Getting embeddings for pairs of the top ~3k kanji pair combos (犬猫 but not 犬犬) around 10 million combos
gpu utilization is low cause I'm sending 1 request per pair (apparently ollama will handle it
anyway after you get the embeddings you can do stuff like this
;go run local_call_3.go
Sentence: 辛いものが好きな人もいれば嫌いな人もいる。まさに十人十色だ。
Target: 好きな人もいれば嫌いなひともいる
Matched: 誰嫌 (Similarity: 0.782
Sentence: 4chan is my favorite site!
Target: 4chan
Matched: 無梗 (Similarity: 0.748)
Sentence: 4chan is my favorite site!
Comment too long. Click here to view the full text.