Эксперимент: Confidence-Tested LoRA Pruning (#469)

{ "worked": true, "confidence": 8, "verdict": "Built Confidence-Tested LoRA Pruning utilities with Bartlett/Newey-West long-run variance estimation, one-sided lower-tail p-values, and fixed-budget component selection. The toy checks confirmed SE scaling near T^-1/2 (observed slope -0.496 vs -0.5), HAC variance inflation with autocorrelation (observed ratios 1.00, 2.14, 4.65, 9.35 vs predicted 1.00, 1.86, 4.00, 9.00), and increasing rejection power as the true contribution moved below Delta (0.0, 0.02, 0.18, 0.71). In the repeated 30-seed toy pruning proxy, statistical selection retained mean utility 3.624 versus 2.976 for latest-minibatch ranking and won in 93.3% of trials; this is promising evidence for the mechanism, not a claim of transformer-level improvement.", "metrics": { "baseline": "Latest-minibatch ranking: mean true retained utility 2.9756 over 30 seeds", "idea": "HAC confidence-tested ranking: mean true retained utility 3.6241 over 30 seeds; win rate 0.9333; mechanism predictions observed: SE slope -0.4959, HAC inflation [1.00, 2.14, 4.65, 9.35], rejection rates [0.00, 0.02, 0.18, 0.71]" }, "how_to_run": "/home/maxwelhelp/main/bin/python3 confidence_lora.py", "files": [ "confidence_lora.py", "results.json" ], "limitations": "No actual LoRA module, optimizer-gradient integration, DeBERTa/MNLI fine-tuning, CUDA benchmark, validation accuracy, matched-FLOP comparison, or three-seed transformer experiment was run. The toy AR(1) contribution process uses known synthetic component utilities and therefore does not establish downstream generalization." }