From e18aa0169e80b8af9241e57f0ab66bf463ac8a4f Mon Sep 17 00:00:00 2001 From: sigoden Date: Mon, 11 Nov 2024 09:46:21 +0800 Subject: refactor: improve RAG (#982) - set max default_chuck_size to 2000 - set default rag_top_k to 4 - change reciprocal_rank_fusion weights --- models.yaml | 10 +++++----- 1 file changed, 5 insertions(+), 5 deletions(-) (limited to 'models.yaml') diff --git a/models.yaml b/models.yaml index de82a60..80e3fc7 100644 --- a/models.yaml +++ b/models.yaml @@ -66,13 +66,13 @@ type: embedding input_price: 0.13 max_tokens_per_chunk: 8191 - default_chunk_size: 3000 + default_chunk_size: 2000 max_batch_size: 100 - name: text-embedding-3-small type: embedding input_price: 0.02 max_tokens_per_chunk: 8191 - default_chunk_size: 3000 + default_chunk_size: 2000 max_batch_size: 100 # Links: @@ -1004,12 +1004,12 @@ - name: text-embedding-3-large type: embedding max_tokens_per_chunk: 8191 - default_chunk_size: 3000 + default_chunk_size: 2000 max_batch_size: 100 - name: text-embedding-3-small type: embedding max_tokens_per_chunk: 8191 - default_chunk_size: 3000 + default_chunk_size: 2000 max_batch_size: 100 - name: meta-llama-3.1-405b-instruct max_input_tokens: 128000 @@ -1714,7 +1714,7 @@ max_input_tokens: 120000 input_price: 0.12 max_tokens_per_chunk: 16000 - default_chunk_size: 3000 + default_chunk_size: 2000 max_batch_size: 128 - name: rerank-2 type: reranker -- cgit v1.2.3