@agentuity/evals 1.0.31 → 1.0.32
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/dist/adversarial.js +1 -1
- package/dist/adversarial.js.map +1 -1
- package/dist/ambiguity.js +1 -1
- package/dist/ambiguity.js.map +1 -1
- package/dist/answer-completeness.js +1 -1
- package/dist/answer-completeness.js.map +1 -1
- package/dist/conciseness.js +1 -1
- package/dist/conciseness.js.map +1 -1
- package/dist/extraneous-content.js +1 -1
- package/dist/extraneous-content.js.map +1 -1
- package/dist/format.js +1 -1
- package/dist/format.js.map +1 -1
- package/dist/knowledge-retention.js +1 -1
- package/dist/knowledge-retention.js.map +1 -1
- package/dist/pii.js +1 -1
- package/dist/pii.js.map +1 -1
- package/dist/politeness.js +1 -1
- package/dist/politeness.js.map +1 -1
- package/dist/role-adherence.js +1 -1
- package/dist/role-adherence.js.map +1 -1
- package/dist/safety.js +1 -1
- package/dist/safety.js.map +1 -1
- package/dist/self-reference.js +1 -1
- package/dist/self-reference.js.map +1 -1
- package/package.json +5 -5
- package/src/adversarial.ts +1 -1
- package/src/ambiguity.ts +1 -1
- package/src/answer-completeness.ts +1 -1
- package/src/conciseness.ts +1 -1
- package/src/extraneous-content.ts +1 -1
- package/src/format.ts +1 -1
- package/src/knowledge-retention.ts +1 -1
- package/src/pii.ts +1 -1
- package/src/politeness.ts +1 -1
- package/src/role-adherence.ts +1 -1
- package/src/safety.ts +1 -1
- package/src/self-reference.ts +1 -1
package/dist/adversarial.js
CHANGED
|
@@ -53,7 +53,7 @@ export const adversarial = createPresetEval({
|
|
|
53
53
|
options: {
|
|
54
54
|
model: DEFAULT_EVAL_MODEL,
|
|
55
55
|
},
|
|
56
|
-
handler: async (
|
|
56
|
+
handler: async (_ctx, input, output, options) => {
|
|
57
57
|
const prompt = interpolatePrompt(adversarialPrompt, {
|
|
58
58
|
USER_REQUEST: input.request,
|
|
59
59
|
MODEL_RESPONSE: output.response,
|
package/dist/adversarial.js.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"adversarial.js","sourceRoot":"","sources":["../src/adversarial.ts"],"names":[],"mappings":"AAAA,OAAO,EACN,gBAAgB,EAChB,iBAAiB,EACjB,kBAAkB,GAGlB,MAAM,UAAU,CAAC;AAClB,OAAO,EAAE,kBAAkB,EAAwB,MAAM,SAAS,CAAC;AAEnE,MAAM,CAAC,MAAM,iBAAiB,GAAG;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;EA8C/B,CAAC;AAEH,MAAM,CAAC,MAAM,WAAW,GAAG,gBAAgB,CAAuD;IACjG,IAAI,EAAE,aAAa;IACnB,WAAW,EAAE,sEAAsE;IACnF,OAAO,EAAE;QACR,KAAK,EAAE,kBAAkB;KACzB;IACD,OAAO,EAAE,KAAK,EAAE,
|
|
1
|
+
{"version":3,"file":"adversarial.js","sourceRoot":"","sources":["../src/adversarial.ts"],"names":[],"mappings":"AAAA,OAAO,EACN,gBAAgB,EAChB,iBAAiB,EACjB,kBAAkB,GAGlB,MAAM,UAAU,CAAC;AAClB,OAAO,EAAE,kBAAkB,EAAwB,MAAM,SAAS,CAAC;AAEnE,MAAM,CAAC,MAAM,iBAAiB,GAAG;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;EA8C/B,CAAC;AAEH,MAAM,CAAC,MAAM,WAAW,GAAG,gBAAgB,CAAuD;IACjG,IAAI,EAAE,aAAa;IACnB,WAAW,EAAE,sEAAsE;IACnF,OAAO,EAAE;QACR,KAAK,EAAE,kBAAkB;KACzB;IACD,OAAO,EAAE,KAAK,EAAE,IAAI,EAAE,KAAK,EAAE,MAAM,EAAE,OAAO,EAAE,EAAE;QAC/C,MAAM,MAAM,GAAG,iBAAiB,CAAC,iBAAiB,EAAE;YACnD,YAAY,EAAE,KAAK,CAAC,OAAO;YAC3B,cAAc,EAAE,MAAM,CAAC,QAAQ;SAC/B,CAAC,CAAC;QAEH,OAAO,kBAAkB,CAAC,EAAE,KAAK,EAAE,OAAO,CAAC,KAAK,EAAE,MAAM,EAAE,CAAC,CAAC;IAC7D,CAAC;CACD,CAAC,CAAC"}
|
package/dist/ambiguity.js
CHANGED
|
@@ -55,7 +55,7 @@ export const ambiguity = createPresetEval({
|
|
|
55
55
|
model: DEFAULT_EVAL_MODEL,
|
|
56
56
|
threshold: 0.7,
|
|
57
57
|
},
|
|
58
|
-
handler: async (
|
|
58
|
+
handler: async (_ctx, input, output, options) => {
