@dalmia/calibrate-mcp 0.0.15 → 0.0.16
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/bin/mcp-server.js +11 -14
- package/bin/mcp-server.js.map +10 -10
- package/esm/landing-page.js +1 -1
- package/esm/lib/config.d.ts +2 -2
- package/esm/lib/config.js +2 -2
- package/esm/mcp-server/mcp-server.js +1 -1
- package/esm/mcp-server/server.js +1 -1
- package/esm/models/annotationitempayload.js +1 -1
- package/esm/models/annotationitempayload.js.map +1 -1
- package/esm/models/annotationtaskcreate.d.ts +0 -4
- package/esm/models/annotationtaskcreate.d.ts.map +1 -1
- package/esm/models/annotationtaskcreate.js +2 -5
- package/esm/models/annotationtaskcreate.js.map +1 -1
- package/esm/models/annotationtaskresponse.d.ts +0 -4
- package/esm/models/annotationtaskresponse.d.ts.map +1 -1
- package/esm/models/annotationtaskresponse.js +2 -5
- package/esm/models/annotationtaskresponse.js.map +1 -1
- package/esm/models/evaluatorcreate.d.ts +1 -1
- package/esm/models/evaluatordetailresponse.d.ts +1 -1
- package/esm/models/listevaluatorsevaluatorsgetop.d.ts +1 -1
- package/esm/models/routersevaluatorsevaluatorresponse.d.ts +1 -1
- package/esm/models/tasksummaryresponse.d.ts +0 -2
- package/esm/models/tasksummaryresponse.d.ts.map +1 -1
- package/esm/models/tasksummaryresponse.js +0 -2
- package/esm/models/tasksummaryresponse.js.map +1 -1
- package/package.json +1 -1
- package/src/landing-page.ts +1 -1
- package/src/lib/config.ts +2 -2
- package/src/mcp-server/mcp-server.ts +1 -1
- package/src/mcp-server/server.ts +1 -1
- package/src/models/annotationitempayload.ts +1 -1
- package/src/models/annotationtaskcreate.ts +2 -6
- package/src/models/annotationtaskresponse.ts +2 -6
- package/src/models/tasksummaryresponse.ts +0 -2
package/esm/landing-page.js
CHANGED
|
@@ -915,7 +915,7 @@ http_headers = { "api-key-auth" = "YOUR_API_KEY_AUTH" }`;
|
|
|
915
915
|
<h1>Instructions</h1>
|
|
916
916
|
<p>One-click installation for Claude Desktop users</p>
|
|
917
917
|
<div class="instruction-item">
|
|
918
|
-
<a href="https://github.com/dalmia/calibrate-mcp/releases/download/v0.0.
|
|
918
|
+
<a href="https://github.com/dalmia/calibrate-mcp/releases/download/v0.0.16/mcp-server.mcpb" download="mcp-server.mcpb" class="action-button header-action" style="display: inline-flex; margin-bottom: 16px;">
|
|
919
919
|
📥 Download MCP Bundle
|
|
920
920
|
</a>
|
|
921
921
|
</div>
|
package/esm/lib/config.d.ts
CHANGED
|
@@ -35,8 +35,8 @@ export declare function serverURLFromOptions(options: SDKOptions): URL | null;
|
|
|
35
35
|
export declare const SDK_METADATA: {
|
|
36
36
|
readonly language: "typescript";
|
|
37
37
|
readonly openapiDocVersion: "0.1.0";
|
|
38
|
-
readonly sdkVersion: "0.0.
|
|
38
|
+
readonly sdkVersion: "0.0.16";
|
|
39
39
|
readonly genVersion: "2.915.1";
|
|
40
|
-
readonly userAgent: "speakeasy-sdk/mcp-typescript 0.0.
|
|
40
|
+
readonly userAgent: "speakeasy-sdk/mcp-typescript 0.0.16 2.915.1 0.1.0 @dalmia/calibrate-mcp";
|
|
41
41
|
};
|
|
42
42
|
//# sourceMappingURL=config.d.ts.map
|
package/esm/lib/config.js
CHANGED
|
@@ -27,8 +27,8 @@ export function serverURLFromOptions(options) {
|
|
|
27
27
|
export const SDK_METADATA = {
|
|
28
28
|
language: "typescript",
|
|
29
29
|
openapiDocVersion: "0.1.0",
|
|
30
|
-
sdkVersion: "0.0.
|
|
30
|
+
sdkVersion: "0.0.16",
|
|
31
31
|
genVersion: "2.915.1",
|
|
32
|
-
userAgent: "speakeasy-sdk/mcp-typescript 0.0.
|
|
32
|
+
userAgent: "speakeasy-sdk/mcp-typescript 0.0.16 2.915.1 0.1.0 @dalmia/calibrate-mcp",
|
|
33
33
|
};
|
|
34
34
|
//# sourceMappingURL=config.js.map
|
package/esm/mcp-server/server.js
CHANGED
|
@@ -42,7 +42,7 @@ import { tool$testsUpdate } from "./tools/testsUpdate.js";
|
|
|
42
42
|
export function createMCPServer(deps) {
|
|
43
43
|
const server = new McpServer({
|
|
44
44
|
name: "CalibrateMcp",
|
|
45
|
-
version: "0.0.
|
|
45
|
+
version: "0.0.16",
|
|
46
46
|
});
|
|
47
47
|
const getClient = deps.getSDK || (() => new CalibrateMcpCore({
|
|
48
48
|
security: deps.security,
|
|
@@ -4,6 +4,6 @@
|
|
|
4
4
|
import * as z from "zod";
|
|
5
5
|
export const AnnotationItemPayload$zodSchema = z.object({
|
|
6
6
|
annotations: z.record(z.string(), z.any()).nullable().optional().describe("Human annotations to seed, keyed by evaluator ID. Each evaluator ID must be linked to the task. Put the judgement in `value`, a bool for binary or a number for rating, with optional `reasoning`. Requires `annotator_id`"),
|
|
7
|
-
payload: z.any().optional().describe("The item's payload.
