From a1c89749757082ae52c94ed53caf5b72dd782fd1 Mon Sep 17 00:00:00 2001 From: Matt Williams Date: Sun, 5 Nov 2023 16:13:48 -0800 Subject: [PATCH] also try other models Signed-off-by: Matt Williams --- examples/langchain-typescript-selfqueryingretrieval/readme.md | 2 ++ 1 file changed, 2 insertions(+) diff --git a/examples/langchain-typescript-selfqueryingretrieval/readme.md b/examples/langchain-typescript-selfqueryingretrieval/readme.md index 9b2f500be..5cf0b54f2 100644 --- a/examples/langchain-typescript-selfqueryingretrieval/readme.md +++ b/examples/langchain-typescript-selfqueryingretrieval/readme.md @@ -107,3 +107,5 @@ When you run this example, you will get a set of documents from the database tha To take this further, you could work on getting more out of the dataset. It turns out that this works best if there is only a single possible value for any given field. Our artists are often referred to by their last name, but sometimes using their full name. It may be Vincent van Gogh, or just van Gogh. Another way to get around this is to build a better query translator that knows that the search could be for a substring of the full name. But that also requires looking into the metadata searching capabilities of the database. Maybe it makes more sense to move the artist name and title of the work into the document itself. Then add some more metadata (there are at least 100 other attributes in the raw API that aren't used in this example.) + +Also try different models. In testing so far, it seems that codellama performs the best, but when a new model comes out you might try that. \ No newline at end of file