|
|
59
59
|
const prompt = interpolatePrompt(ambiguityPrompt, {
|
|
60
60
|
USER_REQUEST: input.request,
|
|
61
61
|
MODEL_RESPONSE: output.response,
|
package/dist/ambiguity.js.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"ambiguity.js","sourceRoot":"","sources":["../src/ambiguity.ts"],"names":[],"mappings":"AAAA,OAAO,EACN,gBAAgB,EAChB,iBAAiB,EACjB,kBAAkB,GAGlB,MAAM,UAAU,CAAC;AAClB,OAAO,EAAE,kBAAkB,EAAwB,MAAM,SAAS,CAAC;AAEnE,MAAM,CAAC,MAAM,eAAe,GAAG;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;EA+C7B,CAAC;AAMH,MAAM,CAAC,MAAM,SAAS,GAAG,gBAAgB,CAIvC;IACD,IAAI,EAAE,WAAW;IACjB,WAAW,EACV,qFAAqF;IACtF,OAAO,EAAE;QACR,KAAK,EAAE,kBAAkB;QACzB,SAAS,EAAE,GAAG;KACd;IACD,OAAO,EAAE,KAAK,EAAE,
|
|
1
|
+
{"version":3,"file":"ambiguity.js","sourceRoot":"","sources":["../src/ambiguity.ts"],"names":[],"mappings":"AAAA,OAAO,EACN,gBAAgB,EAChB,iBAAiB,EACjB,kBAAkB,GAGlB,MAAM,UAAU,CAAC;AAClB,OAAO,EAAE,kBAAkB,EAAwB,MAAM,SAAS,CAAC;AAEnE,MAAM,CAAC,MAAM,eAAe,GAAG;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;EA+C7B,CAAC;AAMH,MAAM,CAAC,MAAM,SAAS,GAAG,gBAAgB,CAIvC;IACD,IAAI,EAAE,WAAW;IACjB,WAAW,EACV,qFAAqF;IACtF,OAAO,EAAE;QACR,KAAK,EAAE,kBAAkB;QACzB,SAAS,EAAE,GAAG;KACd;IACD,OAAO,EAAE,KAAK,EAAE,IAAI,EAAE,KAAK,EAAE,MAAM,EAAE,OAAO,EAAE,EAAE;QAC/C,MAAM,MAAM,GAAG,iBAAiB,CAAC,eAAe,EAAE;YACjD,YAAY,EAAE,KAAK,CAAC,OAAO;YAC3B,cAAc,EAAE,MAAM,CAAC,QAAQ;SAC/B,CAAC,CAAC;QAEH,MAAM,UAAU,GAAG,MAAM,kBAAkB,CAAC,EAAE,KAAK,EAAE,OAAO,CAAC,KAAK,EAAE,MAAM,EAAE,CAAC,CAAC;QAE9E,OAAO;YACN,GAAG,UAAU;YACb,MAAM,EAAE,UAAU,CAAC,MAAM,IAAI,CAAC,UAAU,CAAC,KAAK,IAAI,CAAC,CAAC,IAAI,OAAO,CAAC,SAAS;SACzE,CAAC;IACH,CAAC;CACD,CAAC,CAAC"}
|
|
@@ -57,7 +57,7 @@ export const answerCompleteness = createPresetEval({
|
|
|
57
57
|
model: DEFAULT_EVAL_MODEL,
|
|
58
58
|
threshold: 0.7,
|
|
59
59
|
},
|
|
60
|
-
handler: async (
|
|
60
|
+
handler: async (_ctx, input, output, options) => {
|
|
61
61
|
const prompt = interpolatePrompt(answerCompletenessPrompt, {
|
|
62
62
|
USER_REQUEST: input.request,
|
|
63
63
|
MODEL_RESPONSE: output.response,
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"answer-completeness.js","sourceRoot":"","sources":["../src/answer-completeness.ts"],"names":[],"mappings":"AAAA,OAAO,EACN,gBAAgB,EAChB,iBAAiB,EACjB,kBAAkB,GAGlB,MAAM,UAAU,CAAC;AAClB,OAAO,EAAE,kBAAkB,EAAwB,MAAM,SAAS,CAAC;AAEnE,MAAM,CAAC,MAAM,wBAAwB,GAAG;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;EAiDtC,CAAC;AAMH,MAAM,CAAC,MAAM,kBAAkB,GAAG,gBAAgB,CAIhD;IACD,IAAI,EAAE,qBAAqB;IAC3B,WAAW,EAAE,0EAA0E;IACvF,OAAO,EAAE;QACR,KAAK,EAAE,kBAAkB;QACzB,SAAS,EAAE,GAAG;KACd;IACD,OAAO,EAAE,KAAK,EAAE,
|
|
1
|
+
{"version":3,"file":"answer-completeness.js","sourceRoot":"","sources":["../src/answer-completeness.ts"],"names":[],"mappings":"AAAA,OAAO,EACN,gBAAgB,EAChB,iBAAiB,EACjB,kBAAkB,GAGlB,MAAM,UAAU,CAAC;AAClB,OAAO,EAAE,kBAAkB,EAAwB,MAAM,SAAS,CAAC;AAEnE,MAAM,CAAC,MAAM,wBAAwB,GAAG;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;EAiDtC,CAAC;AAMH,MAAM,CAAC,MAAM,kBAAkB,GAAG,gBAAgB,CAIhD;IACD,IAAI,EAAE,qBAAqB;IAC3B,WAAW,EAAE,0EAA0E;IACvF,OAAO,EAAE;QACR,KAAK,EAAE,kBAAkB;QACzB,SAAS,EAAE,GAAG;KACd;IACD,OAAO,EAAE,KAAK,EAAE,IAAI,EAAE,KAAK,EAAE,MAAM,EAAE,OAAO,EAAE,EAAE;QAC/C,MAAM,MAAM,GAAG,iBAAiB,CAAC,wBAAwB,EAAE;YAC1D,YAAY,EAAE,KAAK,CAAC,OAAO;YAC3B,cAAc,EAAE,MAAM,CAAC,QAAQ;SAC/B,CAAC,CAAC;QAEH,MAAM,UAAU,GAAG,MAAM,kBAAkB,CAAC,EAAE,KAAK,EAAE,OAAO,CAAC,KAAK,EAAE,MAAM,EAAE,CAAC,CAAC;QAE9E,OAAO;YACN,GAAG,UAAU;YACb,MAAM,EAAE,UAAU,CAAC,MAAM,IAAI,CAAC,UAAU,CAAC,KAAK,IAAI,CAAC,CAAC,IAAI,OAAO,CAAC,SAAS;SACzE,CAAC;IACH,CAAC;CACD,CAAC,CAAC"}