|
|
7
|
+
payload: z.any().optional().describe("The item's payload. `name` is required and unique within the task for every type. The other fields depend on the task `type`:\n\n- `stt`: `name`, `reference_transcript`, `predicted_transcript`\n- `llm`: `name`, `chat_history` (list of `{role, content}` turns ending at the user turn), `agent_response` (the reply to judge), optional `evaluator_variables`\n- `llm-general`: `name`, `input`, `output`, optional `evaluator_variables`\n- `conversation`: `name`, `transcript` (list of `{role, content}` turns), optional `evaluator_variables`\n\n`evaluator_variables` maps an evaluator ID to that evaluator's `{{\"{{\"}}variable}}` values for this item"),
|
|
8
8
|
});
|
|
9
9
|
//# sourceMappingURL=annotationitempayload.js.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"annotationitempayload.js","sourceRoot":"","sources":["../../src/models/annotationitempayload.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AAOzB,MAAM,CAAC,MAAM,+BAA+B,GAC1C,CAAC,CAAC,MAAM,CAAC;IACP,WAAW,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CACvE,4NAA4N,CAC7N;IACD,OAAO,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CAClC
|
|
1
|
+
{"version":3,"file":"annotationitempayload.js","sourceRoot":"","sources":["../../src/models/annotationitempayload.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AAOzB,MAAM,CAAC,MAAM,+BAA+B,GAC1C,CAAC,CAAC,MAAM,CAAC;IACP,WAAW,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CACvE,4NAA4N,CAC7N;IACD,OAAO,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CAClC,woBAAwoB,CACzoB;CACF,CAAC,CAAC"}
|
|
@@ -5,14 +5,12 @@ import { ClosedEnum } from "../types/enums.js";
|
|
|
5
5
|
*
|
|
6
6
|
* @remarks
|
|
7
7
|
* - `stt`: judge a transcript on its own
|
|
8
|
-
* - `tts`: judge synthesized speech output
|
|
9
8
|
* - `llm`: judge one response with its conversation history
|
|
10
9
|
* - `llm-general`: judge a standalone `input -> output` pair
|
|
11
10
|
* - `conversation`: judge a full conversation
|
|
12
11
|
*/
|
|
13
12
|
export declare const AnnotationTaskCreateType: {
|
|
14
13
|
readonly Stt: "stt";
|
|
15
|
-
readonly Tts: "tts";
|
|
16
14
|
readonly Llm: "llm";
|
|
17
15
|
readonly LlmGeneral: "llm-general";
|
|
18
16
|
readonly Conversation: "conversation";
|
|
@@ -22,7 +20,6 @@ export declare const AnnotationTaskCreateType: {
|
|
|
22
20
|
*
|
|
23
21
|
* @remarks
|
|
24
22
|
* - `stt`: judge a transcript on its own
|
|
25
|
-
* - `tts`: judge synthesized speech output
|
|
26
23
|
* - `llm`: judge one response with its conversation history
|
|
27
24
|
* - `llm-general`: judge a standalone `input -> output` pair
|
|
28
25
|
* - `conversation`: judge a full conversation
|
|
@@ -31,7 +28,6 @@ export type AnnotationTaskCreateType = ClosedEnum<typeof AnnotationTaskCreateTyp
|
|
|
31
28
|
export declare const AnnotationTaskCreateType$zodSchema: z.ZodEnum<{
|
|
32
29
|
conversation: "conversation";
|
|
33
30
|
stt: "stt";
|
|
34
|
-
tts: "tts";
|
|
35
31
|
llm: "llm";
|
|
36
32
|
"llm-general": "llm-general";
|
|
37
33
|
}>;
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"annotationtaskcreate.d.ts","sourceRoot":"","sources":["../../src/models/annotationtaskcreate.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EAAE,UAAU,EAAE,MAAM,mBAAmB,CAAC;AAE/C
|
|
1
|
+
{"version":3,"file":"annotationtaskcreate.d.ts","sourceRoot":"","sources":["../../src/models/annotationtaskcreate.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EAAE,UAAU,EAAE,MAAM,mBAAmB,CAAC;AAE/C;;;;;;;;GAQG;AACH,eAAO,MAAM,wBAAwB;;;;;CAK3B,CAAC;AACX;;;;;;;;GAQG;AACH,MAAM,MAAM,wBAAwB,GAAG,UAAU,CAC/C,OAAO,wBAAwB,CAChC,CAAC;AAEF,eAAO,MAAM,kCAAkC;;;;;EAO9C,CAAC;AAEF,MAAM,MAAM,oBAAoB,GAAG;IACjC,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE,wBAAwB,CAAC;IAC/B,WAAW,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;IACxC,aAAa,CAAC,EAAE,KAAK,CAAC,MAAM,CAAC,GAAG,IAAI,GAAG,SAAS,CAAC;CAClD,CAAC;AAEF,eAAO,MAAM,8BAA8B,EAAE,CAAC,CAAC,OAAO,CAAC,oBAAoB,CAYvE,CAAC"}
|
|
@@ -7,30 +7,27 @@ import * as z from "zod";
|
|
|
7
7
|
*
|
|
8
8
|
* @remarks
|
|
9
9
|
* - `stt`: judge a transcript on its own
|
|
10
|
-
* - `tts`: judge synthesized speech output
|
|
11
10
|
* - `llm`: judge one response with its conversation history
|
|
12
11
|
