3 ms·
RAG breaks your 2k docs down and then provides only the relevant parts during query/answer time. Current context lengths should be more than sufficient 2000 te
by verdverm 2y ago
RAG breaks your 2k docs down and then provides only the relevant parts during query/answer time. Current context lengths should be more than sufficient
2000 texts is far too low a number to train from scratch. Fine-tuning is often used to refine a previously trained model, not sure 2k is enough for this.
RAG is really the first method you should be implementing. LlamaIndex has the best examples that are easily repurposed (imho)