|
package/dist/conciseness.js
CHANGED
|
@@ -56,7 +56,7 @@ export const conciseness = createPresetEval({
|
|
|
56
56
|
model: DEFAULT_EVAL_MODEL,
|
|
57
57
|
threshold: 0.7,
|
|
58
58
|
},
|
|
59
|
-
handler: async (
|
|
59
|
+
handler: async (_ctx, input, output, options) => {
|
|
60
60
|
const prompt = interpolatePrompt(concisenessPrompt, {
|
|
61
61
|
USER_REQUEST: input.request,
|
|
62
62
|
MODEL_RESPONSE: output.response,
|
package/dist/conciseness.js.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"conciseness.js","sourceRoot":"","sources":["../src/conciseness.ts"],"names":[],"mappings":"AAAA,OAAO,EACN,gBAAgB,EAChB,iBAAiB,EACjB,kBAAkB,GAGlB,MAAM,UAAU,CAAC;AAClB,OAAO,EAAE,kBAAkB,EAAwB,MAAM,SAAS,CAAC;AAEnE,MAAM,CAAC,MAAM,iBAAiB,GAAG;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;EAgD/B,CAAC;AAMH,MAAM,CAAC,MAAM,WAAW,GAAG,gBAAgB,CAIzC;IACD,IAAI,EAAE,aAAa;IACnB,WAAW,EAAE,mFAAmF;IAChG,OAAO,EAAE;QACR,KAAK,EAAE,kBAAkB;QACzB,SAAS,EAAE,GAAG;KACd;IACD,OAAO,EAAE,KAAK,EAAE,
|
|
1
|
+
{"version":3,"file":"conciseness.js","sourceRoot":"","sources":["../src/conciseness.ts"],"names":[],"mappings":"AAAA,OAAO,EACN,gBAAgB,EAChB,iBAAiB,EACjB,kBAAkB,GAGlB,MAAM,UAAU,CAAC;AAClB,OAAO,EAAE,kBAAkB,EAAwB,MAAM,SAAS,CAAC;AAEnE,MAAM,CAAC,MAAM,iBAAiB,GAAG;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;EAgD/B,CAAC;AAMH,MAAM,CAAC,MAAM,WAAW,GAAG,gBAAgB,CAIzC;IACD,IAAI,EAAE,aAAa;IACnB,WAAW,EAAE,mFAAmF;IAChG,OAAO,EAAE;QACR,KAAK,EAAE,kBAAkB;QACzB,SAAS,EAAE,GAAG;KACd;IACD,OAAO,EAAE,KAAK,EAAE,IAAI,EAAE,KAAK,EAAE,MAAM,EAAE,OAAO,EAAE,EAAE;QAC/C,MAAM,MAAM,GAAG,iBAAiB,CAAC,iBAAiB,EAAE;YACnD,YAAY,EAAE,KAAK,CAAC,OAAO;YAC3B,cAAc,EAAE,MAAM,CAAC,QAAQ;SAC/B,CAAC,CAAC;QAEH,MAAM,UAAU,GAAG,MAAM,kBAAkB,CAAC,EAAE,KAAK,EAAE,OAAO,CAAC,KAAK,EAAE,MAAM,EAAE,CAAC,CAAC;QAE9E,OAAO;YACN,GAAG,UAAU;YACb,MAAM,EAAE,UAAU,CAAC,MAAM,IAAI,CAAC,UAAU,CAAC,KAAK,IAAI,CAAC,CAAC,IAAI,OAAO,CAAC,SAAS;SACzE,CAAC;IACH,CAAC;CACD,CAAC,CAAC"}
|
|
@@ -58,7 +58,7 @@ export const extraneousContent = createPresetEval({
|
|
|
58
58
|
model: DEFAULT_EVAL_MODEL,
|
|
59
59
|
threshold: 0.7,
|
|
60
60
|
},
|
|
61
|
-
handler: async (
|
|
61
|
+
handler: async (_ctx, input, output, options) => {
|
|
62
62
|
const prompt = interpolatePrompt(extraneousContentPrompt, {
|
|
63
63
|
USER_REQUEST: input.request,
|
|
64
64
|
MODEL_RESPONSE: output.response,
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"extraneous-content.js","sourceRoot":"","sources":["../src/extraneous-content.ts"],"names":[],"mappings":"AAAA,OAAO,EACN,gBAAgB,EAChB,iBAAiB,EACjB,kBAAkB,GAGlB,MAAM,UAAU,CAAC;AAClB,OAAO,EAAE,kBAAkB,EAAwB,MAAM,SAAS,CAAC;AAEnE,MAAM,CAAC,MAAM,uBAAuB,GAAG;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;EAkDrC,CAAC;AAMH,MAAM,CAAC,MAAM,iBAAiB,GAAG,gBAAgB,CAI/C;IACD,IAAI,EAAE,oBAAoB;IAC1B,WAAW,EAAE,sEAAsE;IACnF,OAAO,EAAE;QACR,KAAK,EAAE,kBAAkB;QACzB,SAAS,EAAE,GAAG;KACd;IACD,OAAO,EAAE,KAAK,EAAE,
|
|
1
|
+
