Replace duplicated definitions with @braintrust/core. (#37)

We have factored out the shared deps between autoevals and sdk to the
`@braintrust/core` (JS) / `braintrust_core` (PY) packages, so we can
replace the redundant definitions in autoevals with imports from these
new packages.

.pre-commit-config.yaml

@@ -14,6 +14,7 @@ repos:
     rev: v0.0.282
     hooks:
       - id: ruff
+        args: [--fix, --exit-non-zero-on-fix]
   - repo: https://github.com/codespell-project/codespell
     rev: v2.2.1
     hooks:

js/index.ts

@@ -27,7 +27,7 @@
  * @module autoevals
  */
 
-export * from "./base.js";
+export { Score, ScorerArgs, Scorer } from "@braintrust/core";
 export * from "./llm.js";
 export * from "./string.js";
 export * from "./number.js";

js/json.ts

@@ -1,4 +1,4 @@
-import { Scorer } from "./base.js";
+import { Scorer } from "@braintrust/core";
 import { NumericDiff } from "./number.js";
 import { LevenshteinScorer } from "./string.js";
 

js/llm.ts

@@ -1,7 +1,7 @@
 import * as yaml from "js-yaml";
 import mustache from "mustache";
 
-import { Score, Scorer, ScorerArgs } from "./base.js";
+import { Score, Scorer, ScorerArgs } from "@braintrust/core";
 import { ChatCache, cachedChatCompletion } from "./oai.js";
 import { templates } from "./templates.js";
 import {

js/number.ts

@@ -1,4 +1,4 @@
-import { Scorer } from "./base.js";
+import { Scorer } from "@braintrust/core";
 
 /**
  * A simple scorer that compares numbers by normalizing their difference.

js/string.ts

@@ -1,4 +1,4 @@
-import { Scorer } from "./base.js";
+import { Scorer } from "@braintrust/core";
 import levenshtein from "js-levenshtein";
 
 /**

package.json

@@ -27,8 +27,8 @@
     "watch": "tsc --watch",
     "docs": "npx typedoc --options typedoc.json js/index.ts",
     "test": "jest",
-    "prepublishOnly": "./scripts/node_prepublish.py",
-    "postpublish": "./scripts/node_postpublish.py"
+    "prepublishOnly": "../../scripts/node_prepublish_autoevals.py",
+    "postpublish": "../../scripts/node_postpublish_autoevals.py"
   },
   "author": "",
   "license": "MIT",
@@ -45,6 +45,7 @@
     "typescript": "^5.1.6"
   },
   "dependencies": {
+    "@braintrust/core": "^0.0.6",
     "@types/node": "^20.4.4",
     "esbuild": "^0.19.1",
     "js-levenshtein": "^1.1.6",

py/autoevals/__init__.py

@@ -29,7 +29,8 @@
 
 """
 
-from .base import *
+from braintrust_core.score import Score, Scorer
+
 from .json import *
 from .llm import *
 from .number import *

py/autoevals/json.py

@@ -1,6 +1,7 @@
 import json
 
-from .base import Score, Scorer
+from braintrust_core.score import Score, Scorer
+
 from .number import NumericDiff
 from .string import Levenshtein
 

py/autoevals/llm.py

@@ -6,8 +6,8 @@
 
 import chevron
 import yaml
+from braintrust_core.score import Score, Scorer
 
-from .base import Score, Scorer
 from .oai import arun_cached_request, run_cached_request
 
 SCRIPT_DIR = os.path.dirname(os.path.abspath(__file__))

py/autoevals/number.py

@@ -1,4 +1,4 @@
-from .base import Score, Scorer
+from braintrust_core.score import Score, Scorer
 
 
 class NumericDiff(Scorer):

py/autoevals/string.py

@@ -1,9 +1,6 @@
+from braintrust_core.score import Score, Scorer
 from Levenshtein import distance
 
-from autoevals.base import Score
-
-from .base import Score, Scorer
-
 
 class Levenshtein(Scorer):
     """

setup.py

@@ -11,7 +11,7 @@
 with open(os.path.join(dir_name, "README.md"), "r", encoding="utf-8") as f:
     long_description = f.read()
 
-install_requires = ["chevron", "levenshtein", "pyyaml"]
+install_requires = ["chevron", "levenshtein", "pyyaml", "braintrust_core"]
 
 extras_require = {
     "dev": [