@empiricalrun/test-gen 0.31.2 → 0.31.3

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/CHANGELOG.md CHANGED
@@ -1,5 +1,13 @@
1
1
  # @empiricalrun/test-gen
2
2
 
3
+ ## 0.31.3
4
+
5
+ ### Patch Changes
6
+
7
+ - cce6ced: feat: base changes for enabling o1 & using it for update using master agent scenario
8
+ - Updated dependencies [cce6ced]
9
+ - @empiricalrun/llm@0.9.3
10
+
3
11
  ## 0.31.2
4
12
 
5
13
  ### Patch Changes
@@ -1 +1 @@
1
- {"version":3,"file":"update-flow.d.ts","sourceRoot":"","sources":["../../../src/agent/codegen/update-flow.ts"],"names":[],"mappings":"AAAA,OAAO,EAKL,WAAW,EACZ,MAAM,mBAAmB,CAAC;AAoB3B,OAAO,EAAE,QAAQ,EAAE,oBAAoB,EAAE,MAAM,aAAa,CAAC;AAI7D,KAAK,eAAe,GAAG,QAAQ,GAAG;IAChC,YAAY,EAAE,MAAM,EAAE,CAAC;CACxB,CAAC;AAqIF,wBAAsB,UAAU,CAC9B,QAAQ,EAAE,QAAQ,EAClB,IAAI,EAAE,MAAM,EACZ,OAAO,EAAE,oBAAoB,GAAG,SAAS,EACzC,OAAO,GAAE,OAAc,EACvB,QAAQ,GAAE,OAAc,GACvB,OAAO,CAAC,eAAe,EAAE,CAAC,CAuF5B;AAED,wBAAsB,qBAAqB,CAAC,EAC1C,QAAQ,EACR,IAAI,EACJ,OAAO,EACP,KAAK,EACL,aAAoB,GACrB,EAAE;IACD,QAAQ,EAAE,QAAQ,CAAC;IACnB,IAAI,EAAE,MAAM,CAAC;IACb,OAAO,CAAC,EAAE,oBAAoB,CAAC;IAC/B,KAAK,CAAC,EAAE,WAAW,CAAC;IACpB,aAAa,CAAC,EAAE,OAAO,CAAC;CACzB,GAAG,OAAO,CAAC,eAAe,EAAE,CAAC,CAwE7B"}
1
+ {"version":3,"file":"update-flow.d.ts","sourceRoot":"","sources":["../../../src/agent/codegen/update-flow.ts"],"names":[],"mappings":"AAAA,OAAO,EAKL,WAAW,EACZ,MAAM,mBAAmB,CAAC;AAoB3B,OAAO,EAAE,QAAQ,EAAE,oBAAoB,EAAE,MAAM,aAAa,CAAC;AAI7D,KAAK,eAAe,GAAG,QAAQ,GAAG;IAChC,YAAY,EAAE,MAAM,EAAE,CAAC;CACxB,CAAC;AAqIF,wBAAsB,UAAU,CAC9B,QAAQ,EAAE,QAAQ,EAClB,IAAI,EAAE,MAAM,EACZ,OAAO,EAAE,oBAAoB,GAAG,SAAS,EACzC,OAAO,GAAE,OAAc,EACvB,QAAQ,GAAE,OAAc,GACvB,OAAO,CAAC,eAAe,EAAE,CAAC,CAuF5B;AAED,wBAAsB,qBAAqB,CAAC,EAC1C,QAAQ,EACR,IAAI,EACJ,OAAO,EACP,KAAK,EACL,aAAoB,GACrB,EAAE;IACD,QAAQ,EAAE,QAAQ,CAAC;IACnB,IAAI,EAAE,MAAM,CAAC;IACb,OAAO,CAAC,EAAE,oBAAoB,CAAC;IAC/B,KAAK,CAAC,EAAE,WAAW,CAAC;IACpB,aAAa,CAAC,EAAE,OAAO,CAAC;CACzB,GAAG,OAAO,CAAC,eAAe,EAAE,CAAC,CAmF7B"}
@@ -221,19 +221,26 @@ async function appendCreateTestBlock({ testCase, file, options, trace, validateT
221
221
  scenarioName: testCase.name,
222
222
  scenarioSteps: testCase.steps.join("\n"),
223
223
  scenarioFile: file,
224
- });
224
+ }, 7);
225
225
  promptSpan.end({ output: { instruction } });
226
+ const [userInstruction] = instruction.filter((s) => s.role === "user");
227
+ const [systemInstruction] = instruction.filter((s) => s.role === "system");
228
+ userInstruction.content = `${systemInstruction?.content}
229
+
230
+ ${userInstruction?.content}
231
+ `;
226
232
  const llm = new llm_1.LLM({
227
233
  trace,
228
- provider: options?.modelProvider || constants_1.DEFAULT_MODEL_PROVIDER,
229
- defaultModel: options?.model || constants_1.DEFAULT_MODEL,
234
+ provider: "openai",
235
+ defaultModel: "o1-preview",
230
236
  providerApiKey: constants_1.MODEL_API_KEYS[options?.modelProvider || constants_1.DEFAULT_MODEL_PROVIDER],
