Add detailed timing logs to score_posterior_sample; warm up with multiple shapes

Fix consistency score always 0: add CONSISTENCY_QUANTILE_CONFIG_PATH to Dockerfile

Rescale Laplace posterior_std to target=0.1; recalibrate all quantile normalizers

Restrict LoRA to q_proj+v_proj on last 2 decoder layers only

Extract shared modules to eliminate code duplication

Add uncertainty scoring script for LoRA-Laplace backend

Fix Seq2SeqTrainer API for transformers 5.x

Add LoRA fine-tuning script for seq2seq models

Clarify calibration artifacts belong to MC Dropout backend

3 styles

3 styles

fix random seeds

Calibrate uncertainty quantiles from real MC Dropout scores

summarize

summarize

summarize

script wiki

script wiki

script wiki

scripts

pipenv

scripts

scripts

scripts