{"version":3,"file":"extraneous-content.js","sourceRoot":"","sources":["../src/extraneous-content.ts"],"names":[],"mappings":"AAAA,OAAO,EACN,gBAAgB,EAChB,iBAAiB,EACjB,kBAAkB,GAGlB,MAAM,UAAU,CAAC;AAClB,OAAO,EAAE,kBAAkB,EAAwB,MAAM,SAAS,CAAC;AAEnE,MAAM,CAAC,MAAM,uBAAuB,GAAG;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;EAkDrC,CAAC;AAMH,MAAM,CAAC,MAAM,iBAAiB,GAAG,gBAAgB,CAI/C;IACD,IAAI,EAAE,oBAAoB;IAC1B,WAAW,EAAE,sEAAsE;IACnF,OAAO,EAAE;QACR,KAAK,EAAE,kBAAkB;QACzB,SAAS,EAAE,GAAG;KACd;IACD,OAAO,EAAE,KAAK,EAAE,IAAI,EAAE,KAAK,EAAE,MAAM,EAAE,OAAO,EAAE,EAAE;QAC/C,MAAM,MAAM,GAAG,iBAAiB,CAAC,uBAAuB,EAAE;YACzD,YAAY,EAAE,KAAK,CAAC,OAAO;YAC3B,cAAc,EAAE,MAAM,CAAC,QAAQ;YAC/B,SAAS,EAAE,OAAO,CAAC,SAAS,CAAC,QAAQ,EAAE;SACvC,CAAC,CAAC;QAEH,MAAM,UAAU,GAAG,MAAM,kBAAkB,CAAC,EAAE,KAAK,EAAE,OAAO,CAAC,KAAK,EAAE,MAAM,EAAE,CAAC,CAAC;QAE9E,OAAO;YACN,GAAG,UAAU;YACb,MAAM,EAAE,UAAU,CAAC,MAAM,IAAI,CAAC,UAAU,CAAC,KAAK,IAAI,CAAC,CAAC,IAAI,OAAO,CAAC,SAAS;SACzE,CAAC;IACH,CAAC;CACD,CAAC,CAAC"}
|
package/dist/format.js
CHANGED
|
@@ -51,7 +51,7 @@ export const format = createPresetEval({
|
|
|
51
51
|
options: {
|
|
52
52
|
model: DEFAULT_EVAL_MODEL,
|
|
53
53
|
},
|
|
54
|
-
handler: async (
|
|
54
|
+
handler: async (_ctx, input, output, options) => {
|
|
55
55
|
const prompt = interpolatePrompt(formatPrompt, {
|
|
56
56
|
USER_REQUEST: input.request,
|
|
57
57
|
MODEL_RESPONSE: output.response,
|
package/dist/format.js.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"format.js","sourceRoot":"","sources":["../src/format.ts"],"names":[],"mappings":"AAAA,OAAO,EACN,gBAAgB,EAChB,iBAAiB,EACjB,kBAAkB,GAGlB,MAAM,UAAU,CAAC;AAClB,OAAO,EAAE,kBAAkB,EAAwB,MAAM,SAAS,CAAC;AAEnE,MAAM,CAAC,MAAM,YAAY,GAAG;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;EA4C1B,CAAC;AAEH,MAAM,CAAC,MAAM,MAAM,GAAG,gBAAgB,CAAuD;IAC5F,IAAI,EAAE,QAAQ;IACd,WAAW,EAAE,yDAAyD;IACtE,OAAO,EAAE;QACR,KAAK,EAAE,kBAAkB;KACzB;IACD,OAAO,EAAE,KAAK,EAAE,
|
|
1
|
+
{"version":3,"file":"format.js","sourceRoot":"","sources":["../src/format.ts"],"names":[],"mappings":"AAAA,OAAO,EACN,gBAAgB,EAChB,iBAAiB,EACjB,kBAAkB,GAGlB,MAAM,UAAU,CAAC;AAClB,OAAO,EAAE,kBAAkB,EAAwB,MAAM,SAAS,CAAC;AAEnE,MAAM,CAAC,MAAM,YAAY,GAAG;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;EA4C1B,CAAC;AAEH,MAAM,CAAC,MAAM,MAAM,GAAG,gBAAgB,CAAuD;IAC5F,IAAI,EAAE,QAAQ;IACd,WAAW,EAAE,yDAAyD;IACtE,OAAO,EAAE;QACR,KAAK,EAAE,kBAAkB;KACzB;IACD,OAAO,EAAE,KAAK,EAAE,IAAI,EAAE,KAAK,EAAE,MAAM,EAAE,OAAO,EAAE,EAAE;QAC/C,MAAM,MAAM,GAAG,iBAAiB,CAAC,YAAY,EAAE;YAC9C,YAAY,EAAE,KAAK,CAAC,OAAO;YAC3B,cAAc,EAAE,MAAM,CAAC,QAAQ;SAC/B,CAAC,CAAC;QAEH,OAAO,kBAAkB,CAAC,EAAE,KAAK,EAAE,OAAO,CAAC,KAAK,EAAE,MAAM,EAAE,CAAC,CAAC;IAC7D,CAAC;CACD,CAAC,CAAC"}
|
|
@@ -61,7 +61,7 @@ export const knowledgeRetention = createPresetEval({
|
|
|
61
61
|
model: DEFAULT_EVAL_MODEL,
|
|
62
62
|
threshold: 0.7,
|
|
63
63
|
},
|
|
64
|
-
handler: async (
|
|
64
|
+
handler: async (_ctx, input, output, options) => {
|
|
65
65
|
const prompt = interpolatePrompt(knowledgeRetentionPrompt, {
|
|
66
66
|
USER_REQUEST: input.request,
|
|
67
67
|
MODEL_RESPONSE: output.response,
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"knowledge-retention.js","sourceRoot":"","sources":["../src/knowledge-retention.ts"],"names":[],"mappings":"AAAA,OAAO,EACN,gBAAgB,EAChB,iBAAiB,EACjB,kBAAkB,GAGlB,MAAM,UAAU,CAAC;AAClB,OAAO,EAAE,kBAAkB,EAAwB,MAAM,SAAS,CAAC;AAEnE,MAAM,CAAC,MAAM,wBAAwB,GAAG;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;EAqDtC,CAAC;AAMH,MAAM,CAAC,MAAM,kBAAkB,GAAG,gBAAgB,CAIhD;IACD,IAAI,EAAE,qBAAqB;IAC3B,WAAW,EAAE,mFAAmF;IAChG,OAAO,EAAE;QACR,KAAK,EAAE,kBAAkB;QACzB,SAAS,EAAE,GAAG;KACd;IACD,OAAO,EAAE,KAAK,EAAE,
|
|
1
|
+
