diff options
| author | YurenHao0426 <Blackhao0426@gmail.com> | 2026-07-22 17:24:19 -0500 |
|---|---|---|
| committer | YurenHao0426 <Blackhao0426@gmail.com> | 2026-07-22 17:24:19 -0500 |
| commit | 9bc58d6d33d33dfd10a0b7dbb90c77f18e0ff5e4 (patch) | |
| tree | 9af0472ca215ab4f1b2a837dffdecd5c1465d52b /results/kp_dynamic_projection_short_gate.json | |
| parent | f0fb6c7b8badc3c4556834c96ac150732d7570f1 (diff) | |
results: pass dynamic projection short validation
Diffstat (limited to 'results/kp_dynamic_projection_short_gate.json')
| -rw-r--r-- | results/kp_dynamic_projection_short_gate.json | 52 |
1 files changed, 52 insertions, 0 deletions
diff --git a/results/kp_dynamic_projection_short_gate.json b/results/kp_dynamic_projection_short_gate.json new file mode 100644 index 0000000..d10dbb1 --- /dev/null +++ b/results/kp_dynamic_projection_short_gate.json @@ -0,0 +1,52 @@ +{ + "checks": { + "accuracy_at_least_0p80": true, + "at_least_70_points_above_mt1_raw_and_matched": true, + "early_alignment_at_least_0p70": true, + "elementwise_and_neutral_cost_reported": true, + "epoch11_to20_feedback_cosine_at_least_0p70": true, + "final_feedback_cosine_at_least_0p80": true, + "initial_traffic_ratio_error_at_most_1e_minus_5": true, + "macs_at_most_1p34x_bp": true, + "one_frozen_64_example_slow_fit": true, + "one_validation_and_zero_test_evaluations": true, + "peak_allocated_memory_at_most_2p5_gib": true, + "post_projection_soma_slope_at_most_1e_minus_5": true, + "post_projection_traffic_ratio_at_most_1e_minus_5": true, + "projection_observes_each_ordinary_example": true, + "record_trajectory_and_diagnostics_finite": true, + "used_instruction_rms_ratio_within_1e_minus_4": true, + "within_3_points_of_clean_kp_short": true, + "zero_instruction_observations_in_fast_fit": true, + "zero_task_loss_queries": true + }, + "confirmation_test_seeds_touched": false, + "full_validation_opened": true, + "metrics": { + "accuracy": 0.8358, + "clean_kp_short_accuracy": 0.8266, + "early_third_alignment": 0.8838296929995219, + "elementwise_operations_estimate": 4070171475968, + "epoch11_to20_feedback_forward_cosine": 0.860594035449781, + "final_feedback_forward_cosine": 0.9040563545728985, + "gain_over_mt1": { + "matched": 0.7358, + "raw": 0.7358 + }, + "loss": 0.5113223815917969, + "mac_ratio_to_bp": 1.3261078696686301, + "maximum_post_projection_soma_slope": 1.4347571131168024e-08, + "maximum_post_projection_traffic_rms_ratio": 3.033539150096096e-08, + "maximum_pre_projection_traffic_rms_ratio": 0.009560643492622747, + "maximum_used_instruction_rms_ratio_error": 4.663103236879351e-10, + "peak_memory_allocated_bytes": 2165061120, + "source_commit": "f0fb6c7b8badc3c4556834c96ac150732d7570f1", + "total_macs": 145192643821568, + "wall_s": 816.9503653049469 + }, + "protocol": "kp_dynamic_neutral_projection_short_v1", + "review_score_after": 5, + "review_score_before": 5, + "score_change_rule": "a one-seed short validation endpoint cannot raise the score", + "status": "passed" +} |
