The encoder-only transformer model — architecture, BIO labeling, decoding, and inference.
View All TagsMost sequence labelers throw away the parts of an address they don't understand, and the output format is built to hide it. We made ours hand every dropped character back, then aggregated them into a ranked list. The top of the list turned out to be the Polish alphabet.
We're trained to be suspicious of a score that looks too good. The scores that say you failed get a free pass — and this week three of them were lying, including one that had us reporting kilometers for a product that lands in meters.
We taught our parser the hardest street cases it gets wrong, the aggregate score climbed, and we almost shipped it. Then a held-out probe caught it forgetting what a town is.
Mailwoman ships two address parsers in the same box.
Our failure taxonomy now ranks boundary instability first among the parser's weaknesses. We
Two days ago we retired a piece of decode-time implementation we'd been leaning on,
Last night our release check failed twice before it passed, and both failures had the
Yesterday we wrote about a lookup table that scored a perfect 100 and nearly talked us
This morning we published a post that ended with a tidy rule: some address tags don't
Our parser's international-order German 'collapsed' to 44%, so we retrained it three times. Then we measured the thing that can't be gamed and found out the model had been right all along. The metric was lying. Native German was sitting at 96%.