{"version":3,"file":"knowledge-retention.js","sourceRoot":"","sources":["../src/knowledge-retention.ts"],"names":[],"mappings":"AAAA,OAAO,EACN,gBAAgB,EAChB,iBAAiB,EACjB,kBAAkB,GAGlB,MAAM,UAAU,CAAC;AAClB,OAAO,EAAE,kBAAkB,EAAwB,MAAM,SAAS,CAAC;AAEnE,MAAM,CAAC,MAAM,wBAAwB,GAAG;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;EAqDtC,CAAC;AAMH,MAAM,CAAC,MAAM,kBAAkB,GAAG,gBAAgB,CAIhD;IACD,IAAI,EAAE,qBAAqB;IAC3B,WAAW,EAAE,mFAAmF;IAChG,OAAO,EAAE;QACR,KAAK,EAAE,kBAAkB;QACzB,SAAS,EAAE,GAAG;KACd;IACD,OAAO,EAAE,KAAK,EAAE,IAAI,EAAE,KAAK,EAAE,MAAM,EAAE,OAAO,EAAE,EAAE;QAC/C,MAAM,MAAM,GAAG,iBAAiB,CAAC,wBAAwB,EAAE;YAC1D,YAAY,EAAE,KAAK,CAAC,OAAO;YAC3B,cAAc,EAAE,MAAM,CAAC,QAAQ;SAC/B,CAAC,CAAC;QAEH,MAAM,UAAU,GAAG,MAAM,kBAAkB,CAAC,EAAE,KAAK,EAAE,OAAO,CAAC,KAAK,EAAE,MAAM,EAAE,CAAC,CAAC;QAE9E,OAAO;YACN,GAAG,UAAU;YACb,MAAM,EAAE,UAAU,CAAC,MAAM,IAAI,CAAC,UAAU,CAAC,KAAK,IAAI,CAAC,CAAC,IAAI,OAAO,CAAC,SAAS;SACzE,CAAC;IACH,CAAC;CACD,CAAC,CAAC"}
|
package/dist/pii.js
CHANGED
|
@@ -48,7 +48,7 @@ export const pii = createPresetEval({
|
|
|
48
48
|
options: {
|
|
49
49
|
model: DEFAULT_EVAL_MODEL,
|
|
50
50
|
},
|
|
51
|
-
handler: async (
|
|
51
|
+
handler: async (_ctx, input, output, options) => {
|
|
52
52
|
const prompt = interpolatePrompt(piiPrompt, {
|
|
53
53
|
USER_REQUEST: input.request,
|
|
54
54
|
MODEL_RESPONSE: output.response,
|
package/dist/pii.js.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"pii.js","sourceRoot":"","sources":["../src/pii.ts"],"names":[],"mappings":"AAAA,OAAO,EACN,gBAAgB,EAChB,iBAAiB,EACjB,kBAAkB,GAGlB,MAAM,UAAU,CAAC;AAClB,OAAO,EAAE,kBAAkB,EAAwB,MAAM,SAAS,CAAC;AAEnE,MAAM,CAAC,MAAM,SAAS,GAAG;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;EAyCvB,CAAC;AAEH,MAAM,CAAC,MAAM,GAAG,GAAG,gBAAgB,CAAuD;IACzF,IAAI,EAAE,KAAK;IACX,WAAW,EAAE,uEAAuE;IACpF,OAAO,EAAE;QACR,KAAK,EAAE,kBAAkB;KACzB;IACD,OAAO,EAAE,KAAK,EAAE,
|
|
1
|
+
{"version":3,"file":"pii.js","sourceRoot":"","sources":["../src/pii.ts"],"names":[],"mappings":"AAAA,OAAO,EACN,gBAAgB,EAChB,iBAAiB,EACjB,kBAAkB,GAGlB,MAAM,UAAU,CAAC;AAClB,OAAO,EAAE,kBAAkB,EAAwB,MAAM,SAAS,CAAC;AAEnE,MAAM,CAAC,MAAM,SAAS,GAAG;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;EAyCvB,CAAC;AAEH,MAAM,CAAC,MAAM,GAAG,GAAG,gBAAgB,CAAuD;IACzF,IAAI,EAAE,KAAK;IACX,WAAW,EAAE,uEAAuE;IACpF,OAAO,EAAE;QACR,KAAK,EAAE,kBAAkB;KACzB;IACD,OAAO,EAAE,KAAK,EAAE,IAAI,EAAE,KAAK,EAAE,MAAM,EAAE,OAAO,EAAE,EAAE;QAC/C,MAAM,MAAM,GAAG,iBAAiB,CAAC,SAAS,EAAE;YAC3C,YAAY,EAAE,KAAK,CAAC,OAAO;YAC3B,cAAc,EAAE,MAAM,CAAC,QAAQ;SAC/B,CAAC,CAAC;QAEH,OAAO,kBAAkB,CAAC,EAAE,KAAK,EAAE,OAAO,CAAC,KAAK,EAAE,MAAM,EAAE,CAAC,CAAC;IAC7D,CAAC;CACD,CAAC,CAAC"}
|
package/dist/politeness.js
CHANGED
|
@@ -47,7 +47,7 @@ export const politeness = createPresetEval({
|
|
|
47
47
|
model: DEFAULT_EVAL_MODEL,
|
|
48
48
|
threshold: 0.8,
|
|
49
49
|
},
|
|
50
|
-
handler: async (
|
|
50
|
+
handler: async (_ctx, input, output, options) => {
|
|
51
51
|
const prompt = interpolatePrompt(politenessPrompt, {
|
|
52
52
|
USER_REQUEST: input.request,
|
|
53
53
|
MODEL_RESPONSE: output.response,
|
package/dist/politeness.js.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"politeness.js","sourceRoot":"","sources":["../src/politeness.ts"],"names":[],"mappings":"AAAA,OAAO,EACN,gBAAgB,EAChB,iBAAiB,EACjB,kBAAkB,GAGlB,MAAM,UAAU,CAAC;AAClB,OAAO,EAAE,kBAAkB,EAAwB,MAAM,SAAS,CAAC;AAEnE,MAAM,CAAC,MAAM,gBAAgB,GAAG;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;EAuC9B,CAAC;AAMH,MAAM,CAAC,MAAM,UAAU,GAAG,gBAAgB,CAIxC;IACD,IAAI,EAAE,YAAY;IAClB,WAAW,EAAE,4DAA4D;IACzE,OAAO,EAAE;QACR,KAAK,EAAE,kBAAkB;QACzB,SAAS,EAAE,GAAG;KACd;IACD,OAAO,EAAE,KAAK,EAAE,
|
|
1
|
+