* - `llm-general`: judge a standalone `input -> output` pair
|
|
13
12
|
* - `conversation`: judge a full conversation
|
|
14
13
|
*/
|
|
15
14
|
export const AnnotationTaskCreateType = {
|
|
16
15
|
Stt: "stt",
|
|
17
|
-
Tts: "tts",
|
|
18
16
|
Llm: "llm",
|
|
19
17
|
LlmGeneral: "llm-general",
|
|
20
18
|
Conversation: "conversation",
|
|
21
19
|
};
|
|
22
20
|
export const AnnotationTaskCreateType$zodSchema = z.enum([
|
|
23
21
|
"stt",
|
|
24
|
-
"tts",
|
|
25
22
|
"llm",
|
|
26
23
|
"llm-general",
|
|
27
24
|
"conversation",
|
|
28
|
-
]).describe("Task type. Determines the shape of each item's payload.\n- `stt`: judge a transcript on its own\n- `
|
|
25
|
+
]).describe("Task type. Determines the shape of each item's payload.\n- `stt`: judge a transcript on its own\n- `llm`: judge one response with its conversation history\n- `llm-general`: judge a standalone `input -> output` pair\n- `conversation`: judge a full conversation");
|
|
29
26
|
export const AnnotationTaskCreate$zodSchema = z
|
|
30
27
|
.object({
|
|
31
28
|
description: z.string().nullable().optional().describe("A description for the task. Omit for none"),
|
|
32
29
|
evaluator_ids: z.array(z.string()).nullable().optional().describe("IDs of evaluators to link when the task is created, in order. Each must be one you created or a built-in default. Omit to create with no linked evaluators"),
|
|
33
30
|
name: z.string().describe("Task name, unique within your workspace"),
|
|
34
|
-
type: AnnotationTaskCreateType$zodSchema.describe("Task type. Determines the shape of each item's payload.\n- `stt`: judge a transcript on its own\n- `
|
|
31
|
+
type: AnnotationTaskCreateType$zodSchema.describe("Task type. Determines the shape of each item's payload.\n- `stt`: judge a transcript on its own\n- `llm`: judge one response with its conversation history\n- `llm-general`: judge a standalone `input -> output` pair\n- `conversation`: judge a full conversation"),
|
|
35
32
|
});
|
|
36
33
|
//# sourceMappingURL=annotationtaskcreate.js.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"annotationtaskcreate.js","sourceRoot":"","sources":["../../src/models/annotationtaskcreate.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AAGzB
|
|
1
|
+
{"version":3,"file":"annotationtaskcreate.js","sourceRoot":"","sources":["../../src/models/annotationtaskcreate.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AAGzB;;;;;;;;GAQG;AACH,MAAM,CAAC,MAAM,wBAAwB,GAAG;IACtC,GAAG,EAAE,KAAK;IACV,GAAG,EAAE,KAAK;IACV,UAAU,EAAE,aAAa;IACzB,YAAY,EAAE,cAAc;CACpB,CAAC;AAcX,MAAM,CAAC,MAAM,kCAAkC,GAAG,CAAC,CAAC,IAAI,CAAC;IACvD,KAAK;IACL,KAAK;IACL,aAAa;IACb,cAAc;CACf,CAAC,CAAC,QAAQ,CACT,qQAAqQ,CACtQ,CAAC;AASF,MAAM,CAAC,MAAM,8BAA8B,GAAoC,CAAC;KAC7E,MAAM,CAAC;IACN,WAAW,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CACpD,2CAA2C,CAC5C;IACD,aAAa,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CAC/D,4JAA4J,CAC7J;IACD,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,yCAAyC,CAAC;IACpE,IAAI,EAAE,kCAAkC,CAAC,QAAQ,CAC/C,qQAAqQ,CACtQ;CACF,CAAC,CAAC"}
|
|
@@ -5,14 +5,12 @@ import { ClosedEnum } from "../types/enums.js";
|
|
|
5
5
|
*
|
|
6
6
|
* @remarks
|
|
7
7
|
* - `stt`: judge a transcript on its own
|
|
8
|
-
* - `tts`: judge synthesized speech output
|
|
9
8
|
* - `llm`: judge one response with its conversation history
|
|
10
9
|
* - `llm-general`: judge a standalone `input -> output` pair
|
|
11
10
|
* - `conversation`: judge a full conversation
|
|
12
11
|
*/
|
|
13
12
|
export declare const AnnotationTaskResponseType: {
|
|
14
13
|
readonly Stt: "stt";
|
|
15
|
-
readonly Tts: "tts";
|
|
16
14
|
readonly Llm: "llm";
|
|
17
15
|
readonly LlmGeneral: "llm-general";
|
|
18
16
|
readonly Conversation: "conversation";
|
|
@@ -22,7 +20,6 @@ export declare const AnnotationTaskResponseType: {
|
|
|
22
20
|
*
|
|
23
21
|
* @remarks
|
|
24
22
|
* - `stt`: judge a transcript on its own
|
|
25
|
-
* - `tts`: judge synthesized speech output
|
|
26
23
|
* - `llm`: judge one response with its conversation history
|
|
27
24
|
* - `llm-general`: judge a standalone `input -> output` pair
|
|
28
25
|
* - `conversation`: judge a full conversation
|
|
@@ -31,7 +28,6 @@ export type AnnotationTaskResponseType = ClosedEnum<typeof AnnotationTaskRespons
|
|
