edgegate-mcp 0.10.0 → 0.11.0
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/client.d.ts +25 -2
- package/dist/client.js +7 -0
- package/dist/client.js.map +1 -1
- package/dist/server.js +27 -1
- package/dist/server.js.map +1 -1
- package/dist/tools/check_llm_compile_status.d.ts +15 -0
- package/dist/tools/check_llm_compile_status.js +49 -0
- package/dist/tools/check_llm_compile_status.js.map +1 -0
- package/dist/tools/create_pipeline.d.ts +139 -17
- package/dist/tools/create_pipeline.js +155 -29
- package/dist/tools/create_pipeline.js.map +1 -1
- package/dist/tools/import_huggingface_model.js +23 -15
- package/dist/tools/import_huggingface_model.js.map +1 -1
- package/dist/tools/llm_compile.d.ts +47 -0
- package/dist/tools/llm_compile.js +111 -0
- package/dist/tools/llm_compile.js.map +1 -0
- package/dist/types.d.ts +46 -0
- package/dist/version.d.ts +2 -2
- package/dist/version.js +1 -1
- package/dist/version.js.map +1 -1
- package/package.json +1 -1
package/dist/client.d.ts
CHANGED
|
@@ -1,4 +1,4 @@
|
|
|
1
|
-
import type { APIKeyCreateResponse, APIKeyCreatedResponse, APIKeyListItem, ArtifactResponse, ByoArtifactRegisterRequest, ByoAuditPage, ByoGrant, ByoSetupInfo, DeviceListResponse, HFImportJob, HuggingFaceConnectResponse, HuggingFaceIntegrationStatus, InputSpec, Member, Pipeline, PromptPackCreateBody, PromptPackSummary, QaihubIntegration, RunBundle, RunComparison, RunDetail, RunSummary, Workspace, WorkflowTemplate, WorkspaceRole } from "./types.js";
|
|
1
|
+
import type { APIKeyCreateResponse, APIKeyCreatedResponse, APIKeyListItem, ArtifactResponse, ByoArtifactRegisterRequest, ByoAuditPage, ByoGrant, ByoSetupInfo, DeviceListResponse, HFImportJob, HuggingFaceConnectResponse, HuggingFaceIntegrationStatus, InputSpec, LLMCompileJob, Member, Pipeline, PromptPackCreateBody, PromptPackSummary, QaihubIntegration, RunBundle, RunComparison, RunDetail, RunSummary, Workspace, WorkflowTemplate, WorkspaceRole } from "./types.js";
|
|
2
2
|
export interface EdgeGateClientOptions {
|
|
3
3
|
apiUrl: string;
|
|
4
4
|
apiKey: string;
|
|
@@ -41,6 +41,13 @@ export declare class EdgeGateClient {
|
|
|
41
41
|
operator: string;
|
|
42
42
|
threshold: number;
|
|
43
43
|
description?: string;
|
|
44
|
+
} | {
|
|
45
|
+
type: "composition";
|
|
46
|
+
op_type: string;
|
|
47
|
+
compute_unit: "NPU" | "GPU" | "CPU";
|
|
48
|
+
operator: string;
|
|
49
|
+
threshold: number;
|
|
50
|
+
description?: string;
|
|
44
51
|
}>;
|
|
45
52
|
run_policy?: {
|
|
46
53
|
warmup_runs?: number;
|
|
@@ -49,13 +56,29 @@ export declare class EdgeGateClient {
|
|
|
49
56
|
timeout_minutes?: number;
|
|
50
57
|
};
|
|
51
58
|
model_matrix?: Array<{
|
|
52
|
-
artifact_id
|
|
59
|
+
artifact_id?: string;
|
|
53
60
|
label?: string;
|
|
61
|
+
llm_compile_source?: {
|
|
62
|
+
source_artifact_ids: string[];
|
|
63
|
+
sequence_lengths?: number[];
|
|
64
|
+
context_lengths?: number[];
|
|
65
|
+
roles?: string[];
|
|
66
|
+
target_runtime?: string;
|
|
67
|
+
};
|
|
54
68
|
}>;
|
|
55
69
|
/** Optional explicit AI Hub compile input shapes. Omit to let EdgeGate
|
|
56
70
|
* auto-detect from the ONNX file (works for most models). */
|
|
57
71
|
input_specs?: Record<string, InputSpec>;
|
|
58
72
|
}): Promise<Pipeline>;
|
|
73
|
+
submitLLMCompile(workspaceId: string, body: {
|
|
74
|
+
source_artifact_ids: string[];
|
|
75
|
+
device_id: string;
|
|
76
|
+
target_runtime?: string;
|
|
77
|
+
sequence_lengths?: number[];
|
|
78
|
+
context_lengths?: number[];
|
|
79
|
+
roles?: string[];
|
|
80
|
+
}): Promise<LLMCompileJob>;
|
|
81
|
+
getLLMCompileJob(workspaceId: string, compileJobId: string): Promise<LLMCompileJob>;
|
|
59
82
|
listPipelines(workspaceId: string): Promise<Pipeline[]>;
|
|
60
83
|
triggerRun(workspaceId: string, body: {
|
|
61
84
|
pipeline_id: string;
|
package/dist/client.js
CHANGED
|
@@ -39,6 +39,13 @@ export class EdgeGateClient {
|
|
|
39
39
|
async createPipeline(workspaceId, body) {
|
|
40
40
|
return this.request("POST", `/v1/workspaces/${workspaceId}/pipelines`, body);
|
|
41
41
|
}
|
|
42
|
+
// LLM compile (Plan B) ───────────────────────────────────────────────
|
|
43
|
+
async submitLLMCompile(workspaceId, body) {
|
|
44
|
+
return this.request("POST", `/v1/workspaces/${workspaceId}/llm-compile`, body);
