': 0.0011, 'sign_acc': 0.4789, 'harm_brier': 0.1553, 'harm_ece': 0.0189, 'n_labels': 26916, 'oracle_opportunity_per_state': 0.0136, 'regret_per_state': 0.018, 'captured_fraction': -0.3198, 'top1_match': 0.0541, 'n_audit_states': 1645}
step 1500 loss -0.12260 gain 0.00123 harm 0.3940
step 1600 loss -0.12072 gain 0.00048 harm 0.4509
step 1700 loss -0.08131 gain 0.00089 harm 0.4556
  calib: {'huber': 0.0011, 'sign_acc': 0.6076, 'harm_brier': 0.1546, 'harm_ece': 0.029, 'n_labels': 26916, 'oracle_opportunity_per_state': 0.0136, 'regret_per_state': 0.0172, 'captured_fraction': -0.2634, 'top1_match': 0.0693, 'n_audit_states': 1645}
step 1800 loss -0.05758 gain 0.00079 harm 0.4626
step 1900 loss -0.08881 gain 0.00047 harm 0.4962
step 1999 loss -0.10400 gain 0.00060 harm 0.5061
  calib: {'huber': 0.001, 'sign_acc': 0.6257, 'harm_brier': 0.1534, 'harm_ece': 0.0172, 'n_labels': 26916, 'oracle_opportunity_per_state': 0.0136, 'regret_per_state': 0.0194, 'captured_fraction': -0.427, 'top1_match': 0.0821, 'n_audit_states': 1645}
restored controller from step 250 (calibration score -0.01538)
sweep {'lambda': 0.0, 'kappa': 0.0, 'loss': 0.44644779327501183, 'edits_per_token': 3.2716488730723605, 'extra_ops_ratio': 1.009114049366219}
sweep {'lambda': 0.0, 'kappa': 0.5, 'loss': 0.43412450878645603, 'edits_per_token': 1.6144721233689205, 'extra_ops_ratio': 0.4928484992483191}
sweep {'lambda': 0.01, 'kappa': 0.0, 'loss': 0.4457159031045536, 'edits_per_token': 3.0355871886120998, 'extra_ops_ratio': 0.9343256107570155}
sweep {'lambda': 0.01, 'kappa': 0.5, 'loss': 0.43284321396941955, 'edits_per_token': 1.4572953736654803, 'extra_ops_ratio': 0.44381220852635794}
sweep {'lambda': 0.03, 'kappa': 0.0, 'loss': 0.4455074115721498, 'edits_per_token': 2.6565836298932384, 'extra_ops_ratio': 0.8122402747754949}
sweep {'lambda': 0.03, 'kappa': 0.5, 'loss': 0.4333898039757852, 'edits_per_token': 1.1986951364175564, 'extra_ops_ratio': 0.3607135243263109}
sweep {'lambda': 0.1, 'kappa': 0.0, 'loss': 0.4371275036360445, 'edits_per_token': 1.547449584816133, 'extra_ops_ratio': 0.45874506496732625}
sweep {'lambda': 0.1, 'kappa': 0.5, 'loss': 0.4322353082478117, 'edits_per_token': 0.5830367734282325, 'extra_ops_ratio': 0.17120895068982034}
sweep {'lambda': 0.3, 'kappa': 0.0, 'loss': 0.43019490903806856, 'edits_per_token': 0.199288256227758, 'extra_ops_ratio': 0.058016694878909654}
sweep {'lambda': 0.3, 'kappa': 0.5, 'loss': 0.43101718671805495, 'edits_per_token': 0.03914590747330961, 'extra_ops_ratio': 0.011396136837497725}
selected {'lambda': 0.3, 'kappa': 0.0, 'loss': 0.43019490903806856, 'edits_per_token': 0.199288256227758, 'extra_ops_ratio': 0.058016694878909654, 'loss_reduction': 0.0008682667289905965, 'reduction_per_ops': 0.014965808217838184} 
run controller_v1ctrl_seed01_20261030004131