Closed Dhanushkmr closed 8 months ago
reranker context length is only 512, AMR chunks can get as big as 2048 which won't fit. how best to approach this?
the current steps are: retrieve, do first level aggregation (max 512 tokens), rerank, then second level aggregation, then generate.
@hungchiayu1 @victoriachong
reranker context length is only 512, AMR chunks can get as big as 2048 which won't fit. how best to approach this?