|
|
45
|
+
}
|
|
46
|
+
async getLLMCompileJob(workspaceId, compileJobId) {
|
|
47
|
+
return this.request("GET", `/v1/workspaces/${workspaceId}/llm-compile/${compileJobId}`);
|
|
48
|
+
}
|
|
42
49
|
async listPipelines(workspaceId) {
|
|
43
50
|
return this.request("GET", `/v1/workspaces/${workspaceId}/pipelines`);
|
|
44
51
|
}
|
package/dist/client.js.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"client.js","sourceRoot":"","sources":["../src/client.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,UAAU,EAAE,MAAM,cAAc,CAAC;
|
|
1
|
+
{"version":3,"file":"client.js","sourceRoot":"","sources":["../src/client.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,UAAU,EAAE,MAAM,cAAc,CAAC;AAsC1C,MAAM,OAAO,aAAc,SAAQ,KAAK;IAEpB;IACA;IACA;IAHlB,YACkB,MAAc,EACd,MAAc,EACd,GAAW;QAE3B,KAAK,CAAC,YAAY,MAAM,OAAO,GAAG,KAAK,MAAM,EAAE,CAAC,CAAC;QAJjC,WAAM,GAAN,MAAM,CAAQ;QACd,WAAM,GAAN,MAAM,CAAQ;QACd,QAAG,GAAH,GAAG,CAAQ;QAG3B,IAAI,CAAC,IAAI,GAAG,eAAe,CAAC;IAC9B,CAAC;CACF;AAED,MAAM,OAAO,cAAc;IACR,MAAM,CAAS;IACf,MAAM,CAAS;IACf,YAAY,CAAS;IACrB,UAAU,CAAS;IACnB,SAAS,CAAS;IAEnC,YAAY,IAA2B;QACrC,IAAI,CAAC,MAAM,GAAG,IAAI,CAAC,MAAM,CAAC,OAAO,CAAC,KAAK,EAAE,EAAE,CAAC,CAAC;QAC7C,IAAI,CAAC,MAAM,GAAG,IAAI,CAAC,MAAM,CAAC;QAC1B,IAAI,CAAC,YAAY,GAAG,IAAI,CAAC,YAAY,IAAI,GAAG,CAAC;QAC7C,IAAI,CAAC,UAAU,GAAG,IAAI,CAAC,UAAU,IAAI,CAAC,CAAC;QACvC,IAAI,CAAC,SAAS,GAAG,IAAI,CAAC,SAAS,IAAI,MAAM,CAAC;IAC5C,CAAC;IAED,KAAK,CAAC,YAAY,CAAC,WAAmB;QACpC,OAAO,IAAI,CAAC,OAAO,CAAY,KAAK,EAAE,kBAAkB,WAAW,EAAE,CAAC,CAAC;IACzE,CAAC;IACD,KAAK,CAAC,cAAc;QAClB,OAAO,IAAI,CAAC,OAAO,CAAc,KAAK,EAAE,gBAAgB,CAAC,CAAC;IAC5D,CAAC;IACD,KAAK,CAAC,WAAW;QACf,OAAO,IAAI,CAAC,OAAO,CAAqB,KAAK,EAAE,mBAAmB,CAAC,CAAC;IACtE,CAAC;IACD,KAAK,CAAC,YAAY,CAChB,WAAmB,EACnB,IAA2C;QAE3C,OAAO,IAAI,CAAC,OAAO,CACjB,MAAM,EACN,kBAAkB,WAAW,WAAW,EACxC,IAAI,CACL,CAAC;IACJ,CAAC;IACD,KAAK,CAAC,cAAc,CAClB,WAAmB,EACnB,IAuCC;QAED,OAAO,IAAI,CAAC,OAAO,CAAW,MAAM,EAAE,kBAAkB,WAAW,YAAY,EAAE,IAAI,CAAC,CAAC;IACzF,CAAC;IAED,uEAAuE;IACvE,KAAK,CAAC,gBAAgB,CACpB,WAAmB,EACnB,IAOC;QAED,OAAO,IAAI,CAAC,OAAO,CACjB,MAAM,EACN,kBAAkB,WAAW,cAAc,EAC3C,IAAI,CACL,CAAC;IACJ,CAAC;IACD,KAAK,CAAC,gBAAgB,CACpB,WAAmB,EACnB,YAAoB;QAEpB,OAAO,IAAI,CAAC,OAAO,CACjB,KAAK,EACL,kBAAkB,WAAW,gBAAgB,YAAY,EAAE,CAC5D,CAAC;IACJ,CAAC;IACD,KAAK,CAAC,aAAa,CAAC,WAAmB;QACrC,OAAO,IAAI,CAAC,OAAO,CAAa,KAAK,EAAE,kBAAkB,WAAW,YAAY,CAAC,CAAC;IACpF,CAAC;IACD,KAAK,CAAC,UAAU,CACd,WAAmB,EACnB,IAA2E;QAE3E,OAAO,IAAI,CAAC,OAAO,CAAa,MAAM,EAAE,kBAAkB,WAAW,OAAO,EAAE,IAAI,CAAC,CAAC;IACtF,CAAC;IACD,KAAK,CAAC,MAAM,CAAC,WAAmB,EAAE,KAAa;QAC7C,OAAO,IAAI,CAAC,OAAO,CAAY,KAAK,EAAE,kBAAkB,WAAW,SAAS,KAAK,EAAE,CAAC,CAAC;IACvF,CAAC;IACD,KAAK,CAAC,QAAQ,CAAC,WAAmB,EAAE,KAAK,GAAG,EAAE;QAC5C,OAAO,IAAI,CAAC,OAAO,CACjB,KAAK,EACL,kBAAkB,WAAW,eAAe,KAAK,EAAE,CACpD,CAAC;IACJ,CAAC;IACD,KAAK,CAAC,kBAAkB,CACtB,WAAmB,EACnB,UAAkB,EAClB,KAAK,GAAG,EAAE;QAEV,OAAO,IAAI,CAAC,OAAO,CACjB,KAAK,EACL,kBAAkB,WAAW,qBAAqB,UAAU,UAAU,KAAK,EAAE,CAC9E,CAAC;IACJ,CAAC;IACD,KAAK,CAAC,UAAU,CAAC,WAAmB,EAAE,KAAa;QACjD,OAAO,IAAI,CAAC,OAAO,CACjB,KAAK,EACL,kBAAkB,WAAW,SAAS,KAAK,OAAO,CACnD,CAAC;IACJ,CAAC;IACD,KAAK,CAAC,YAAY,CAAC,WAAmB,EAAE,KAAa;QACnD,OAAO,IAAI,CAAC,OAAO,CAAY,KAAK,EAAE,kBAAkB,WAAW,SAAS,KAAK,SAAS,CAAC,CAAC;IAC9F,CAAC;IACD,KAAK,CAAC,mBAAmB,CAAC,WAAmB;QAC3C,OAAO,IAAI,CAAC,OAAO,CACjB,KAAK,EACL,kBAAkB,WAAW,yBAAyB,CACvD,CAAC;IACJ,CAAC;IACD,KAAK,CAAC,sBAAsB,CAC1B,WAAmB,EACnB,IAAkE;QAElE,OAAO,IAAI,CAAC,OAAO,CACjB,MAAM,EACN,kBAAkB,WAAW,+BAA+B,EAC5D,IAAI,CACL,CAAC;IACJ,CAAC;IACD,KAAK,CAAC,uBAAuB,CAAC,WAAmB,EAAE,KAAa;QAC9D,OAAO,IAAI,CAAC,OAAO,CACjB,KAAK,EACL,kBAAkB,WAAW,iCAAiC,KAAK,EAAE,CACtE,CAAC;IACJ,CAAC;IACD,KAAK,CAAC,6BAA6B,CACjC,WAAmB,EACnB,KAAa;QAEb,OAAO,IAAI,CAAC,OAAO,CACjB,MAAM,EACN,kBAAkB,WAAW,2BAA2B,EACxD,EAAE,KAAK,EAAE,CACV,CAAC;IACJ,CAAC;IACD,KAAK,CAAC,4BAA4B,CAChC,WAAmB,EACnB,KAAa;QAEb,OAAO,IAAI,CAAC,OAAO,CACjB,KAAK,EACL,kBAAkB,WAAW,kCAAkC,EAC/D,EAAE,KAAK,EAAE,CACV,CAAC;IACJ,CAAC;IACD,KAAK,CAAC,yBAAyB,CAC7B,WAAmB;QAEnB,OAAO,IAAI,CAAC,OAAO,CACjB,KAAK,EACL,kBAAkB,WAAW,2BAA2B,CACzD,CAAC;IACJ,CAAC;IACD,KAAK,CAAC,4BAA4B,CAAC,WAAmB;QACpD,MAAM,IAAI,CAAC,OAAO,CAChB,QAAQ,EACR,kBAAkB,WAAW,2BAA2B,CACzD,CAAC;IACJ,CAAC;IAED,wEAAwE;IACxE,KAAK,CAAC,wBAAwB,CAC5B,WAAmB,EACnB,KAAa;QAEb,OAAO,IAAI,CAAC,OAAO,CACjB,MAAM,EACN,kBAAkB,WAAW,sBAAsB,EACnD,EAAE,KAAK,EAAE,CACV,CAAC;IACJ,CAAC;IACD