|
31
28
|
export declare const AnnotationTaskResponseType$zodSchema: z.ZodEnum<{
|
|
32
29
|
conversation: "conversation";
|
|
33
30
|
stt: "stt";
|
|
34
|
-
tts: "tts";
|
|
35
31
|
llm: "llm";
|
|
36
32
|
"llm-general": "llm-general";
|
|
37
33
|
}>;
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"annotationtaskresponse.d.ts","sourceRoot":"","sources":["../../src/models/annotationtaskresponse.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EAAE,UAAU,EAAE,MAAM,mBAAmB,CAAC;AAE/C
|
|
1
|
+
{"version":3,"file":"annotationtaskresponse.d.ts","sourceRoot":"","sources":["../../src/models/annotationtaskresponse.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EAAE,UAAU,EAAE,MAAM,mBAAmB,CAAC;AAE/C;;;;;;;;GAQG;AACH,eAAO,MAAM,0BAA0B;;;;;CAK7B,CAAC;AACX;;;;;;;;GAQG;AACH,MAAM,MAAM,0BAA0B,GAAG,UAAU,CACjD,OAAO,0BAA0B,CAClC,CAAC;AAEF,eAAO,MAAM,oCAAoC;;;;;EAOhD,CAAC;AAEF,MAAM,MAAM,sBAAsB,GAAG;IACnC,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE,0BAA0B,CAAC;IACjC,WAAW,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;IACxC,UAAU,EAAE,MAAM,CAAC;IACnB,UAAU,EAAE,MAAM,CAAC;IACnB,UAAU,CAAC,EAAE,KAAK,CAAC;QAAE,CAAC,CAAC,EAAE,MAAM,GAAG,GAAG,CAAA;KAAE,CAAC,GAAG,SAAS,CAAC;IACrD,UAAU,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;IACvC,KAAK,CAAC,EAAE,KAAK,CAAC;QAAE,CAAC,CAAC,EAAE,MAAM,GAAG,GAAG,CAAA;KAAE,CAAC,GAAG,SAAS,CAAC;IAChD,IAAI,CAAC,EAAE,KAAK,CAAC;QAAE,CAAC,CAAC,EAAE,MAAM,GAAG,GAAG,CAAA;KAAE,CAAC,GAAG,SAAS,CAAC;CAChD,CAAC;AAEF,eAAO,MAAM,gCAAgC,EAAE,CAAC,CAAC,OAAO,CACtD,sBAAsB,CA0BtB,CAAC"}
|
|
@@ -7,25 +7,22 @@ import * as z from "zod";
|
|
|
7
7
|
*
|
|
8
8
|
* @remarks
|
|
9
9
|
* - `stt`: judge a transcript on its own
|
|
10
|
-
* - `tts`: judge synthesized speech output
|
|
11
10
|
* - `llm`: judge one response with its conversation history
|
|
12
11
|
* - `llm-general`: judge a standalone `input -> output` pair
|
|
13
12
|
* - `conversation`: judge a full conversation
|
|
14
13
|
*/
|
|
15
14
|
export const AnnotationTaskResponseType = {
|
|
16
15
|
Stt: "stt",
|
|
17
|
-
Tts: "tts",
|
|
18
16
|
Llm: "llm",
|
|
19
17
|
LlmGeneral: "llm-general",
|
|
20
18
|
Conversation: "conversation",
|
|
21
19
|
};
|
|
22
20
|
export const AnnotationTaskResponseType$zodSchema = z.enum([
|
|
23
21
|
"stt",
|
|
24
|
-
"tts",
|
|
25
22
|
"llm",
|
|
26
23
|
"llm-general",
|
|
27
24
|
"conversation",
|
|
28
|
-
]).describe("Task type. Determines the shape of each item's payload.\n- `stt`: judge a transcript on its own\n- `
|
|
25
|
+
]).describe("Task type. Determines the shape of each item's payload.\n- `stt`: judge a transcript on its own\n- `llm`: judge one response with its conversation history\n- `llm-general`: judge a standalone `input -> output` pair\n- `conversation`: judge a full conversation");
|
|
29
26
|
export const AnnotationTaskResponse$zodSchema = z.object({
|
|
30
27
|
created_at: z.string().describe("When the task was created (ISO 8601 UTC)"),
|
|
31
28
|
description: z.string().nullable().optional().describe("The task's description"),
|
|
@@ -34,7 +31,7 @@ export const AnnotationTaskResponse$zodSchema = z.object({
|
|
|
34
31
|
items: z.array(z.record(z.string(), z.any())).optional().describe("The task's items, each with its agreement stats. You get these when you fetch one task by ID, not when you list tasks"),
|
|
35
32
|
jobs: z.array(z.record(z.string(), z.any())).optional().describe("The task's labelling jobs. You get these when you fetch one task by ID, not when you list tasks"),
|
|
36
33
|
name: z.string().describe("Name of the task"),
|
|
37
|
-
type: AnnotationTaskResponseType$zodSchema.describe("Task type. Determines the shape of each item's payload.\n- `stt`: judge a transcript on its own\n- `
|
|
34
|
+
type: AnnotationTaskResponseType$zodSchema.describe("Task type. Determines the shape of each item's payload.\n- `stt`: judge a transcript on its own\n- `llm`: judge one response with its conversation history\n- `llm-general`: judge a standalone `input -> output` pair\n- `conversation`: judge a full conversation"),
|
|
38
35
|
updated_at: z.string().describe("When the task was last updated (ISO 8601 UTC)"),
|
|
39
36
|
uuid: z.string().describe("Unique ID for the task"),
|
|
40
37
|
});
