From 2071a3d9919583485a8914fda026241f9f91d9c2 Mon Sep 17 00:00:00 2001 From: Arjun Attam Date: Tue, 16 Apr 2024 17:54:55 +0530 Subject: [PATCH 1/5] chore: update spider example to avoid rate limit logs --- examples/spider/empiricalrc.json | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/examples/spider/empiricalrc.json b/examples/spider/empiricalrc.json index 87b8ac17..419ce1c7 100644 --- a/examples/spider/empiricalrc.json +++ b/examples/spider/empiricalrc.json @@ -34,7 +34,7 @@ { "type": "model", "provider": "google", - "model": "gemini-1.5-pro-latest", + "model": "gemini-1.0-pro", "prompt": "You are an SQLite expert who can convert natural language questions to SQL queries for the database schema given below.\n\nDatabase schema:\n{{schema}}\n\nAnswer the following question with only the SQL query.\n\nQuestion: {{question}}", "scorers": [ { From 1ae48bc634e0d590a6371d7292ffe4f02749e09e Mon Sep 17 00:00:00 2001 From: Arjun Attam Date: Tue, 16 Apr 2024 17:58:10 +0530 Subject: [PATCH 2/5] sql message --- packages/scorer/src/provider/deterministic/sql.ts | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/packages/scorer/src/provider/deterministic/sql.ts b/packages/scorer/src/provider/deterministic/sql.ts index 21108dac..2a8f5fe0 100644 --- a/packages/scorer/src/provider/deterministic/sql.ts +++ b/packages/scorer/src/provider/deterministic/sql.ts @@ -28,7 +28,7 @@ export const checkSqlSyntax: ScoringFn = async ({ output }) => { { score: isSQLQuery ? 1 : 0, name: syntaxName, - message: isSQLQuery ? "" : errorMsg, + message: isSQLQuery ? "SQL is valid" : errorMsg, }, ]; }; From 190d30949f1e6f9e371ad05c43a27d040f4965d7 Mon Sep 17 00:00:00 2001 From: Arjun Attam Date: Tue, 16 Apr 2024 18:11:43 +0530 Subject: [PATCH 3/5] fix sql message --- packages/scorer/src/provider/deterministic/sql.test.ts | 4 ++-- packages/scorer/src/provider/deterministic/sql.ts | 2 +- 2 files changed, 3 insertions(+), 3 deletions(-) diff --git a/packages/scorer/src/provider/deterministic/sql.test.ts b/packages/scorer/src/provider/deterministic/sql.test.ts index 0dc5a2d9..2cc54405 100644 --- a/packages/scorer/src/provider/deterministic/sql.test.ts +++ b/packages/scorer/src/provider/deterministic/sql.test.ts @@ -16,7 +16,7 @@ test("sql-syntax works with backticks", async () => { { score: 1, name: "sql-syntax", - message: "", + message: "Output is valid SQL", }, ]); }); @@ -85,7 +85,7 @@ WHERE stadium_id = ( { score: 1, name: "sql-syntax", - message: "", + message: "Output is valid SQL", }, ]); }); diff --git a/packages/scorer/src/provider/deterministic/sql.ts b/packages/scorer/src/provider/deterministic/sql.ts index 2a8f5fe0..a4d86c80 100644 --- a/packages/scorer/src/provider/deterministic/sql.ts +++ b/packages/scorer/src/provider/deterministic/sql.ts @@ -28,7 +28,7 @@ export const checkSqlSyntax: ScoringFn = async ({ output }) => { { score: isSQLQuery ? 1 : 0, name: syntaxName, - message: isSQLQuery ? "SQL is valid" : errorMsg, + message: isSQLQuery ? "Output is valid SQL" : errorMsg, }, ]; }; From d577eaec2e0e7611ed0ed5439383cc443f7a4f20 Mon Sep 17 00:00:00 2001 From: Arjun Attam Date: Tue, 16 Apr 2024 18:12:13 +0530 Subject: [PATCH 4/5] cs --- .changeset/tall-meals-smile.md | 5 +++++ 1 file changed, 5 insertions(+) create mode 100644 .changeset/tall-meals-smile.md diff --git a/.changeset/tall-meals-smile.md b/.changeset/tall-meals-smile.md new file mode 100644 index 00000000..edc17077 --- /dev/null +++ b/.changeset/tall-meals-smile.md @@ -0,0 +1,5 @@ +--- +"@empiricalrun/scorer": patch +--- + +chore: improve message string for valid sql-syntax From 01e5402939139375dc017f9f16801250dbd7958e Mon Sep 17 00:00:00 2001 From: Arjun Attam Date: Tue, 16 Apr 2024 18:37:22 +0530 Subject: [PATCH 5/5] header copy --- apps/web/components/ui/page-header.tsx | 5 ++--- packages/cli/src/bin/index.ts | 2 +- 2 files changed, 3 insertions(+), 4 deletions(-) diff --git a/apps/web/components/ui/page-header.tsx b/apps/web/components/ui/page-header.tsx index d1f33b26..c4e6a451 100644 --- a/apps/web/components/ui/page-header.tsx +++ b/apps/web/components/ui/page-header.tsx @@ -13,9 +13,8 @@ export function PageHeader() {
- - compare and evaluate multiple AI model completions on different - prompts and model + + Compare and evaluate AI models across all the scenarios that matter
diff --git a/packages/cli/src/bin/index.ts b/packages/cli/src/bin/index.ts index 0a8510e8..c12e7c2b 100644 --- a/packages/cli/src/bin/index.ts +++ b/packages/cli/src/bin/index.ts @@ -44,7 +44,7 @@ const outputFilePath = `${cwd}/${cacheDir}/${outputFileName}`; program .name("Empirical.run CLI") .description( - "CLI to compare and evaluate multiple AI model completions on different prompts and models", + "CLI to compare and evaluate AI models across all the scenarios that matter", ) .version(packageJSON.version);