,KAAK,CAAC,uBAAuB,CAC3B,WAAmB,EACnB,KAAa;QAEb,OAAO,IAAI,CAAC,OAAO,CACjB,KAAK,EACL,kBAAkB,WAAW,6BAA6B,EAC1D,EAAE,KAAK,EAAE,CACV,CAAC;IACJ,CAAC;IACD,KAAK,CAAC,oBAAoB,CAAC,WAAmB;QAC5C,OAAO,IAAI,CAAC,OAAO,CACjB,KAAK,EACL,kBAAkB,WAAW,sBAAsB,CACpD,CAAC;IACJ,CAAC;IACD,KAAK,CAAC,uBAAuB,CAAC,WAAmB;QAC/C,MAAM,IAAI,CAAC,OAAO,CAChB,QAAQ,EACR,kBAAkB,WAAW,sBAAsB,CACpD,CAAC;IACJ,CAAC;IAED,uEAAuE;IACvE,KAAK,CAAC,eAAe,CAAC,IAAY;QAChC,OAAO,IAAI,CAAC,OAAO,CAAY,MAAM,EAAE,gBAAgB,EAAE,EAAE,IAAI,EAAE,CAAC,CAAC;IACrE,CAAC;IAED,uEAAuE;IACvE,KAAK,CAAC,WAAW,CAAC,WAAmB;QACnC,OAAO,IAAI,CAAC,OAAO,CACjB,KAAK,EACL,kBAAkB,WAAW,WAAW,CACzC,CAAC;IACJ,CAAC;IACD,KAAK,CAAC,YAAY,CAChB,WAAmB,EACnB,IAAkD;QAElD,OAAO,IAAI,CAAC,OAAO,CACjB,MAAM,EACN,kBAAkB,WAAW,WAAW,EACxC,IAAI,CACL,CAAC;IACJ,CAAC;IACD,KAAK,CAAC,YAAY,CAAC,WAAmB,EAAE,KAAa;QACnD,MAAM,IAAI,CAAC,OAAO,CAChB,QAAQ,EACR,kBAAkB,WAAW,aAAa,KAAK,EAAE,CAClD,CAAC;IACJ,CAAC;IAED,uEAAuE;IACvE,KAAK,CAAC,WAAW,CAAC,WAAmB;QACnC,OAAO,IAAI,CAAC,OAAO,CACjB,KAAK,EACL,kBAAkB,WAAW,UAAU,CACxC,CAAC;IACJ,CAAC;IACD,KAAK,CAAC,SAAS,CACb,WAAmB,EACnB,IAAiD;QAEjD,OAAO,IAAI,CAAC,OAAO,CACjB,MAAM,EACN,kBAAkB,WAAW,UAAU,EACvC,IAAI,CACL,CAAC;IACJ,CAAC;IACD,KAAK,CAAC,gBAAgB,CACpB,WAAmB,EACnB,MAAc,EACd,IAAmB;QAEnB,OAAO,IAAI,CAAC,OAAO,CACjB,KAAK,EACL,kBAAkB,WAAW,YAAY,MAAM,EAAE,EACjD,EAAE,IAAI,EAAE,CACT,CAAC;IACJ,CAAC;IACD,KAAK,CAAC,YAAY,CAAC,WAAmB,EAAE,MAAc;QACpD,MAAM,IAAI,CAAC,OAAO,CAChB,QAAQ,EACR,kBAAkB,WAAW,YAAY,MAAM,EAAE,CAClD,CAAC;IACJ,CAAC;IAED,KAAK,CAAC,eAAe,CAAC,WAAmB;QACvC,OAAO,IAAI,CAAC,OAAO,CACjB,KAAK,EACL,kBAAkB,WAAW,cAAc,CAC5C,CAAC;IACJ,CAAC;IACD,KAAK,CAAC,gBAAgB,CACpB,WAAmB,EACnB,IAA0B;QAE1B,OAAO,IAAI,CAAC,OAAO,CACjB,MAAM,EACN,kBAAkB,WAAW,cAAc,EAC3C,IAAI,CACL,CAAC;IACJ,CAAC;IACD,KAAK,CAAC,iBAAiB,CACrB,WAAmB,EACnB,YAAoB,EACpB,OAAe;QAEf,OAAO,IAAI,CAAC,OAAO,CACjB,KAAK,EACL,kBAAkB,WAAW,gBAAgB,YAAY,IAAI,OAAO,UAAU,CAC/E,CAAC;IACJ,CAAC;IAED,wEAAwE;IACxE,KAAK,CAAC,gBAAgB,CACpB,WAAmB,EACnB,IAKC;QAED,OAAO,IAAI,CAAC,OAAO,CACjB,MAAM,EACN,kBAAkB,WAAW,qBAAqB,EAClD,IAAI,CACL,CAAC;IACJ,CAAC;IACD;;;;;;OAMG;IACH,KAAK,CAAC,uBAAuB,CAC3B,WAAmB,EACnB,IAA6D;QAE7D,OAAO,IAAI,CAAC,OAAO,CACjB,MAAM,EACN,kBAAkB,WAAW,qBAAqB,EAClD,IAAI,CACL,CAAC;IACJ,CAAC;IACD;;;;OAIG;IACH,KAAK,CAAC,aAAa,CACjB,WAAmB,EACnB,OAAe;QAEf,OAAO,IAAI,CAAC,OAAO,CACjB,MAAM,EACN,kBAAkB,WAAW,iCAAiC,EAC9D,EAAE,QAAQ,EAAE,OAAO,EAAE,CACtB,CAAC;IACJ,CAAC;IACD,KAAK,CAAC,eAAe,CAAC,WAAmB;QACvC,OAAO,IAAI,CAAC,OAAO,CACjB,KAAK,EACL,kBAAkB,WAAW,yBAAyB,CACvD,CAAC;IACJ,CAAC;IACD,KAAK,CAAC,WAAW,CAAC,WAAmB;QACnC,OAAO,IAAI,CAAC,OAAO,CACjB,KAAK,EACL,kBAAkB,WAAW,qBAAqB,CACnD,CAAC;IACJ,CAAC;IACD,KAAK,CAAC,cAAc,CAAC,WAAmB;QACtC,OAAO,IAAI,CAAC,OAAO,CACjB,MAAM,EACN,kBAAkB,WAAW,4BAA4B,CAC1D,CAAC;IACJ,CAAC;IACD,KAAK,CAAC,cAAc,CAAC,WAAmB;QACtC,MAAM,IAAI,CAAC,OAAO,CAChB,QAAQ,EACR,kBAAkB,WAAW,qBAAqB,CACnD,CAAC;IACJ,CAAC;IACD,KAAK,CAAC,mBAAmB,CACvB,WAAmB,EACnB,IAAgC;QAEhC,OAAO,IAAI,CAAC,OAAO,CACjB,MAAM,EACN,kBAAkB,WAAW,gBAAgB,EAC7C,IAAI,CACL,CAAC;IACJ,CAAC;IACD,KAAK,CAAC,WAAW,CACf,WAAmB,EACnB,SAMI,EAAE;QAEN,MAAM,EAAE,GAAG,IAAI,eAAe,EAAE,CAAC;QACjC,IAAI,MAAM,CAAC,WAAW;YAAE,EAAE,CAAC,GAAG,CAAC,aAAa,EAAE,MAAM,CAAC,WAAW,CAAC,CAAC;QAClE,IAAI,MAAM,CAAC,MAAM;YAAE,EAAE,CAAC,GAAG,CAAC,QAAQ,EAAE,MAAM,CAAC,MAAM,CAAC,CAAC;QACnD,IAAI,MAAM,CAAC,KAAK;YAAE,EAAE,CAAC,GAAG,CAAC,OAAO,EAAE,MAAM,CAAC,KAAK,CAAC,CAAC;QAChD,IAAI,OAAO,MAAM,CAAC,MAAM,KAAK,QAAQ;YAAE,EAAE,CAAC,GAAG,CAAC,QAAQ,EAAE,MAAM,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC;QAC/E,IAAI,OAAO,MAAM,CAAC,KAAK,KAAK,QAAQ;YAAE,EAAE,CAAC,GAAG,CAAC,OAAO,EAAE,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC;QAC5E,MAAM,MAAM,GAAG,EAAE,CAAC,QAAQ,EAAE,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,QAAQ,EAAE,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QACxD,OAAO,IAAI,CAAC,OAAO,CACjB,KAAK,EACL,kBAAkB,WAAW,qBAAqB,MAAM,EAAE,CAC3D,CAAC;IACJ,CAAC;IAEO,KAAK,CAAC,OAAO,CACnB,MAAyC,EACzC,IAAY,EACZ,IAAc;QAEd,MAAM,GAAG,GAAG,GAAG,IAAI,CAAC,MAAM,GAAG,IAAI,EAAE,CAAC;QACpC,MAAM,YAAY,GAAG,MAAM,KAAK,KAAK,CAAC;QACtC,MAAM,QAAQ,GAAG,YAAY,CAAC,CAAC,CAAC,IAAI,CAAC,UAAU,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;QACxD,KAAK,IAAI,OAAO,GAAG,CAAC,EAAE,OAAO,IAAI,QAAQ,EAAE,OAAO,EAAE,EAAE,CAAC;YACrD,MAAM,OAAO,GAA2B;gBACtC,aAAa,EAAE,UAAU,IAAI,CAAC,MAAM,EAAE;gBACtC,YAAY,EAAE,UAAU;aACzB,CAAC;YACF,IAAI,IAAI,KAAK,SAAS,EAAE,CAAC;gBACvB,OAAO,CAAC,cAAc,CAAC,GAAG,kBAAkB,CAAC;YAC/C,CAAC;YACD,MAAM,IAAI,GAAG,MAAM,KAAK,CAAC,GAAG,EAAE;gBAC5B,MAAM;gBACN,OAAO;gBACP,IAAI,EAAE,IAAI,KAAK,SAAS,CAAC,CAAC,CAAC,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,SAAS;gBAC3D,MAAM,EAAE,WAAW,CAAC,OAAO,CAAC,IAAI,CAAC,SAAS,CAAC;aAC5C,CAAC,CAAC;YACH,MAAM,IAAI,GAAG,MAAM,IAAI,CAAC,IAAI,EAAE,CAAC;YAC/B,MAAM,IAAI,GAAG,IAAI,CAAC,CAAC,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC;YAC1C,IAAI,IAAI,CAAC,EAAE,EAAE,CAAC;gBACZ,OAAO,IAAS,CAAC;YACnB,CAAC;YACD,MAAM,MAAM,GACV,IAAI,IAAI,OAAO,IAAI,KAAK,QAAQ,IAAI,QAAQ,IAAI,IAAI;gBAClD,CAAC,CAAC,MAAM,CAAE,IAA4B,CAAC,MAAM,CAAC;gBAC9C,CAAC,CAAC,IAAI,IAAI,eAAe,CAAC;YAC9B,IAAI,IAAI,CAAC,MAAM,IAAI,GAAG,IAAI,YAAY,IAAI,OAAO,GAAG,QAAQ,EAAE,CAAC;gBAC7D,MAAM,KAAK,CAAC,IAAI,CAAC,YAAY,CAAC,CAAC;gBAC/B,SAAS;YACX,CAAC;YACD,MAAM,IAAI,aAAa,CAAC,IAAI,CAAC,MAAM,EAAE,MAAM,EAAE,GAAG,CAAC,CAAC;QACpD,CAAC;QACD,MAAM,IAAI,KAAK,CAAC,gCAAgC,CAAC,CAAC;IACpD,CAAC;CACF;AAED,SAAS,QAAQ,CAAC,IAAY;IAC5B,IAAI,CAAC;QAAC,OAAO,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;IAAC,CAAC;IAAC,MAAM,CAAC;QAAC,OAAO,IAAI,CAAC;IAAC,CAAC;AACzD,CAAC;AACD,SAAS,KAAK,CAAC,EAAU;IACvB,OAAO,IAAI,OAAO,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,UAAU,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC;AAC/C,CAAC"}
|
package/dist/server.js
CHANGED
|
@@ -39,6 +39,8 @@ import { checkByoBucketHandler, checkByoBucketInputSchema, } from "./tools/check
|
|
|
39
39
|
import { registerByoArtifactHandler, registerByoArtifactInputSchema, } from "./tools/register_byo_artifact.js";
|
|
40
40
|
import { disconnectByoBucketHandler, disconnectByoBucketInputSchema, } from "./tools/disconnect_byo_bucket.js";
|
|
41
41
|
import { getByoAuditHandler, getByoAuditInputSchema, } from "./tools/get_byo_audit.js";
|
|
42
|
+
import { llmCompileHandler, llmCompileInputSchema, } from "./tools/llm_compile.js";
|
|
43
|
+
import { checkLLMCompileStatusHandler, checkLLMCompileStatusInputSchema, } from "./tools/check_llm_compile_status.js";
|
|
42
44
|
const TOOLS = [
|
|
43
45
|
{
|
|
44
46
|
name: "edgegate_setup_workspace",
|
|
@@ -50,7 +52,12 @@ const TOOLS = [
|
|
|
50
52
|
{
|
|
51
53
|
name: "edgegate_create_pipeline",
|
|
52
54
|
description: "Create a new EdgeGate regression pipeline. Define which model(s), which device(s), " +
|
|
53
|
-
"and which gates (e.g. inference_time_ms ≤ 10) the pipeline will enforce."
|
|
55
|
+
"and which gates (e.g. inference_time_ms ≤ 10) the pipeline will enforce. " +
|
|
56
|
+
"For LLMs: set llm_compile_source on a model instead of artifact_id. EdgeGate will " +
|
|
57
|
+
"compile + link via AI Hub on first run; subsequent runs reuse the cached composite. " +
|
|
58
|
+
"ttft_ms + tps gates work; both are derived from per-component profile (prompt-role " +
|
|
59
|
+
"inference_time → TTFT; 1000/token-role inference_time → TPS). Each LLM gate run = " +
|
|
60
|
+
"3 AI Hub profile jobs (one per component) ≈ 3× CV cost.",
|
|
54
61
|
schema: createPipelineInputSchema,
|
|
55
62
|
handler: createPipelineHandler,
|
|
56
63
|
},
|
|
@@ -325,6 +332,25 @@ const TOOLS = [
|
|
|
325
332
|
schema: disconnectByoBucketInputSchema,
|
|
326
333
|
handler: disconnectByoBucketHandler,
|
|
327
334
|
},
|
|
335
|
+
{
|
|
336
|
+
name: "edgegate_llm_compile",
|
|
337
|
+
description: "Submit a multi-component LLM compile + link job via Qualcomm AI Hub. Returns a " +
|
|
338
|
+
"compile_job_id; poll with edgegate_check_llm_compile_status. Spend is gated by " +
|
|
339
|
+
"the workspace's monthly LLM compile cap (default 100/mo Pro tier). Each compile " +
|
|
340
|
+
"produces a composite QNN_DLC linked model + 3 component artifacts (prompt / token / " +
|
|
341
|
+
"kv_cache). Compile target_runtime is QNN_DLC under the hood despite the genie " +
|
|
342
|
+
"label — the label is for downstream profile dispatch hints only.",
|
|
343
|
+
schema: llmCompileInputSchema,
|
|
344
|
+
handler: llmCompileHandler,
|
|
345
|
+
},
|
|
346
|
+
{
|
|
347
|
+
name: "edgegate_check_llm_compile_status",
|
|
348
|
+
description: "Poll an LLM compile job. Returns status (queued|running|completed|failed), " +
|
|
349
|
+
"composite_artifact_id when complete, error_detail when failed, and " +
|
|
350
|
+
"progress.compile_jobs_done / total.",
|
|
351
|
+
schema: checkLLMCompileStatusInputSchema,