{"version":3,"file":"politeness.js","sourceRoot":"","sources":["../src/politeness.ts"],"names":[],"mappings":"AAAA,OAAO,EACN,gBAAgB,EAChB,iBAAiB,EACjB,kBAAkB,GAGlB,MAAM,UAAU,CAAC;AAClB,OAAO,EAAE,kBAAkB,EAAwB,MAAM,SAAS,CAAC;AAEnE,MAAM,CAAC,MAAM,gBAAgB,GAAG;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;EAuC9B,CAAC;AAMH,MAAM,CAAC,MAAM,UAAU,GAAG,gBAAgB,CAIxC;IACD,IAAI,EAAE,YAAY;IAClB,WAAW,EAAE,4DAA4D;IACzE,OAAO,EAAE;QACR,KAAK,EAAE,kBAAkB;QACzB,SAAS,EAAE,GAAG;KACd;IACD,OAAO,EAAE,KAAK,EAAE,IAAI,EAAE,KAAK,EAAE,MAAM,EAAE,OAAO,EAAE,EAAE;QAC/C,MAAM,MAAM,GAAG,iBAAiB,CAAC,gBAAgB,EAAE;YAClD,YAAY,EAAE,KAAK,CAAC,OAAO;YAC3B,cAAc,EAAE,MAAM,CAAC,QAAQ;SAC/B,CAAC,CAAC;QAEH,MAAM,UAAU,GAAG,MAAM,kBAAkB,CAAC,EAAE,KAAK,EAAE,OAAO,CAAC,KAAK,EAAE,MAAM,EAAE,CAAC,CAAC;QAE9E,OAAO;YACN,GAAG,UAAU;YACb,MAAM,EAAE,UAAU,CAAC,MAAM,IAAI,CAAC,UAAU,CAAC,KAAK,IAAI,CAAC,CAAC,IAAI,OAAO,CAAC,SAAS;SACzE,CAAC;IACH,CAAC;CACD,CAAC,CAAC"}
|
package/dist/role-adherence.js
CHANGED
|
@@ -62,7 +62,7 @@ export const roleAdherence = createPresetEval({
|
|
|
62
62
|
model: DEFAULT_EVAL_MODEL,
|
|
63
63
|
threshold: 0.7,
|
|
64
64
|
},
|
|
65
|
-
handler: async (
|
|
65
|
+
handler: async (_ctx, input, output, options) => {
|
|
66
66
|
const prompt = interpolatePrompt(roleAdherencePrompt, {
|
|
67
67
|
USER_REQUEST: input.request,
|
|
68
68
|
MODEL_RESPONSE: output.response,
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"role-adherence.js","sourceRoot":"","sources":["../src/role-adherence.ts"],"names":[],"mappings":"AAAA,OAAO,EACN,gBAAgB,EAChB,iBAAiB,EACjB,kBAAkB,GAGlB,MAAM,UAAU,CAAC;AAClB,OAAO,EAAE,kBAAkB,EAAwB,MAAM,SAAS,CAAC;AAEnE,MAAM,CAAC,MAAM,mBAAmB,GAAG;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;EAsDjC,CAAC;AAMH,MAAM,CAAC,MAAM,aAAa,GAAG,gBAAgB,CAI3C;IACD,IAAI,EAAE,gBAAgB;IACtB,WAAW,EAAE,6EAA6E;IAC1F,OAAO,EAAE;QACR,KAAK,EAAE,kBAAkB;QACzB,SAAS,EAAE,GAAG;KACd;IACD,OAAO,EAAE,KAAK,EAAE,
|
|
1
|
+
{"version":3,"file":"role-adherence.js","sourceRoot":"","sources":["../src/role-adherence.ts"],"names":[],"mappings":"AAAA,OAAO,EACN,gBAAgB,EAChB,iBAAiB,EACjB,kBAAkB,GAGlB,MAAM,UAAU,CAAC;AAClB,OAAO,EAAE,kBAAkB,EAAwB,MAAM,SAAS,CAAC;AAEnE,MAAM,CAAC,MAAM,mBAAmB,GAAG;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;EAsDjC,CAAC;AAMH,MAAM,CAAC,MAAM,aAAa,GAAG,gBAAgB,CAI3C;IACD,IAAI,EAAE,gBAAgB;IACtB,WAAW,EAAE,6EAA6E;IAC1F,OAAO,EAAE;QACR,KAAK,EAAE,kBAAkB;QACzB,SAAS,EAAE,GAAG;KACd;IACD,OAAO,EAAE,KAAK,EAAE,IAAI,EAAE,KAAK,EAAE,MAAM,EAAE,OAAO,EAAE,EAAE;QAC/C,MAAM,MAAM,GAAG,iBAAiB,CAAC,mBAAmB,EAAE;YACrD,YAAY,EAAE,KAAK,CAAC,OAAO;YAC3B,cAAc,EAAE,MAAM,CAAC,QAAQ;SAC/B,CAAC,CAAC;QAEH,MAAM,UAAU,GAAG,MAAM,kBAAkB,CAAC,EAAE,KAAK,EAAE,OAAO,CAAC,KAAK,EAAE,MAAM,EAAE,CAAC,CAAC;QAE9E,OAAO;YACN,GAAG,UAAU;YACb,MAAM,EAAE,UAAU,CAAC,MAAM,IAAI,CAAC,UAAU,CAAC,KAAK,IAAI,CAAC,CAAC,IAAI,OAAO,CAAC,SAAS;SACzE,CAAC;IACH,CAAC;CACD,CAAC,CAAC"}
|
package/dist/safety.js
CHANGED
|
@@ -53,7 +53,7 @@ export const safety = createPresetEval({
|
|
|
53
53
|
options: {
|
|
54
54
|
model: DEFAULT_EVAL_MODEL,
|
|
55
55
|
},
|
|
56
|
-
handler: async (
|
|
56
|
+
handler: async (_ctx, input, output, options) => {
|
|
57
57
|
const prompt = interpolatePrompt(safetyPrompt, {
|
|
58
58
|
USER_REQUEST: input.request,
|
|
59
59
|
MODEL_RESPONSE: output.response,
|
package/dist/safety.js.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"safety.js","sourceRoot":"","sources":["../src/safety.ts"],"names":[],"mappings":"AAAA,OAAO,EACN,gBAAgB,EAChB,iBAAiB,EACjB,kBAAkB,GAGlB,MAAM,UAAU,CAAC;AAClB,OAAO,EAAE,kBAAkB,EAAwB,MAAM,SAAS,CAAC;AAEnE,MAAM,CAAC,MAAM,YAAY,GAAG;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;EA8C1B,CAAC;AAEH,MAAM,CAAC,MAAM,MAAM,GAAG,gBAAgB,CAAuD;IAC5F,IAAI,EAAE,QAAQ;IACd,WAAW,EAAE,wEAAwE;IACrF,OAAO,EAAE;QACR,KAAK,EAAE,kBAAkB;KACzB;IACD,OAAO,EAAE,KAAK,EAAE,
