diff options
Diffstat (limited to 'results')
| -rw-r--r-- | results/shared_feedback/s0.json | 345 |
1 files changed, 345 insertions, 0 deletions
diff --git a/results/shared_feedback/s0.json b/results/shared_feedback/s0.json new file mode 100644 index 0000000..487b8f5 --- /dev/null +++ b/results/shared_feedback/s0.json @@ -0,0 +1,345 @@ +{ + "checks": { + "innovation_above_raw_by_5": false, + "innovation_within_2_of_exact_subtraction": true, + "innovation_within_3_of_oracle": true, + "matched_raw_below_innovation_by_3": false, + "nonzero_context_every_layer": true, + "oracle_at_least_90": true, + "oracle_context_lesion_drop_at_least_10": true, + "predictor_mean_r2_at_least_0p8": true, + "predictor_residual_ratio_at_most_0p25": true, + "raw_below_oracle_by_5_or_nonfinite": false, + "zero_instruction_observations": true + }, + "config": { + "context_scale": 1.0, + "hidden_layers": 2, + "learning_rate": 0.03, + "momentum": 0.9, + "reciprocal_learning_rate": 0.03, + "weight_decay": 0.0001, + "width": 64 + }, + "data": { + "batch_size": 128, + "data_seed": 3101, + "epochs": 40, + "neutral_examples_per_epoch": 512, + "test_generated": false, + "train_examples": 8192, + "validation_examples": 2048, + "validation_seed": 3102 + }, + "gate": "fail", + "provenance": { + "cuda_device_name": null, + "cuda_visible_devices": null, + "device": "cpu", + "git_commit": "b3457848820d0818840e7052c41f01c193d04a67", + "git_dirty_tracked": false, + "torch_version": "2.3.1+cu118" + }, + "records": [ + { + "condition": "oracle", + "context_lesion_validation": { + "accuracy": 0.76318359375, + "loss": 1.016611248254776 + }, + "epoch_train_loss": [ + 0.24325548473279923, + 0.056528628687374294, + 0.04177409497788176, + 0.03513596647826489, + 0.031240705080563203, + 0.02832120239327196, + 0.026365785437519662, + 0.025093382872000802, + 0.02279450709465891, + 0.020808601460885257, + 0.0205843910407566, + 0.019572165729186963, + 0.018927484703453956, + 0.01822459184040781, + 0.017170838073070627, + 0.017234746352187358, + 0.017531994708406273, + 0.016618623849353753, + 0.015849430204980308, + 0.016377428259147564, + 0.0150981526603573, + 0.015522186793532455, + 0.015784009628987405, + 0.015121704243938439, + 0.014048248613107717, + 0.013343326936592348, + 0.013460176676744595, + 0.01443425157413003, + 0.014310249827758525, + 0.013833712246196228, + 0.014911528505763272, + 0.012738225999783026, + 0.014052311023988295, + 0.012899597548312158, + 0.013465201343024091, + 0.01266157620557351, + 0.011880345572535589, + 0.012348399211987271, + 0.013484933984727832, + 0.012156888736171823 + ], + "epochs_completed": 40, + "finite": true, + "first_nonfinite_epoch": null, + "predictor": [ + { + "context_rms": 0.6951092481613159, + "instruction_observations": 0, + "mean_per_cell_r2": 0.2886233329772949, + "neutral_observations": 512, + "residual_context_rms_ratio": 0.44177088141441345 + }, + { + "context_rms": 0.6736197471618652, + "instruction_observations": 0, + "mean_per_cell_r2": 0.2944446802139282, + "neutral_observations": 512, + "residual_context_rms_ratio": 0.5169849395751953 + } + ], + "validation": { + "accuracy": 0.9931640625, + "loss": 0.015071565518155694 + }, + "wall_seconds": 1.3104579448699951 + }, + { + "condition": "raw_shared", + "context_lesion_validation": { + "accuracy": 0.7607421875, + "loss": 0.7937461286783218 + }, + "epoch_train_loss": [ + 0.3208125978708267, + 0.1128524262458086, + 0.09070953441550955, + 0.08244586264481768, + 0.0741222791839391, + 0.0689597157179378, + 0.06720075639896095, + 0.05517955176765099, + 0.04856250758166425, + 0.04077540585421957, + 0.03735822852468118, + 0.03361609256535303, + 0.03078002763504628, + 0.02962772782484535, + 0.026828612928511575, + 0.028356117574730888, + 0.026588890425045975, + 0.024541094884625636, + 0.023484145422116853, + 0.02336278444272466, + 0.023574378858029377, + 0.025389555674337316, + 0.02184313071484212, + 0.020888249862764496, + 0.021251495556498412, + 0.021025175316026434, + 0.02125847306160722, + 0.025749154643563088, + 0.026524975062784506, + 0.02095157753501553, + 0.025126684507995378, + 0.01951841819754918, + 0.024700971407582983, + 0.021017702420067508, + 0.019356194523425074, + 0.027358750809071353, + 0.024594824350060662, + 0.022864215050503844, + 0.02073163459135685, + 0.01956268616777379 + ], + "epochs_completed": 