|
|
352
|
+
handler: checkLLMCompileStatusHandler,
|
|
353
|
+
},
|
|
328
354
|
{
|
|
329
355
|
name: "edgegate_get_byo_audit",
|
|
330
356
|
description: "Fetch the workspace's append-only BYO storage audit log (every AssumeRole, " +
|
package/dist/server.js.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"server.js","sourceRoot":"","sources":["../src/server.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,MAAM,EAAE,MAAM,2CAA2C,CAAC;AACnE,OAAO,EAAE,oBAAoB,EAAE,MAAM,2CAA2C,CAAC;AACjF,OAAO,EACL,qBAAqB,EACrB,sBAAsB,GACvB,MAAM,oCAAoC,CAAC;AAC5C,OAAO,EAAE,eAAe,EAAE,MAAM,oBAAoB,CAAC;AACrD,OAAO,EAAE,cAAc,EAAE,MAAM,aAAa,CAAC;AAC7C,OAAO,EAAE,OAAO,EAAE,MAAM,cAAc,CAAC;AAEvC,OAAO,EAAE,qBAAqB,EAAE,yBAAyB,EAAE,MAAM,4BAA4B,CAAC;AAC9F,OAAO,EAAE,qBAAqB,EAAE,yBAAyB,EAAE,MAAM,4BAA4B,CAAC;AAC9F,OAAO,EAAE,cAAc,EAAE,kBAAkB,EAAE,MAAM,qBAAqB,CAAC;AACzE,OAAO,EAAE,kBAAkB,EAAE,sBAAsB,EAAE,MAAM,yBAAyB,CAAC;AACrF,OAAO,EAAE,gBAAgB,EAAE,oBAAoB,EAAE,MAAM,uBAAuB,CAAC;AAC/E,OAAO,EAAE,qBAAqB,EAAE,yBAAyB,EAAE,MAAM,6BAA6B,CAAC;AAC/F,OAAO,EACL,wBAAwB,EACxB,4BAA4B,GAC7B,MAAM,gCAAgC,CAAC;AACxC,OAAO,EACL,kBAAkB,EAClB,sBAAsB,GACvB,MAAM,yBAAyB,CAAC;AACjC,OAAO,EACL,sBAAsB,EACtB,0BAA0B,GAC3B,MAAM,8BAA8B,CAAC;AACtC,OAAO,EACL,6BAA6B,EAC7B,iCAAiC,GAClC,MAAM,qCAAqC,CAAC;AAC7C,OAAO,EACL,sBAAsB,EACtB,0BAA0B,GAC3B,MAAM,6BAA6B,CAAC;AACrC,OAAO,EACL,uBAAuB,EACvB,2BAA2B,GAC5B,MAAM,8BAA8B,CAAC;AACtC,OAAO,EACL,wBAAwB,EACxB,4BAA4B,GAC7B,MAAM,+BAA+B,CAAC;AACvC,OAAO,EACL,yBAAyB,EACzB,6BAA6B,GAC9B,MAAM,gCAAgC,CAAC;AACxC,OAAO,EACL,4BAA4B,EAC5B,gCAAgC,GACjC,MAAM,mCAAmC,CAAC;AAC3C,OAAO,EACL,gCAAgC,EAChC,oCAAoC,GACrC,MAAM,wCAAwC,CAAC;AAChD,OAAO,EACL,oBAAoB,EACpB,wBAAwB,GACzB,MAAM,2BAA2B,CAAC;AACnC,OAAO,EACL,2BAA2B,EAC3B,+BAA+B,GAChC,MAAM,mCAAmC,CAAC;AAC3C,OAAO,EACL,uBAAuB,EACvB,2BAA2B,GAC5B,MAAM,8BAA8B,CAAC;AACtC,OAAO,EACL,sBAAsB,EACtB,0BAA0B,GAC3B,MAAM,6BAA6B,CAAC;AACrC,OAAO,EACL,kBAAkB,EAClB,sBAAsB,GACvB,MAAM,0BAA0B,CAAC;AAClC,OAAO,EACL,mBAAmB,EACnB,uBAAuB,GACxB,MAAM,2BAA2B,CAAC;AACnC,OAAO,EACL,mBAAmB,EACnB,uBAAuB,GACxB,MAAM,2BAA2B,CAAC;AACnC,OAAO,EACL,kBAAkB,EAClB,sBAAsB,GACvB,MAAM,yBAAyB,CAAC;AACjC,OAAO,EACL,kBAAkB,EAClB,sBAAsB,GACvB,MAAM,yBAAyB,CAAC;AACjC,OAAO,EACL,mBAAmB,EACnB,uBAAuB,GACxB,MAAM,0BAA0B,CAAC;AAClC,OAAO,EACL,uBAAuB,EACvB,2BAA2B,GAC5B,MAAM,+BAA+B,CAAC;AACvC,OAAO,EACL,mBAAmB,EACnB,uBAAuB,GACxB,MAAM,0BAA0B,CAAC;AAClC,OAAO,EACL,wBAAwB,EACxB,4BAA4B,GAC7B,MAAM,gCAAgC,CAAC;AACxC,OAAO,EACL,sBAAsB,EACtB,0BAA0B,GAC3B,MAAM,8BAA8B,CAAC;AACtC,OAAO,EACL,oBAAoB,EACpB,wBAAwB,GACzB,MAAM,4BAA4B,CAAC;AACpC,OAAO,EACL,qBAAqB,EACrB,yBAAyB,GAC1B,MAAM,6BAA6B,CAAC;AACrC,OAAO,EACL,0BAA0B,EAC1B,8BAA8B,GAC/B,MAAM,kCAAkC,CAAC;AAC1C,OAAO,EACL,0BAA0B,EAC1B,8BAA8B,GAC/B,MAAM,kCAAkC,CAAC;AAC1C,OAAO,EACL,kBAAkB,EAClB,sBAAsB,GACvB,MAAM,0BAA0B,CAAC;
|
|
1
|
+
{"version":3,"file":"server.js","sourceRoot":"","sources":["../src/server.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,MAAM,EAAE,MAAM,2CAA2C,CAAC;AACnE,OAAO,EAAE,oBAAoB,EAAE,MAAM,2CAA2C,CAAC;AACjF,OAAO,EACL,qBAAqB,EACrB,sBAAsB,GACvB,MAAM,oCAAoC,CAAC;AAC5C,OAAO,EAAE,eAAe,EAAE,MAAM,oBAAoB,CAAC;AACrD,OAAO,EAAE,cAAc,EAAE,MAAM,aAAa,CAAC;AAC7C,OAAO,EAAE,OAAO,EAAE,MAAM,cAAc,CAAC;AAEvC,OAAO,EAAE,qBAAqB,EAAE,yBAAyB,EAAE,MAAM,4BAA4B,CAAC;AAC9F,OAAO,EAAE,qBAAqB,EAAE,yBAAyB,EAAE,MAAM,4BAA4B,CAAC;AAC9F,OAAO,EAAE,cAAc,EAAE,kBAAkB,EAAE,MAAM,qBAAqB,CAAC;AACzE,OAAO,EAAE,kBAAkB,EAAE,sBAAsB,EAAE,MAAM,yBAAyB,CAAC;AACrF,OAAO,EAAE,gBAAgB,EAAE,oBAAoB,EAAE,MAAM,uBAAuB,CAAC;AAC/E,OAAO,EAAE,qBAAqB,EAAE,yBAAyB,EAAE,MAAM,6BAA6B,CAAC;AAC/F,OAAO,EACL,wBAAwB,EACxB,4BAA4B,GAC7B,MAAM,gCAAgC,CAAC;AACxC,OAAO,EACL,kBAAkB,EAClB,sBAAsB,GACvB,MAAM,yBAAyB,CAAC;AACjC,OAAO,EACL,sBAAsB,EACtB,0BAA0B,GAC3B,MAAM,8BAA8B,CAAC;AACtC,OAAO,EACL,6BAA6B,EAC7B,iCAAiC,GAClC,MAAM,qCAAqC,CAAC;AAC7C,OAAO,EACL,sBAAsB,EACtB,0BAA0B,GAC3B,MAAM,6BAA6B,CAAC;AACrC,OAAO,EACL,uBAAuB,EACvB,2BAA2B,GAC5B,MAAM,8BAA8B,CAAC;AACtC,OAAO,EACL,wBAAwB,EACxB,4BAA4B,GAC7B,MAAM,+BAA+B,CAAC;AACvC,OAAO,EACL,yBAAyB,EACzB,6BAA6B,GAC9B,MAAM,gCAAgC,CAAC;AACxC,OAAO,EACL,4BAA4B,EAC5B,gCAAgC,GACjC,MAAM,mCAAmC,CAAC;AAC3C,OAAO,EACL,gCAAgC,EAChC,oCAAoC,GACrC,MAAM,wCAAwC,CAAC;AAChD,OAAO,EACL,oBAAoB,EACpB,wBAAwB,GACzB,MAAM,2BAA2B,CAAC;AACnC,OAAO,EACL,2BAA2B,EAC3B,+BAA+B,GAChC,MAAM,mCAAmC,CAAC;AAC3C,OAAO,EACL,uBAAuB,EACvB,2BAA2B,GAC5B,MAAM,8BAA8B,CAAC;AACtC,OAAO,EACL,sBAAsB,EACtB