|
|
1
|
+
{"version":3,"file":"safety.js","sourceRoot":"","sources":["../src/safety.ts"],"names":[],"mappings":"AAAA,OAAO,EACN,gBAAgB,EAChB,iBAAiB,EACjB,kBAAkB,GAGlB,MAAM,UAAU,CAAC;AAClB,OAAO,EAAE,kBAAkB,EAAwB,MAAM,SAAS,CAAC;AAEnE,MAAM,CAAC,MAAM,YAAY,GAAG;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;EA8C1B,CAAC;AAEH,MAAM,CAAC,MAAM,MAAM,GAAG,gBAAgB,CAAuD;IAC5F,IAAI,EAAE,QAAQ;IACd,WAAW,EAAE,wEAAwE;IACrF,OAAO,EAAE;QACR,KAAK,EAAE,kBAAkB;KACzB;IACD,OAAO,EAAE,KAAK,EAAE,IAAI,EAAE,KAAK,EAAE,MAAM,EAAE,OAAO,EAAE,EAAE;QAC/C,MAAM,MAAM,GAAG,iBAAiB,CAAC,YAAY,EAAE;YAC9C,YAAY,EAAE,KAAK,CAAC,OAAO;YAC3B,cAAc,EAAE,MAAM,CAAC,QAAQ;SAC/B,CAAC,CAAC;QAEH,OAAO,kBAAkB,CAAC,EAAE,KAAK,EAAE,OAAO,CAAC,KAAK,EAAE,MAAM,EAAE,CAAC,CAAC;IAC7D,CAAC;CACD,CAAC,CAAC"}
|
package/dist/self-reference.js
CHANGED
|
@@ -43,7 +43,7 @@ export const selfReference = createPresetEval({
|
|
|
43
43
|
options: {
|
|
44
44
|
model: DEFAULT_EVAL_MODEL,
|
|
45
45
|
},
|
|
46
|
-
handler: async (
|
|
46
|
+
handler: async (_ctx, input, output, options) => {
|
|
47
47
|
const prompt = interpolatePrompt(selfReferencePrompt, {
|
|
48
48
|
USER_REQUEST: input.request,
|
|
49
49
|
MODEL_RESPONSE: output.response,
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"self-reference.js","sourceRoot":"","sources":["../src/self-reference.ts"],"names":[],"mappings":"AAAA,OAAO,EACN,gBAAgB,EAChB,iBAAiB,EACjB,kBAAkB,GAGlB,MAAM,UAAU,CAAC;AAClB,OAAO,EAAE,kBAAkB,EAAwB,MAAM,SAAS,CAAC;AAEnE,MAAM,CAAC,MAAM,mBAAmB,GAAG;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;EAoCjC,CAAC;AAEH,MAAM,CAAC,MAAM,aAAa,GAAG,gBAAgB,CAC5C;IACC,IAAI,EAAE,gBAAgB;IACtB,WAAW,EAAE,2EAA2E;IACxF,OAAO,EAAE;QACR,KAAK,EAAE,kBAAkB;KACzB;IACD,OAAO,EAAE,KAAK,EAAE,
|
|
1
|
+
{"version":3,"file":"self-reference.js","sourceRoot":"","sources":["../src/self-reference.ts"],"names":[],"mappings":"AAAA,OAAO,EACN,gBAAgB,EAChB,iBAAiB,EACjB,kBAAkB,GAGlB,MAAM,UAAU,CAAC;AAClB,OAAO,EAAE,kBAAkB,EAAwB,MAAM,SAAS,CAAC;AAEnE,MAAM,CAAC,MAAM,mBAAmB,GAAG;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;EAoCjC,CAAC;AAEH,MAAM,CAAC,MAAM,aAAa,GAAG,gBAAgB,CAC5C;IACC,IAAI,EAAE,gBAAgB;IACtB,WAAW,EAAE,2EAA2E;IACxF,OAAO,EAAE;QACR,KAAK,EAAE,kBAAkB;KACzB;IACD,OAAO,EAAE,KAAK,EAAE,IAAI,EAAE,KAAK,EAAE,MAAM,EAAE,OAAO,EAAE,EAAE;QAC/C,MAAM,MAAM,GAAG,iBAAiB,CAAC,mBAAmB,EAAE;YACrD,YAAY,EAAE,KAAK,CAAC,OAAO;YAC3B,cAAc,EAAE,MAAM,CAAC,QAAQ;SAC/B,CAAC,CAAC;QAEH,OAAO,kBAAkB,CAAC,EAAE,KAAK,EAAE,OAAO,CAAC,KAAK,EAAE,MAAM,EAAE,CAAC,CAAC;IAC7D,CAAC;CACD,CACD,CAAC"}
|
package/package.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "@agentuity/evals",
|
|
3
|
-
"version": "1.0.
|
|
3
|
+
"version": "1.0.32",
|
|
4
4
|
"license": "Apache-2.0",
|
|
5
5
|
"author": "Agentuity employees and contributors",
|
|
6
6
|
"type": "module",
|
|
@@ -26,14 +26,14 @@
|
|
|
26
26
|
"prepublishOnly": "bun run clean && bun run build"
|
|
27
27
|
},
|
|
28
28
|
"dependencies": {
|
|
29
|
-
"@agentuity/core": "1.0.
|
|
30
|
-
"@agentuity/runtime": "1.0.
|
|
31
|
-
"@agentuity/schema": "1.0.
|
|
29
|
+
"@agentuity/core": "1.0.32",
|
|
30
|
+
"@agentuity/runtime": "1.0.32",
|
|
31
|
+
"@agentuity/schema": "1.0.32",
|
|
32
32
|
"@ai-sdk/groq": "latest",
|
|
33
33
|
"ai": "^6.0.0"
|
|
34
34
|
},
|
|
35
35
|
"devDependencies": {
|
|
36
|
-
"@agentuity/test-utils": "1.0.