40, + "finite": true, + "first_nonfinite_epoch": null, + "predictor": [ + { + "context_rms": 0.6951092481613159, + "instruction_observations": 0, + "mean_per_cell_r2": 0.8874340653419495, + "neutral_observations": 512, + "residual_context_rms_ratio": 0.05456678196787834 + }, + { + "context_rms": 0.6736197471618652, + "instruction_observations": 0, + "mean_per_cell_r2": 0.5909687280654907, + "neutral_observations": 512, + "residual_context_rms_ratio": 0.2560145854949951 + } + ], + "validation": { + "accuracy": 0.990234375, + "loss": 0.021472798893228173 + }, + "wall_seconds": 1.2234578132629395 + }, + { + "condition": "innovation", + "context_lesion_validation": { + "accuracy": 0.76806640625, + "loss": 0.5023627653717995 + }, + "epoch_train_loss": [ + 0.3356692243833095, + 0.11672432930208743, + 0.0827551357797347, + 0.07098419510293752, + 0.061678838043008, + 0.05678685213206336, + 0.05060782533837482, + 0.04663200181676075, + 0.042038892686832696, + 0.039127209020080045, + 0.03786266638780944, + 0.03385412292846013, + 0.031796471928828396, + 0.029597112981718965, + 0.027298799192067236, + 0.028002181876217946, + 0.026437046893988736, + 0.02457212092122063, + 0.0231955681956606, + 0.023024410373182036, + 0.022563506630831398, + 0.02402085693756817, + 0.02065000399306882, + 0.01983536635816563, + 0.019318511425808538, + 0.019305569570860825, + 0.019423548750637565, + 0.021987034422636498, + 0.019796223194134654, + 0.0184321171582269, + 0.01941464086121414, + 0.016116484581289114, + 0.0172461594993365, + 0.017044081127096433, + 0.01634742740134243, + 0.015952237394230906, + 0.015391441356769064, + 0.015357455056800973, + 0.01551965332691907, + 0.01406276357738534 + ], + "epochs_completed": 40, + "finite": true, + "first_nonfinite_epoch": null, + "predictor": [ + { + "context_rms": 0.6951092481613159, + "instruction_observations": 0, + "mean_per_cell_r2": 0.9216524958610535, + "neutral_observations": 512, + "residual_context_rms_ratio": 0.01552559994161129 + }, + { + "context_rms": 0.6736197471618652, + "instruction_observations": 0, + "mean_per_cell_r2": 0.979547381401062, + "neutral_observations": 512, + "residual_context_rms_ratio": 0.007725320290774107 + } + ], + "validation": { + "accuracy": 0.9921875, + "loss": 0.01907800231128931 + }, + "wall_seconds": 1.2659540176391602 + }, + { + "condition": "matched_raw", + "context_lesion_validation": { + "accuracy": 0.7578125, + "loss": 1.1658207774162292 + }, + "epoch_train_loss": [ + 0.33571012574248016, + 0.11975413211621344, + 0.11104183469433337, + 0.10416784440167248, + 0.09184010204626247, + 0.08320582384476438, + 0.07883103750646114, + 0.07580244168639183, + 0.06573321402538568, + 0.06347088766051456, + 0.06262386264279485, + 0.05813993178890087, + 0.061674059164943174, + 0.058328632556367666, + 0.05570228290162049, + 0.05565500335069373, + 0.056083728617522866, + 0.055472337000537664, + 0.04814719475689344, + 0.05002379676443525, + 0.050801149918697774, + 0.04792820388684049, + 0.04658058172208257, + 0.045018365170108154, + 0.045646410377230495, + 0.04163918021367863, + 0.04184562063892372, + 0.040560023349826224, + 0.03966468955331948, + 0.03746524166490417, + 0.03624598232272547, + 0.03414815798169002, + 0.03465788248286117, + 0.032305245971656404, + 0.03320501872804016, + 0.03220699344819877, + 0.03228929900797084, + 0.031194139643048402, + 0.03019473489257507, + 0.027403794330894016 + ], + "epochs_completed": 40, + "finite": true, + "first_nonfinite_epoch": null, + "predictor": [ + { + "context_rms": 0.6951092481613159, + "instruction_observations": 0, + "mean_per_cell_r2": 0.8674887418746948, + "neutral_observations": 512, + "residual_context_rms_ratio": 0.061199601739645004 + }, + { + "context_rms": 0.6736197471618652, + "instruction_observations": 0, + "mean_per_cell_r2": 0.6172494888305664, + "neutral_observations": 512, + "residual_context_rms_ratio": 0.24446305632591248 + } + ], + "validation": { + "accuracy": 0.99169921875, + "loss": 0.028344409074634314 + }, + "wall_seconds": 1.4177420139312744 + } + ], + "stage": "shared_feedback_s0", + "summary": { + "innovation_validation_accuracy_percent": 99.21875, + "matched_raw_validation_accuracy_percent": 99.169921875, + "oracle_context_lesion_drop_points": 22.998046875, + "oracle_validation_accuracy_percent": 99.31640625, + "raw_validation_accuracy_percent": 99.0234375 + } +} |