Rohit Ghumare 613d23cedd fix(phase-18/21): address CodeRabbit critical findings ...

train() shuffled data in place but indexed sample_weights by the
enumerate position on the shuffled list, so weights were misaligned
after the first shuffle. Pair each example with its weight before
shuffling so weights stay bound to their examples. Also avoids
mutating the caller's list.

2026-04-24 21:35:55 +01:00

..

main.py

fix(phase-18/21): address CodeRabbit critical findings

2026-04-24 21:35:55 +01:00