Metric computed with vector retrieval only (no Semantic ranking).
Retrieval Configuration | Recall@50
512 input tokens per vector | 42.4
1024 input tokens per vector | 37.5
4096 input tokens per vector | 36.4
8191 input tokens per vector | 34.9
512 input tokens per vector | 42.4
1024 input tokens per vector | 37.5
4096 input tokens per vector | 36.4
8191 input tokens per vector | 34.9
Table 4: Recall@50 comparison of different chunk sizes with Ada-002 embedding model