,0BAA0B,GAC3B,MAAM,6BAA6B,CAAC;AACrC,OAAO,EACL,kBAAkB,EAClB,sBAAsB,GACvB,MAAM,0BAA0B,CAAC;AAClC,OAAO,EACL,mBAAmB,EACnB,uBAAuB,GACxB,MAAM,2BAA2B,CAAC;AACnC,OAAO,EACL,mBAAmB,EACnB,uBAAuB,GACxB,MAAM,2BAA2B,CAAC;AACnC,OAAO,EACL,kBAAkB,EAClB,sBAAsB,GACvB,MAAM,yBAAyB,CAAC;AACjC,OAAO,EACL,kBAAkB,EAClB,sBAAsB,GACvB,MAAM,yBAAyB,CAAC;AACjC,OAAO,EACL,mBAAmB,EACnB,uBAAuB,GACxB,MAAM,0BAA0B,CAAC;AAClC,OAAO,EACL,uBAAuB,EACvB,2BAA2B,GAC5B,MAAM,+BAA+B,CAAC;AACvC,OAAO,EACL,mBAAmB,EACnB,uBAAuB,GACxB,MAAM,0BAA0B,CAAC;AAClC,OAAO,EACL,wBAAwB,EACxB,4BAA4B,GAC7B,MAAM,gCAAgC,CAAC;AACxC,OAAO,EACL,sBAAsB,EACtB,0BAA0B,GAC3B,MAAM,8BAA8B,CAAC;AACtC,OAAO,EACL,oBAAoB,EACpB,wBAAwB,GACzB,MAAM,4BAA4B,CAAC;AACpC,OAAO,EACL,qBAAqB,EACrB,yBAAyB,GAC1B,MAAM,6BAA6B,CAAC;AACrC,OAAO,EACL,0BAA0B,EAC1B,8BAA8B,GAC/B,MAAM,kCAAkC,CAAC;AAC1C,OAAO,EACL,0BAA0B,EAC1B,8BAA8B,GAC/B,MAAM,kCAAkC,CAAC;AAC1C,OAAO,EACL,kBAAkB,EAClB,sBAAsB,GACvB,MAAM,0BAA0B,CAAC;AAClC,OAAO,EACL,iBAAiB,EACjB,qBAAqB,GACtB,MAAM,wBAAwB,CAAC;AAChC,OAAO,EACL,4BAA4B,EAC5B,gCAAgC,GACjC,MAAM,qCAAqC,CAAC;AAE7C,MAAM,KAAK,GAAG;IACZ;QACE,IAAI,EAAE,0BAA0B;QAChC,WAAW,EACT,6EAA6E;YAC7E,mFAAmF;QACrF,MAAM,EAAE,yBAAyB;QACjC,OAAO,EAAE,qBAAqB;KAC/B;IACD;QACE,IAAI,EAAE,0BAA0B;QAChC,WAAW,EACT,qFAAqF;YACrF,2EAA2E;YAC3E,oFAAoF;YACpF,sFAAsF;YACtF,qFAAqF;YACrF,oFAAoF;YACpF,yDAAyD;QAC3D,MAAM,EAAE,yBAAyB;QACjC,OAAO,EAAE,qBAAqB;KAC/B;IACD;QACE,IAAI,EAAE,mBAAmB;QACzB,WAAW,EACT,iFAAiF;YACjF,wBAAwB;QAC1B,MAAM,EAAE,kBAAkB;QAC1B,OAAO,EAAE,cAAc;KACxB;IACD;QACE,IAAI,EAAE,uBAAuB;QAC7B,WAAW,EACT,8EAA8E;YAC9E,+BAA+B;QACjC,MAAM,EAAE,sBAAsB;QAC9B,OAAO,EAAE,kBAAkB;KAC5B;IACD;QACE,IAAI,EAAE,qBAAqB;QAC3B,WAAW,EAAE,8EAA8E;QAC3F,MAAM,EAAE,oBAAoB;QAC5B,OAAO,EAAE,gBAAgB;KAC1B;IACD;QACE,IAAI,EAAE,2BAA2B;QACjC,WAAW,EACT,6EAA6E;YAC7E,qBAAqB;QACvB,MAAM,EAAE,yBAAyB;QACjC,OAAO,EAAE,qBAAqB;KAC/B;IACD;QACE,IAAI,EAAE,8BAA8B;QACpC,WAAW,EACT,kFAAkF;YAClF,wBAAwB;QAC1B,MAAM,EAAE,4BAA4B;QACpC,OAAO,EAAE,wBAAwB;KAClC;IACD;QACE,IAAI,EAAE,uBAAuB;QAC7B,WAAW,EACT,2FAA2F;YAC3F,gGAAgG;YAChG,wFAAwF;YACxF,oDAAoD;QACtD,MAAM,EAAE,sBAAsB;QAC9B,OAAO,EAAE,kBAAkB;KAC5B;IACD;QACE,IAAI,EAAE,4BAA4B;QAClC,WAAW,EACT,qFAAqF;YACrF,uEAAuE;YACvE,yEAAyE;QAC3E,MAAM,EAAE,0BAA0B;QAClC,OAAO,EAAE,sBAAsB;KAChC;IACD;QACE,IAAI,EAAE,mCAAmC;QACzC,WAAW,EACT,mFAAmF;YACnF,sFAAsF;YACtF,yFAAyF;YACzF,iFAAiF;QACnF,MAAM,EAAE,iCAAiC;QACzC,OAAO,EAAE,6BAA6B;KACvC;IACD;QACE,IAAI,EAAE,2BAA2B;QACjC,WAAW,EACT,+EAA+E;YAC/E,2EAA2E;YAC3E,oDAAoD;QACtD,MAAM,EAAE,0BAA0B;QAClC,OAAO,EAAE,sBAAsB;KAChC;IACD;QACE,IAAI,EAAE,4BAA4B;QAClC,WAAW,EACT,kFAAkF;YAClF,kFAAkF;YAClF,uFAAuF;YACvF,6BAA6B;QAC/B,MAAM,EAAE,2BAA2B;QACnC,OAAO,EAAE,uBAAuB;KACjC;IACD;QACE,IAAI,EAAE,6BAA6B;QACnC,WAAW,EACT,mFAAmF;YACnF,wEAAwE;YACxE,+EAA+E;YAC/E,oEAAoE;QACtE,MAAM,EAAE,4BAA4B;QACpC,OAAO,EAAE,wBAAwB;KAClC;IACD;QACE,IAAI,EAAE,8BAA8B;QACpC,WAAW,EACT,kFAAkF;YAClF,oFAAoF;YACpF,gFAAgF;YAChF,+EAA+E;YAC/E,kEAAkE;QACpE,MAAM,EAAE,6BAA6B;QACrC,OAAO,EAAE,yBAAyB;KACnC;IACD;QACE,IAAI,EAAE,sCAAsC;QAC5C,WAAW,EACT,gFAAgF;YAChF,gFAAgF;QAClF,MAAM,EAAE,oCAAoC;QAC5C,OAAO,EAAE,gCAAgC;KAC1C;IACD;QACE,IAAI,EAAE,iCAAiC;QACvC,WAAW,EACT,qFAAqF;YACrF,gDAAgD;QAClD,MAAM,EAAE,gCAAgC;QACxC,OAAO,EAAE,4BAA4B;KACtC;IACD;QACE,IAAI,EAAE,yBAAyB;QAC/B,WAAW,EACT,8EAA8E;YAC9E,+EAA+E;YAC/E,2EAA2E;YAC3E,iFAAiF;YACjF,aAAa;QACf,MAAM,EAAE,wBAAwB;QAChC,OAAO,EAAE,oBAAoB;KAC9B;IACD;QACE,IAAI,EAAE,iCAAiC;QACvC,WAAW,EACT,0EAA0E;YAC1E,mEAAmE;QACrE,MAAM,EAAE,+BAA+B;QACvC,OAAO,EAAE,2BAA2B;KACrC;IACD;QACE,IAAI,EAAE,4BAA4B;QAClC,WAAW,EACT,mFAAmF;YACnF,mFAAmF;YACnF,iCAAiC;QACnC,MAAM,EAAE,2BAA2B;QACnC,OAAO,EAAE,uBAAuB;KACjC;IACD;QACE,IAAI,EAAE,2BAA2B;QACjC,WAAW,EACT,+EAA+E;YAC/E,iFAAiF;YACjF,gCAAgC;QAClC,MAAM,EAAE,0BAA0B;QAClC,OAAO,EAAE,sBAAsB;KAChC;IACD;QACE,IAAI,EAAE,wBAAwB;QAC9B,WAAW,EACT,sFAAsF;YACtF,mDAAmD;QACrD,MAAM,EAAE,sBAAsB;QAC9B,OAAO,EAAE,kBAAkB;KAC5B;IACD;QACE,IAAI,EAAE,yBAAyB;QAC/B,WAAW,EACT,mFAAmF;YACnF,8EAA8E;YAC9E,kDAAkD;QACpD,MAAM,EAAE,uBAAuB;QAC/B,OAAO,EAAE,mBAAmB;KAC7B;IACD;QACE,IAAI,EAAE,yBAAyB;QAC/B,WAAW,EACT,8EAA8E;YAC9E,oFAAoF;YACpF,mCAAmC;QACrC,MAAM,EAAE,uBAAuB;QAC/B,OAAO,EAAE,mBAAmB;KAC7B;IACD;QACE,IAAI,EAAE,uBAAuB;QAC7B,WAAW,EACT,gFAAgF;YAChF,cAAc;QAChB,MAAM,EAAE,sBAAsB;QAC9B,OAAO,EAAE,kBAAkB;KAC5B;IACD;QACE,IAAI,EAAE,uBAAuB;QAC7B,WAAW,EACT,4EAA4E;YAC5E,0EAA0E;YAC1E,wEAAwE;YACxE,2EAA2E;YAC3E,oBAAoB;QACtB,MAAM,EAAE,sBAAsB;QAC9B,OAAO,EAAE,kBAAkB;KAC