|
|
36
|
+
"@agentuity/test-utils": "1.0.32",
|
|
37
37
|
"@types/bun": "latest",
|
|
38
38
|
"bun-types": "latest",
|
|
39
39
|
"typescript": "^5.9.0"
|
package/src/adversarial.ts
CHANGED
|
@@ -61,7 +61,7 @@ export const adversarial = createPresetEval<DefaultEvalInput, DefaultEvalOutput,
|
|
|
61
61
|
options: {
|
|
62
62
|
model: DEFAULT_EVAL_MODEL,
|
|
63
63
|
},
|
|
64
|
-
handler: async (
|
|
64
|
+
handler: async (_ctx, input, output, options) => {
|
|
65
65
|
const prompt = interpolatePrompt(adversarialPrompt, {
|
|
66
66
|
USER_REQUEST: input.request,
|
|
67
67
|
MODEL_RESPONSE: output.response,
|
package/src/ambiguity.ts
CHANGED
|
@@ -72,7 +72,7 @@ export const ambiguity = createPresetEval<
|
|
|
72
72
|
model: DEFAULT_EVAL_MODEL,
|
|
73
73
|
threshold: 0.7,
|
|
74
74
|
},
|
|
75
|
-
handler: async (
|
|
75
|
+
handler: async (_ctx, input, output, options) => {
|
|
76
76
|
const prompt = interpolatePrompt(ambiguityPrompt, {
|
|
77
77
|
USER_REQUEST: input.request,
|
|
78
78
|
MODEL_RESPONSE: output.response,
|
|
@@ -73,7 +73,7 @@ export const answerCompleteness = createPresetEval<
|
|
|
73
73
|
model: DEFAULT_EVAL_MODEL,
|
|
74
74
|
threshold: 0.7,
|
|
75
75
|
},
|
|
76
|
-
handler: async (
|
|
76
|
+
handler: async (_ctx, input, output, options) => {
|
|
77
77
|
const prompt = interpolatePrompt(answerCompletenessPrompt, {
|
|
78
78
|
USER_REQUEST: input.request,
|
|
79
79
|
MODEL_RESPONSE: output.response,
|
package/src/conciseness.ts
CHANGED
|
@@ -72,7 +72,7 @@ export const conciseness = createPresetEval<
|
|
|
72
72
|
model: DEFAULT_EVAL_MODEL,
|
|
73
73
|
threshold: 0.7,
|
|
74
74
|
},
|
|
75
|
-
handler: async (
|
|
75
|
+
handler: async (_ctx, input, output, options) => {
|
|
76
76
|
const prompt = interpolatePrompt(concisenessPrompt, {
|
|
77
77
|
USER_REQUEST: input.request,
|
|
78
78
|
MODEL_RESPONSE: output.response,
|
|
@@ -74,7 +74,7 @@ export const extraneousContent = createPresetEval<
|
|
|
74
74
|
model: DEFAULT_EVAL_MODEL,
|
|
75
75
|
threshold: 0.7,
|
|
76
76
|
},
|
|
77
|
-
handler: async (
|
|
77
|
+
handler: async (_ctx, input, output, options) => {
|
|
78
78
|
const prompt = interpolatePrompt(extraneousContentPrompt, {
|
|
79
79
|
USER_REQUEST: input.request,
|
|
80
80
|
MODEL_RESPONSE: output.response,
|
package/src/format.ts
CHANGED
|
@@ -59,7 +59,7 @@ export const format = createPresetEval<DefaultEvalInput, DefaultEvalOutput, Base
|
|
|
59
59
|
options: {
|
|
60
60
|
model: DEFAULT_EVAL_MODEL,
|
|
61
61
|
},
|
|
62
|
-
handler: async (
|
|
62
|
+
handler: async (_ctx, input, output, options) => {
|
|
63
63
|
const prompt = interpolatePrompt(formatPrompt, {
|
|
64
64
|
USER_REQUEST: input.request,
|
|
65
65
|
MODEL_RESPONSE: output.response,
|
|
@@ -77,7 +77,7 @@ export const knowledgeRetention = createPresetEval<
|
|
|
77
77
|
model: DEFAULT_EVAL_MODEL,
|
|
78
78
|
threshold: 0.7,
|
|
79
79
|
},
|
|
80
|
-
handler: async (
|
|
80
|
+
handler: async (_ctx, input, output, options) => {
|
|
81
81
|
const prompt = interpolatePrompt(knowledgeRetentionPrompt, {
|
|
82
82
|
USER_REQUEST: input.request,
|
|
83
83
|
MODEL_RESPONSE: output.response,
|
package/src/pii.ts
CHANGED
|
@@ -56,7 +56,7 @@ export const pii = createPresetEval<DefaultEvalInput, DefaultEvalOutput, BaseEva
|
|
|
56
56
|
options: {
|
|
57
57
|
model: DEFAULT_EVAL_MODEL,
|
|
58
58
|
},
|
|
59
|
-
handler: async (
|
|
59
|
+
handler: async (_ctx, input, output, options) => {
|
|
60
60
|
const prompt = interpolatePrompt(piiPrompt, {
|
|
61
61
|
USER_REQUEST: input.request,
|
|
62
62
|
MODEL_RESPONSE: output.response,
|
package/src/politeness.ts
CHANGED
|
@@ -63,7 +63,7 @@ export const politeness = createPresetEval<
|
|
|
63
63
|
model: DEFAULT_EVAL_MODEL,
|
|
64
64
|
threshold: 0.8,
|
|
65
65
|
},
|
|
66
|
-
handler: async (
|
|
66
|
+
handler: async (_ctx, input, output, options) => {
|
|
67
67
|
const prompt = interpolatePrompt(politenessPrompt, {
|
|
68
68
|
USER_REQUEST: input.request,
|
|
69
69
|
MODEL_RESPONSE: output.response,
|
package/src/role-adherence.ts
CHANGED
|
@@ -78,7 +78,7 @@ export const roleAdherence = createPresetEval<
|
|
|
78
78
|
model: DEFAULT_EVAL_MODEL,
|
|
79
79
|
threshold: 0.7,
|
|
80
80
|
},
|
|
81
|
-
handler: async (
|
|
81
|
+
handler: async (_ctx, input, output, options) => {
|
|
82
82
|
const prompt = interpolatePrompt(roleAdherencePrompt, {
|
|
83
83
|
USER_REQUEST: input.request,
|
|
84
84
|
MODEL_RESPONSE: output.response,
|
package/src/safety.ts
CHANGED
|
@@ -61,7 +61,7 @@ export const safety = createPresetEval<DefaultEvalInput, DefaultEvalOutput, Base
|
|
|
61
61
|
options: {
|
|
62
62
|
model: DEFAULT_EVAL_MODEL,
|
|
63
63
|
},
|
|
64
|
-
handler: async (
|
|
64
|
+
handler: async (_ctx, input, output, options) => {
|
|
65
65
|
const prompt = interpolatePrompt(safetyPrompt, {
|
|
66
66
|
USER_REQUEST: input.request,
|
|
67
67
|
MODEL_RESPONSE: output.response,
|
package/src/self-reference.ts
CHANGED
|
@@ -52,7 +52,7 @@ export const selfReference = createPresetEval<DefaultEvalInput, DefaultEvalOutpu
|
|
|
52
52
|
options: {
|
|
53
53
|
model: DEFAULT_EVAL_MODEL,
|
|
54
54
|
},
|
|
55
|
-
handler: async (
|
|
55
|
+
handler: async (_ctx, input, output, options) => {
|
|
56
56
|
const prompt = interpolatePrompt(selfReferencePrompt, {
|
|
57
57
|
USER_REQUEST: input.request,
|
|
58
58
|
MODEL_RESPONSE: output.response,
|