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"annotationtaskresponse.js","sourceRoot":"","sources":["../../src/models/annotationtaskresponse.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AAGzB
|
|
1
|
+
{"version":3,"file":"annotationtaskresponse.js","sourceRoot":"","sources":["../../src/models/annotationtaskresponse.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AAGzB;;;;;;;;GAQG;AACH,MAAM,CAAC,MAAM,0BAA0B,GAAG;IACxC,GAAG,EAAE,KAAK;IACV,GAAG,EAAE,KAAK;IACV,UAAU,EAAE,aAAa;IACzB,YAAY,EAAE,cAAc;CACpB,CAAC;AAcX,MAAM,CAAC,MAAM,oCAAoC,GAAG,CAAC,CAAC,IAAI,CAAC;IACzD,KAAK;IACL,KAAK;IACL,aAAa;IACb,cAAc;CACf,CAAC,CAAC,QAAQ,CACT,qQAAqQ,CACtQ,CAAC;AAeF,MAAM,CAAC,MAAM,gCAAgC,GAEzC,CAAC,CAAC,MAAM,CAAC;IACX,UAAU,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,0CAA0C,CAAC;IAC3E,WAAW,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CACpD,wBAAwB,CACzB;IACD,UAAU,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC,CAAC,QAAQ,EAAE,CAAC,QAAQ,CACpE,2FAA2F,CAC5F;IACD,UAAU,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CAChD,6BAA6B,CAC9B;IACD,KAAK,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC,CAAC,QAAQ,EAAE,CAAC,QAAQ,CAC/D,uHAAuH,CACxH;IACD,IAAI,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC,CAAC,QAAQ,EAAE,CAAC,QAAQ,CAC9D,iGAAiG,CAClG;IACD,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,kBAAkB,CAAC;IAC7C,IAAI,EAAE,oCAAoC,CAAC,QAAQ,CACjD,qQAAqQ,CACtQ;IACD,UAAU,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAC7B,+CAA+C,CAChD;IACD,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,wBAAwB,CAAC;CACpD,CAAC,CAAC"}
|
|
@@ -34,9 +34,9 @@ export type EvaluatorCreateEvaluatorType = ClosedEnum<typeof EvaluatorCreateEval
|
|
|
34
34
|
export declare const EvaluatorCreateEvaluatorType$zodSchema: z.ZodEnum<{
|
|
35
35
|
conversation: "conversation";
|
|
36
36
|
stt: "stt";
|
|
37
|
-
tts: "tts";
|
|
38
37
|
llm: "llm";
|
|
39
38
|
"llm-general": "llm-general";
|
|
39
|
+
tts: "tts";
|
|
40
40
|
}>;
|
|
41
41
|
/**
|
|
42
42
|
* The modality the judge reads:
|
|
@@ -34,9 +34,9 @@ export type EvaluatorDetailResponseEvaluatorType = ClosedEnum<typeof EvaluatorDe
|
|
|
34
34
|
export declare const EvaluatorDetailResponseEvaluatorType$zodSchema: z.ZodEnum<{
|
|
35
35
|
conversation: "conversation";
|
|
36
36
|
stt: "stt";
|
|
37
|
-
tts: "tts";
|
|
38
37
|
llm: "llm";
|
|
39
38
|
"llm-general": "llm-general";
|
|
39
|
+
tts: "tts";
|
|
40
40
|
}>;
|
|
41
41
|
/**
|
|
42
42
|
* The modality the judge reads:
|
|
@@ -19,9 +19,9 @@ export type ListEvaluatorsEvaluatorsGetEvaluatorType = ClosedEnum<typeof ListEva
|
|
|
19
19
|
export declare const ListEvaluatorsEvaluatorsGetEvaluatorType$zodSchema: z.ZodEnum<{
|
|
20
20
|
conversation: "conversation";
|
|
21
21
|
stt: "stt";
|
|
22
|
-
tts: "tts";
|
|
23
22
|
llm: "llm";
|
|
24
23
|
"llm-general": "llm-general";
|
|
24
|
+
tts: "tts";
|
|
25
25
|
}>;
|
|
26
26
|
/**
|
|
27
27
|
* Filter by modality. Omit for all
|
|
@@ -34,9 +34,9 @@ export type RoutersEvaluatorsEvaluatorResponseEvaluatorType = ClosedEnum<typeof
|
|
|
34
34
|
export declare const RoutersEvaluatorsEvaluatorResponseEvaluatorType$zodSchema: z.ZodEnum<{
|
|
35
35
|
conversation: "conversation";
|
|
36
36
|
stt: "stt";
|
|
37
|
-
tts: "tts";
|
|
38
37
|
llm: "llm";
|
|
39
38
|
"llm-general": "llm-general";
|
|
39
|
+
tts: "tts";
|
|
40
40
|
}>;
|
|
41
41
|
/**
|
|
42
42
|
* The modality the judge reads:
|
|
@@ -6,7 +6,6 @@ import { PaginationMeta } from "./paginationmeta.js";
|
|
|
6
6
|
*/
|
|
7
7
|
export declare const TaskType: {
|
|
8
8
|
readonly Stt: "stt";
|
|
9
|
-
readonly Tts: "tts";
|
|
10
9
|
readonly Llm: "llm";
|
|
11
10
|
readonly LlmGeneral: "llm-general";
|
|
12
11
|
readonly Conversation: "conversation";
|
|
@@ -18,7 +17,6 @@ export type TaskType = ClosedEnum<typeof TaskType>;
|
|
|
18
17
|
export declare const TaskType$zodSchema: z.ZodEnum<{
|
|
19
18
|
conversation: "conversation";
|
|
20
19
|
stt: "stt";
|
|
21
|
-
tts: "tts";
|
|
22
20
|
llm: "llm";
|
|
23
21
|
"llm-general": "llm-general";
|
|
24
22
|
}>;
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"tasksummaryresponse.d.ts","sourceRoot":"","sources":["../../src/models/tasksummaryresponse.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AAEzB,OAAO,EAAE,UAAU,EAAE,MAAM,mBAAmB,CAAC;AAC/C,OAAO,EAAE,cAAc,EAA4B,MAAM,qBAAqB,CAAC;AAE/E;;GAEG;AACH,eAAO,MAAM,QAAQ
