We rewrote one weight matrix to teach a 0.8B model conversation classificationgiga.ai·1 pts·namanbnsl·0