231
237
  });
232
238
  const firstShotMessage = await llm.createChatCompletion({
233
- messages: instruction,
239
+ messages: [userInstruction],
234
240
  modelParameters: {
235
241
  ...constants_1.DEFAULT_MODEL_PARAMETERS,
236
242
  ...options?.modelParameters,
243
+ temperature: 1,
237
244
  },
238
245
  });
239
246
  let response = firstShotMessage?.content || "";
@@ -10,5 +10,5 @@ exports.DEFAULT_MODEL_PROVIDER = "openai";
10
10
  exports.DEFAULT_MODEL = "gpt-4o-2024-08-06";
11
11
  exports.DEFAULT_MODEL_PARAMETERS = {
12
12
  temperature: 0.5,
13
- max_tokens: 4000,
13
+ max_completion_tokens: 4000,
14
14
  };
@@ -1,4 +1,3 @@
1
- /// <reference types="node" />
2
1
  import { Reporter } from "@empiricalrun/reporter";
3
2
  type ReporterConfigType = {
4
3
  testSessionId: number;
@@ -1 +1 @@
1
- {"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/reporter/index.ts"],"names":[],"mappings":";AACA,OAAO,EAA4B,QAAQ,EAAE,MAAM,wBAAwB,CAAC;AAa5E,KAAK,kBAAkB,GAAG;IACxB,aAAa,EAAE,MAAM,CAAC;IACtB,YAAY,EAAE,MAAM,CAAC;IACrB,eAAe,EAAE,MAAM,CAAC;IACxB,YAAY,EAAE,MAAM,CAAC;IACrB,SAAS,EAAE,MAAM,CAAC;IAClB,UAAU,EAAE,MAAM,CAAC;CACpB,CAAC;AAKF,wBAAgB,WAAW,IAAI,QAAQ,GAAG,SAAS,CAUlD;AAED;;;;;;;;;GASG;AACH,wBAAgB,iBAAiB,CAAC,MAAM,EAAE,kBAAkB,GAAG,IAAI,CAGlE;AAED,qBAAa,sBAAsB;;IAE3B,YAAY,CAAC,KAAK,EAAE,MAAM,GAAG,OAAO,CAAC,IAAI,CAAC;IAK1C,eAAe,CAAC,EACpB,eAAe,EACf,QAAQ,GACT,EAAE;QACD,eAAe,EAAE,MAAM,CAAC;QACxB,QAAQ,EAAE,MAAM,CAAC;KAClB;IA6CK,eAAe,CAAC,MAAM,EAAE,MAAM,GAAG,OAAO,CAAC,IAAI,CAAC;IA6B9C,WAAW,CAAC,OAAO,EAAE,MAAM,GAAG,OAAO,CAAC,IAAI,CAAC;CAUlD"}
1
+ {"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/reporter/index.ts"],"names":[],"mappings":"AACA,OAAO,EAA4B,QAAQ,EAAE,MAAM,wBAAwB,CAAC;AAa5E,KAAK,kBAAkB,GAAG;IACxB,aAAa,EAAE,MAAM,CAAC;IACtB,YAAY,EAAE,MAAM,CAAC;IACrB,eAAe,EAAE,MAAM,CAAC;IACxB,YAAY,EAAE,MAAM,CAAC;IACrB,SAAS,EAAE,MAAM,CAAC;IAClB,UAAU,EAAE,MAAM,CAAC;CACpB,CAAC;AAKF,wBAAgB,WAAW,IAAI,QAAQ,GAAG,SAAS,CAUlD;AAED;;;;;;;;;GASG;AACH,wBAAgB,iBAAiB,CAAC,MAAM,EAAE,kBAAkB,GAAG,IAAI,CAGlE;AAED,qBAAa,sBAAsB;;IAE3B,YAAY,CAAC,KAAK,EAAE,MAAM,GAAG,OAAO,CAAC,IAAI,CAAC;IAK1C,eAAe,CAAC,EACpB,eAAe,EACf,QAAQ,GACT,EAAE;QACD,eAAe,EAAE,MAAM,CAAC;QACxB,QAAQ,EAAE,MAAM,CAAC;KAClB;IA6CK,eAAe,CAAC,MAAM,EAAE,MAAM,GAAG,OAAO,CAAC,IAAI,CAAC;IA6B9C,WAAW,CAAC,OAAO,EAAE,MAAM,GAAG,OAAO,CAAC,IAAI,CAAC;CAUlD"}
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@empiricalrun/test-gen",
3
- "version": "0.31.2",
3
+ "version": "0.31.3",
4
4
  "publishConfig": {
5
5
  "registry": "https://registry.npmjs.org/",
6
6
  "access": "public"
@@ -44,7 +44,7 @@
44
44
  "ts-morph": "^23.0.0",
45
45
  "tsx": "^4.16.2",
46
46
  "typescript": "^5.3.3",
47
- "@empiricalrun/llm": "^0.9.2",
47
+ "@empiricalrun/llm": "^0.9.3",
48
48
  "@empiricalrun/r2-uploader": "^0.3.3",
49
49
  "@empiricalrun/reporter": "^0.20.0"
50
50
  },