|
|
1
|
+
{"version":3,"file":"tasksummaryresponse.d.ts","sourceRoot":"","sources":["../../src/models/tasksummaryresponse.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AAEzB,OAAO,EAAE,UAAU,EAAE,MAAM,mBAAmB,CAAC;AAC/C,OAAO,EAAE,cAAc,EAA4B,MAAM,qBAAqB,CAAC;AAE/E;;GAEG;AACH,eAAO,MAAM,QAAQ;;;;;CAKX,CAAC;AACX;;GAEG;AACH,MAAM,MAAM,QAAQ,GAAG,UAAU,CAAC,OAAO,QAAQ,CAAC,CAAC;AAEnD,eAAO,MAAM,kBAAkB;;;;;EAKA,CAAC;AAEhC,MAAM,MAAM,mBAAmB,GAAG;IAChC,OAAO,EAAE,MAAM,CAAC;IAChB,SAAS,EAAE,QAAQ,CAAC;IACpB,UAAU,EAAE,KAAK,CAAC;QAAE,CAAC,CAAC,EAAE,MAAM,GAAG,GAAG,CAAA;KAAE,CAAC,CAAC;IACxC,UAAU,EAAE,KAAK,CAAC;QAAE,CAAC,CAAC,EAAE,MAAM,GAAG,GAAG,CAAA;KAAE,CAAC,CAAC;IACxC,IAAI,EAAE,KAAK,CAAC;QAAE,CAAC,CAAC,EAAE,MAAM,GAAG,GAAG,CAAA;KAAE,CAAC,CAAC;IAClC,aAAa,EAAE;QAAE,CAAC,CAAC,EAAE,MAAM,GAAG,GAAG,CAAA;KAAE,CAAC;IACpC,UAAU,EAAE,cAAc,CAAC;IAC3B,oBAAoB,CAAC,EAAE;QAAE,CAAC,CAAC,EAAE,MAAM,GAAG,GAAG,CAAA;KAAE,CAAC;CAC7C,CAAC;AAEF,eAAO,MAAM,6BAA6B,EAAE,CAAC,CAAC,OAAO,CAAC,mBAAmB,CAqBtE,CAAC;AAEJ,eAAO,MAAM,qCAAqC,EAAE,CAAC,CAAC,OAAO,CAAC,GAAG,CAe/D,CAAC"}
|
|
@@ -9,14 +9,12 @@ import { PaginationMeta$zodSchema } from "./paginationmeta.js";
|
|
|
9
9
|
*/
|
|
10
10
|
export const TaskType = {
|
|
11
11
|
Stt: "stt",
|
|
12
|
-
Tts: "tts",
|
|
13
12
|
Llm: "llm",
|
|
14
13
|
LlmGeneral: "llm-general",
|
|
15
14
|
Conversation: "conversation",
|
|
16
15
|
};
|
|
17
16
|
export const TaskType$zodSchema = z.enum([
|
|
18
17
|
"stt",
|
|
19
|
-
"tts",
|
|
20
18
|
"llm",
|
|
21
19
|
"llm-general",
|
|
22
20
|
"conversation",
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"tasksummaryresponse.js","sourceRoot":"","sources":["../../src/models/tasksummaryresponse.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EAAE,gBAAgB,IAAI,iBAAiB,EAAE,MAAM,mBAAmB,CAAC;AAE1E,OAAO,EAAkB,wBAAwB,EAAE,MAAM,qBAAqB,CAAC;AAE/E;;GAEG;AACH,MAAM,CAAC,MAAM,QAAQ,GAAG;IACtB,GAAG,EAAE,KAAK;IACV,GAAG,EAAE,KAAK;IACV,
|
|
1
|
+
{"version":3,"file":"tasksummaryresponse.js","sourceRoot":"","sources":["../../src/models/tasksummaryresponse.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EAAE,gBAAgB,IAAI,iBAAiB,EAAE,MAAM,mBAAmB,CAAC;AAE1E,OAAO,EAAkB,wBAAwB,EAAE,MAAM,qBAAqB,CAAC;AAE/E;;GAEG;AACH,MAAM,CAAC,MAAM,QAAQ,GAAG;IACtB,GAAG,EAAE,KAAK;IACV,GAAG,EAAE,KAAK;IACV,UAAU,EAAE,aAAa;IACzB,YAAY,EAAE,cAAc;CACpB,CAAC;AAMX,MAAM,CAAC,MAAM,kBAAkB,GAAG,CAAC,CAAC,IAAI,CAAC;IACvC,KAAK;IACL,KAAK;IACL,aAAa;IACb,cAAc;CACf,CAAC,CAAC,QAAQ,CAAC,kBAAkB,CAAC,CAAC;AAahC,MAAM,CAAC,MAAM,6BAA6B,GACxC,iBAAiB,CACf,CAAC,CAAC,MAAM,CAAC;IACP,UAAU,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC,CAAC,QAAQ,CACzD,oFAAoF,CACrF;IACD,UAAU,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC,CAAC,QAAQ,CACzD,uEAAuE,CACxE;IACD,aAAa,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC,QAAQ,CACnD,sCAAsC,CACvC;IACD,UAAU,EAAE,wBAAwB;IACpC,IAAI,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC,CAAC,QAAQ,CACnD,8GAA8G,CAC/G;IACD,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,gBAAgB,CAAC;IAC9C,SAAS,EAAE,kBAAkB,CAAC,QAAQ,CAAC,kBAAkB,CAAC;CAC3D,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,GAAG,EAAE,CAAC,EACpB,sBAAsB,EACtB,IAAI,CACL,CAAC;AAEJ,MAAM,CAAC,MAAM,qCAAqC,GAAmB,CAAC,CAAC,MAAM,CAAC;IAC5E,oBAAoB,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC,QAAQ,EAAE;IAC9D,UAAU,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC;IAClD,UAAU,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC;IAClD,aAAa,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC;IAC5C,UAAU,EAAE,wBAAwB;IACpC,IAAI,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC;IAC5C,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE;IACnB,SAAS,EAAE,kBAAkB;CAC9B,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,EAAE,EAAE;IACjB,MAAM,EAAE,oBAAoB,EAAE,GAAG,IAAI,EAAE,GAAG,CAAC,CAAC;IAC5C,OAAO;QACL,GAAG,IAAI;QACP,GAAG,oBAAoB;KACxB,CAAC;AACJ,CAAC,CAAC,CAAC"}
|
package/package.json
CHANGED
package/src/landing-page.ts
CHANGED
|
@@ -930,7 +930,7 @@ http_headers = { "api-key-auth" = "YOUR_API_KEY_AUTH" }`;
|
|
|
930
930
|
<h1>Instructions</h1>
|
|
931
931
|
<p>One-click installation for Claude Desktop users</p>
|
|
932
932
|
<div class="instruction-item">
|
|
933
|
-
<a href="https://github.com/dalmia/calibrate-mcp/releases/download/v0.0.