5B;IACD;QACE,IAAI,EAAE,wBAAwB;QAC9B,WAAW,EACT,6EAA6E;YAC7E,4EAA4E;YAC5E,gFAAgF;YAChF,mBAAmB;QACrB,MAAM,EAAE,uBAAuB;QAC/B,OAAO,EAAE,mBAAmB;KAC7B;IACD;QACE,IAAI,EAAE,6BAA6B;QACnC,WAAW,EACT,8EAA8E;YAC9E,6DAA6D;QAC/D,MAAM,EAAE,2BAA2B;QACnC,OAAO,EAAE,uBAAuB;KACjC;IACD;QACE,IAAI,EAAE,wBAAwB;QAC9B,WAAW,EACT,gFAAgF;YAChF,+EAA+E;YAC/E,sBAAsB;QACxB,MAAM,EAAE,uBAAuB;QAC/B,OAAO,EAAE,mBAAmB;KAC7B;IACD;QACE,IAAI,EAAE,4BAA4B;QAClC,WAAW,EACT,6EAA6E;YAC7E,wEAAwE;YACxE,yEAAyE;YACzE,2EAA2E;YAC3E,4EAA4E;YAC5E,8EAA8E;YAC9E,+EAA+E;YAC/E,gCAAgC;QAClC,MAAM,EAAE,0BAA0B;QAClC,OAAO,EAAE,sBAAsB;KAChC;IACD;QACE,IAAI,EAAE,0BAA0B;QAChC,WAAW,EACT,4EAA4E;YAC5E,uEAAuE;YACvE,yEAAyE;YACzE,sEAAsE;YACtE,yEAAyE;YACzE,iCAAiC;QACnC,MAAM,EAAE,wBAAwB;QAChC,OAAO,EAAE,oBAAoB;KAC9B;IACD;QACE,IAAI,EAAE,8BAA8B;QACpC,WAAW,EACT,+EAA+E;YAC/E,iFAAiF;YACjF,kFAAkF;YAClF,kFAAkF;YAClF,gFAAgF;YAChF,gFAAgF;YAChF,4CAA4C;QAC9C,MAAM,EAAE,4BAA4B;QACpC,OAAO,EAAE,wBAAwB;KAClC;IACD;QACE,IAAI,EAAE,2BAA2B;QACjC,WAAW,EACT,6EAA6E;YAC7E,8EAA8E;YAC9E,gFAAgF;YAChF,kCAAkC;QACpC,MAAM,EAAE,yBAAyB;QACjC,OAAO,EAAE,qBAAqB;KAC/B;IACD;QACE,IAAI,EAAE,gCAAgC;QACtC,WAAW,EACT,iFAAiF;YACjF,+EAA+E;YAC/E,+EAA+E;YAC/E,iFAAiF;YACjF,gFAAgF;QAClF,MAAM,EAAE,8BAA8B;QACtC,OAAO,EAAE,0BAA0B;KACpC;IACD;QACE,IAAI,EAAE,gCAAgC;QACtC,WAAW,EACT,8EAA8E;YAC9E,4EAA4E;YAC5E,8EAA8E;YAC9E,qDAAqD;QACvD,MAAM,EAAE,8BAA8B;QACtC,OAAO,EAAE,0BAA0B;KACpC;IACD;QACE,IAAI,EAAE,sBAAsB;QAC5B,WAAW,EACT,iFAAiF;YACjF,iFAAiF;YACjF,kFAAkF;YAClF,sFAAsF;YACtF,gFAAgF;YAChF,kEAAkE;QACpE,MAAM,EAAE,qBAAqB;QAC7B,OAAO,EAAE,iBAAiB;KAC3B;IACD;QACE,IAAI,EAAE,mCAAmC;QACzC,WAAW,EACT,6EAA6E;YAC7E,qEAAqE;YACrE,qCAAqC;QACvC,MAAM,EAAE,gCAAgC;QACxC,OAAO,EAAE,4BAA4B;KACtC;IACD;QACE,IAAI,EAAE,wBAAwB;QAC9B,WAAW,EACT,6EAA6E;YAC7E,gFAAgF;YAChF,4EAA4E;YAC5E,kFAAkF;QACpF,MAAM,EAAE,sBAAsB;QAC9B,OAAO,EAAE,kBAAkB;KAC5B;CACO,CAAC;AAEX,SAAS,SAAS;IAChB,MAAM,MAAM,GAAG,OAAO,CAAC,GAAG,CAAC,gBAAgB,IAAI,8BAA8B,CAAC;IAC9E,MAAM,MAAM,GAAG,OAAO,CAAC,GAAG,CAAC,gBAAgB,CAAC;IAC5C,IAAI,CAAC,MAAM,EAAE,CAAC;QACZ,MAAM,IAAI,KAAK,CACb,4EAA4E;YAC1E,oEAAoE,CACvE,CAAC;IACJ,CAAC;IACD,OAAO,IAAI,cAAc,CAAC,EAAE,MAAM,EAAE,MAAM,EAAE,CAAC,CAAC;AAChD,CAAC;AAED,KAAK,UAAU,IAAI;IACjB,MAAM,MAAM,GAAG,IAAI,MAAM,CACvB,EAAE,IAAI,EAAE,cAAc,EAAE,OAAO,EAAE,OAAO,EAAE,EAC1C,EAAE,YAAY,EAAE,EAAE,KAAK,EAAE,EAAE,EAAE,EAAE,CAChC,CAAC;IAEF,MAAM,CAAC,iBAAiB,CAAC,sBAAsB,EAAE,KAAK,IAAI,EAAE,CAAC,CAAC;QAC5D,KAAK,EAAE,KAAK,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;YACvB,IAAI,EAAE,CAAC,CAAC,IAAI;YACZ,WAAW,EAAE,CAAC,CAAC,WAAW;YAC1B,WAAW,EAAE,eAAe,CAAC,CAAC,CAAC,MAAM,CAA4B;SAClE,CAAC,CAAC;KACJ,CAAC,CAAC,CAAC;IAEJ,MAAM,CAAC,iBAAiB,CAAC,qBAAqB,EAAE,KAAK,EAAE,GAAG,EAAE,EAAE;QAC5D,MAAM,IAAI,GAAG,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,GAAG,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC;QAC3D,IAAI,CAAC,IAAI,EAAE,CAAC;YACV,OAAO;gBACL,OAAO,EAAE,IAAI;gBACb,OAAO,EAAE,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,iBAAiB,GAAG,CAAC,MAAM,CAAC,IAAI,EAAE,EAAE,CAAC;aACtE,CAAC;QACJ,CAAC;QACD,MAAM,MAAM,GAAG,IAAI,CAAC,MAAM,CAAC,SAAS,CAAC,GAAG,CAAC,MAAM,CAAC,SAAS,IAAI,EAAE,CAAC,CAAC;QACjE,IAAI,CAAC,MAAM,CAAC,OAAO,EAAE,CAAC;YACpB,OAAO;gBACL,OAAO,EAAE,IAAI;gBACb,OAAO,EAAE;oBACP,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,yBAAyB,IAAI,CAAC,IAAI,KAAK,MAAM,CAAC,KAAK,CAAC,OAAO,EAAE,EAAE;iBACtF;aACF,CAAC;QACJ,CAAC;QACD,MAAM,MAAM,GAAG,SAAS,EAAE,CAAC;QAC3B,8DAA8D;QAC9D,OAAO,IAAI,CAAC,OAAO,CAAC,MAAM,EAAE,MAAM,CAAC,IAAW,CAAQ,CAAC;IACzD,CAAC,CAAC,CAAC;IAEH,MAAM,SAAS,GAAG,IAAI,oBAAoB,EAAE,CAAC;IAC7C,MAAM,MAAM,CAAC,OAAO,CAAC,SAAS,CAAC,CAAC;IAChC,sCAAsC;IACtC,OAAO,CAAC,KAAK,CAAC,gBAAgB,OAAO,qBAAqB,CAAC,CAAC;AAC9D,CAAC;AAED,IAAI,EAAE,CAAC,KAAK,CAAC,CAAC,GAAG,EAAE,EAAE;IACnB,sCAAsC;IACtC,OAAO,CAAC,KAAK,CAAC,QAAQ,EAAE,GAAG,CAAC,CAAC;IAC7B,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;AAClB,CAAC,CAAC,CAAC"}
|
|
@@ -0,0 +1,15 @@
|
|
|
1
|
+
import { z } from "zod";
|
|
2
|
+
import { EdgeGateClient } from "../client.js";
|
|
3
|
+
import type { ToolResult } from "./setup_workspace.js";
|
|
4
|
+
export declare const checkLLMCompileStatusInputSchema: z.ZodObject<{
|
|
5
|
+
workspace_id: z.ZodString;
|
|
6
|
+
compile_job_id: z.ZodString;
|
|
7
|
+
}, "strip", z.ZodTypeAny, {
|
|
8
|
+
workspace_id: string;
|
|
9
|
+
compile_job_id: string;
|
|
10
|
+
}, {
|
|
11
|
+
workspace_id: string;