|
|
933
|
+
<a href="https://github.com/dalmia/calibrate-mcp/releases/download/v0.0.16/mcp-server.mcpb" download="mcp-server.mcpb" class="action-button header-action" style="display: inline-flex; margin-bottom: 16px;">
|
|
934
934
|
📥 Download MCP Bundle
|
|
935
935
|
</a>
|
|
936
936
|
</div>
|
package/src/lib/config.ts
CHANGED
|
@@ -65,8 +65,8 @@ export function serverURLFromOptions(options: SDKOptions): URL | null {
|
|
|
65
65
|
export const SDK_METADATA = {
|
|
66
66
|
language: "typescript",
|
|
67
67
|
openapiDocVersion: "0.1.0",
|
|
68
|
-
sdkVersion: "0.0.
|
|
68
|
+
sdkVersion: "0.0.16",
|
|
69
69
|
genVersion: "2.915.1",
|
|
70
70
|
userAgent:
|
|
71
|
-
"speakeasy-sdk/mcp-typescript 0.0.
|
|
71
|
+
"speakeasy-sdk/mcp-typescript 0.0.16 2.915.1 0.1.0 @dalmia/calibrate-mcp",
|
|
72
72
|
} as const;
|
package/src/mcp-server/server.ts
CHANGED
|
@@ -15,6 +15,6 @@ export const AnnotationItemPayload$zodSchema: z.ZodType<AnnotationItemPayload> =
|
|
|
15
15
|
"Human annotations to seed, keyed by evaluator ID. Each evaluator ID must be linked to the task. Put the judgement in `value`, a bool for binary or a number for rating, with optional `reasoning`. Requires `annotator_id`",
|
|
16
16
|
),
|
|
17
17
|
payload: z.any().optional().describe(
|
|
18
|
-
"The item's payload.
|
|
18
|
+
"The item's payload. `name` is required and unique within the task for every type. The other fields depend on the task `type`:\n\n- `stt`: `name`, `reference_transcript`, `predicted_transcript`\n- `llm`: `name`, `chat_history` (list of `{role, content}` turns ending at the user turn), `agent_response` (the reply to judge), optional `evaluator_variables`\n- `llm-general`: `name`, `input`, `output`, optional `evaluator_variables`\n- `conversation`: `name`, `transcript` (list of `{role, content}` turns), optional `evaluator_variables`\n\n`evaluator_variables` maps an evaluator ID to that evaluator's `{{\"{{\"}}variable}}` values for this item",
|
|
19
19
|
),
|
|
20
20
|
});
|
|
@@ -10,14 +10,12 @@ import { ClosedEnum } from "../types/enums.js";
|
|
|
10
10
|
*
|
|
11
11
|
* @remarks
|
|
12
12
|
* - `stt`: judge a transcript on its own
|
|
13
|
-
* - `tts`: judge synthesized speech output
|
|
14
13
|
* - `llm`: judge one response with its conversation history
|
|
15
14
|
* - `llm-general`: judge a standalone `input -> output` pair
|
|
16
15
|
* - `conversation`: judge a full conversation
|
|
17
16
|
*/
|
|
18
17
|
export const AnnotationTaskCreateType = {
|
|
19
18
|
Stt: "stt",
|
|
20
|
-
Tts: "tts",
|
|
21
19
|
Llm: "llm",
|
|
22
20
|
LlmGeneral: "llm-general",
|
|
23
21
|
Conversation: "conversation",
|
|
@@ -27,7 +25,6 @@ export const AnnotationTaskCreateType = {
|
|
|
27
25
|
*
|
|
28
26
|
* @remarks
|
|
29
27
|
* - `stt`: judge a transcript on its own
|
|
30
|
-
* - `tts`: judge synthesized speech output
|
|
31
28
|
* - `llm`: judge one response with its conversation history
|
|
32
29
|
* - `llm-general`: judge a standalone `input -> output` pair
|
|
33
30
|
* - `conversation`: judge a full conversation
|
|
@@ -38,12 +35,11 @@ export type AnnotationTaskCreateType = ClosedEnum<
|
|
|
38
35
|
|
|
39
36
|
export const AnnotationTaskCreateType$zodSchema = z.enum([
|
|
40
37
|
"stt",
|
|
41
|
-
"tts",
|
|
42
38
|
"llm",
|
|
43
39
|
"llm-general",
|
|
44
40
|
"conversation",
|
|
45
41
|
]).describe(
|
|
46
|
-
"Task type. Determines the shape of each item's payload.\n- `stt`: judge a transcript on its own\n- `
|
|
42
|
+
"Task type. Determines the shape of each item's payload.\n- `stt`: judge a transcript on its own\n- `llm`: judge one response with its conversation history\n- `llm-general`: judge a standalone `input -> output` pair\n- `conversation`: judge a full conversation",
|
|
47
43
|
);
|
|
48
44
|
|
|
49
45
|
export type AnnotationTaskCreate = {
|
|
@@ -63,6 +59,6 @@ export const AnnotationTaskCreate$zodSchema: z.ZodType<AnnotationTaskCreate> = z
|
|
|
63
59
|
),
|
|
64
60
|
name: z.string().describe("Task name, unique within your workspace"),
|
|
65
61
|
type: AnnotationTaskCreateType$zodSchema.describe(
|
|
66
|
-
"Task type. Determines the shape of each item's payload.\n- `stt`: judge a transcript on its own\n- `
|
|
62
|
+
"Task type. Determines the shape of each item's payload.\n- `stt`: judge a transcript on its own\n- `llm`: judge one response with its conversation history\n- `llm-general`: judge a standalone `input -> output` pair\n- `conversation`: judge a full conversation",
|
|
67
63
|
),
|
|
68
64
|
});
|
|
@@ -10,14 +10,12 @@ import { ClosedEnum } from "../types/enums.js";
|
|
|
10
10
|
*
|
|
11
11
|
* @remarks
|
|
12
12
|
* - `stt`: judge a transcript on its own
|
|
13
|
-
* - `tts`: judge synthesized speech output
|
|
14
13
|
* - `llm`: judge one response with its conversation history
|
|
15
14
|
* - `llm-general`: judge a standalone `input -> output` pair
|
|
16
15
|
* - `conversation`: judge a full conversation
|
|
17
16
|
*/
|
|
18
17
|
export const AnnotationTaskResponseType = {
|
|
19
18
|
Stt: "stt",
|
|
20
|
-
Tts: "tts",
|
|
21
19
|
Llm: "llm",
|
|
22
20
|
LlmGeneral: "llm-general",
|
|
23
21
|
Conversation: "conversation",
|
|
@@ -27,7 +25,6 @@ export const AnnotationTaskResponseType = {
|
|
|
27
25
|
*
|
|
28
26
|
* @remarks
|
|
29
27
|
* - `stt`: judge a transcript on its own
|
|
30
|
-
* - `tts`: judge synthesized speech output
|
|
31
28
|
* - `llm`: judge one response with its conversation history
|
|
32
29
|
* - `llm-general`: judge a standalone `input -> output` pair
|
|
33
30
|
* - `conversation`: judge a full conversation
|
|
@@ -38,12 +35,11 @@ export type AnnotationTaskResponseType = ClosedEnum<
|
|
|
38
35
|
|
|
39
36
|
export const AnnotationTaskResponseType$zodSchema = z.enum([
|
|
40
37
|
"stt",
|
|
41
|
-
"tts",
|
|
42
38
|
"llm",
|
|
43
39
|
"llm-general",
|
|
44
40
|
"conversation",
|
|
45
41
|
]).describe(
|
|
46
|
-
"Task type. Determines the shape of each item's payload.\n- `stt`: judge a transcript on its own\n- `
|
|
42
|
+
"Task type. Determines the shape of each item's payload.\n- `stt`: judge a transcript on its own\n- `llm`: judge one response with its conversation history\n- `llm-general`: judge a standalone `input -> output` pair\n- `conversation`: judge a full conversation",
|
|
47
43
|
);
|
|
48
44
|
|
|
49
45
|
export type AnnotationTaskResponse = {
|
|
@@ -80,7 +76,7 @@ export const AnnotationTaskResponse$zodSchema: z.ZodType<
|
|
|
80
76
|
),
|
|
81
77
|
name: z.string().describe("Name of the task"),
|
|
82
78
|
type: AnnotationTaskResponseType$zodSchema.describe(
|
|
83
|
-
"Task type. Determines the shape of each item's payload.\n- `stt`: judge a transcript on its own\n- `
|
|
79
|
+
"Task type. Determines the shape of each item's payload.\n- `stt`: judge a transcript on its own\n- `llm`: judge one response with its conversation history\n- `llm-general`: judge a standalone `input -> output` pair\n- `conversation`: judge a full conversation",
|
|
84
80
|
),
|
|
85
81
|
updated_at: z.string().describe(
|
|
86
82
|
"When the task was last updated (ISO 8601 UTC)",
|
|
@@ -12,7 +12,6 @@ import { PaginationMeta, PaginationMeta$zodSchema } from "./paginationmeta.js";
|
|
|
12
12
|
*/
|
|
13
13
|
export const TaskType = {
|
|
14
14
|
Stt: "stt",
|
|
15
|
-
Tts: "tts",
|
|
16
15
|
Llm: "llm",
|
|
17
16
|
LlmGeneral: "llm-general",
|
|
18
17
|
Conversation: "conversation",
|
|
@@ -24,7 +23,6 @@ export type TaskType = ClosedEnum<typeof TaskType>;
|
|
|
24
23
|
|
|
25
24
|
export const TaskType$zodSchema = z.enum([
|
|
26
25
|
"stt",
|
|
27
|
-
"tts",
|
|
28
26
|
"llm",
|
|
29
27
|
"llm-general",
|
|
30
28
|
"conversation",
|