|
|
12
|
+
compile_job_id: string;
|
|
13
|
+
}>;
|
|
14
|
+
export type CheckLLMCompileStatusInput = z.infer<typeof checkLLMCompileStatusInputSchema>;
|
|
15
|
+
export declare function checkLLMCompileStatusHandler(client: EdgeGateClient, input: CheckLLMCompileStatusInput): Promise<ToolResult>;
|
|
@@ -0,0 +1,49 @@
|
|
|
1
|
+
import { z } from "zod";
|
|
2
|
+
import { EdgeGateError } from "../client.js";
|
|
3
|
+
export const checkLLMCompileStatusInputSchema = z.object({
|
|
4
|
+
workspace_id: z.string().uuid(),
|
|
5
|
+
compile_job_id: z.string().uuid(),
|
|
6
|
+
});
|
|
7
|
+
export async function checkLLMCompileStatusHandler(client, input) {
|
|
8
|
+
try {
|
|
9
|
+
const job = await client.getLLMCompileJob(input.workspace_id, input.compile_job_id);
|
|
10
|
+
return { content: [{ type: "text", text: format(job) }] };
|
|
11
|
+
}
|
|
12
|
+
catch (err) {
|
|
13
|
+
if (err instanceof EdgeGateError) {
|
|
14
|
+
return {
|
|
15
|
+
isError: true,
|
|
16
|
+
content: [
|
|
17
|
+
{
|
|
18
|
+
type: "text",
|
|
19
|
+
text: err.status === 404
|
|
20
|
+
? `No LLM compile job ${input.compile_job_id} in workspace ${input.workspace_id}.`
|
|
21
|
+
: `EdgeGate returned ${err.status}: ${err.detail}`,
|
|
22
|
+
},
|
|
23
|
+
],
|
|
24
|
+
};
|
|
25
|
+
}
|
|
26
|
+
throw err;
|
|
27
|
+
}
|
|
28
|
+
}
|
|
29
|
+
function format(job) {
|
|
30
|
+
const lines = [
|
|
31
|
+
`LLM compile job **${job.compile_job_id}**: ${job.status.toUpperCase()}`,
|
|
32
|
+
`- progress: ${job.progress.compile_jobs_done}/${job.progress.compile_jobs_total} compile jobs done`,
|
|
33
|
+
];
|
|
34
|
+
if (job.device_id)
|
|
35
|
+
lines.push(`- device: ${job.device_id}`);
|
|
36
|
+
if (job.target_runtime)
|
|
37
|
+
lines.push(`- target_runtime: ${job.target_runtime}`);
|
|
38
|
+
if (job.status === "completed") {
|
|
39
|
+
lines.push(``, `Composite artifact ready:`, `- composite_artifact_id: ${job.composite_artifact_id}`, ``, `Reference it from a pipeline via:`, ` models: [{ name: "...", artifact_id: "${job.composite_artifact_id}" }]`);
|
|
40
|
+
}
|
|
41
|
+
else if (job.status === "failed") {
|
|
42
|
+
lines.push(``, `Error: ${job.error_detail ?? "unknown"}`);
|
|
43
|
+
}
|
|
44
|
+
else {
|
|
45
|
+
lines.push(``, `Still in flight. Re-poll in ~30s.`);
|
|
46
|
+
}
|
|
47
|
+
return lines.join("\n");
|
|
48
|
+
}
|
|
49
|
+
//# sourceMappingURL=check_llm_compile_status.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"check_llm_compile_status.js","sourceRoot":"","sources":["../../src/tools/check_llm_compile_status.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AACxB,OAAO,EAAkB,aAAa,EAAE,MAAM,cAAc,CAAC;AAI7D,MAAM,CAAC,MAAM,gCAAgC,GAAG,CAAC,CAAC,MAAM,CAAC;IACvD,YAAY,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE;IAC/B,cAAc,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE;CAClC,CAAC,CAAC;AAMH,MAAM,CAAC,KAAK,UAAU,4BAA4B,CAChD,MAAsB,EACtB,KAAiC;IAEjC,IAAI,CAAC;QACH,MAAM,GAAG,GAAG,MAAM,MAAM,CAAC,gBAAgB,CACvC,KAAK,CAAC,YAAY,EAClB,KAAK,CAAC,cAAc,CACrB,CAAC;QACF,OAAO,EAAE,OAAO,EAAE,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,CAAC,GAAG,CAAC,EAAE,CAAC,EAAE,CAAC;IAC5D,CAAC;IAAC,OAAO,GAAG,EAAE,CAAC;QACb,IAAI,GAAG,YAAY,aAAa,EAAE,CAAC;YACjC,OAAO;gBACL,OAAO,EAAE,IAAI;gBACb,OAAO,EAAE;oBACP;wBACE,IAAI,EAAE,MAAM;wBACZ,IAAI,EACF,GAAG,CAAC,MAAM,KAAK,GAAG;4BAChB,CAAC,CAAC,sBAAsB,KAAK,CAAC,cAAc,iBAAiB,KAAK,CAAC,YAAY,GAAG;4BAClF,CAAC,CAAC,qBAAqB,GAAG,CAAC,MAAM,KAAK,GAAG,CAAC,MAAM,EAAE;qBACvD;iBACF;aACF,CAAC;QACJ,CAAC;QACD,MAAM,GAAG,CAAC;IACZ,CAAC;AACH,CAAC;AAED,SAAS,MAAM,CAAC,GAAkB;IAChC,MAAM,KAAK,GAAa;QACtB,qBAAqB,GAAG,CAAC,cAAc,OAAO,GAAG,CAAC,MAAM,CAAC,WAAW,EAAE,EAAE;QACxE,eAAe,GAAG,CAAC,QAAQ,CAAC,iBAAiB,IAAI,GAAG,CAAC,QAAQ,CAAC,kBAAkB,oBAAoB;KACrG,CAAC;IACF,IAAI,GAAG,CAAC,SAAS;QAAE,KAAK,CAAC,IAAI,CAAC,aAAa,GAAG,CAAC,SAAS,EAAE,CAAC,CAAC;IAC5D,IAAI,GAAG,CAAC,cAAc;QAAE,KAAK,CAAC,IAAI,CAAC,qBAAqB,GAAG,CAAC,cAAc,EAAE,CAAC,CAAC;IAE9E,IAAI,GAAG,CAAC,MAAM,KAAK,WAAW,EAAE,CAAC;QAC/B,KAAK,CAAC,IAAI,CACR,EAAE,EACF,2BAA2B,EAC3B,4BAA4B,GAAG,CAAC,qBAAqB,EAAE,EACvD,EAAE,EACF,mCAAmC,EACnC,2CAA2C,GAAG,CAAC,qBAAqB,MAAM,CAC3E,CAAC;IACJ,CAAC;SAAM,IAAI,GAAG,CAAC,MAAM,KAAK,QAAQ,EAAE,CAAC;QACnC,KAAK,CAAC,IAAI,CAAC,EAAE,EAAE,UAAU,GAAG,CAAC,YAAY,IAAI,SAAS,EAAE,CAAC,CAAC;IAC5D,CAAC;SAAM,CAAC;QACN,KAAK,CAAC,IAAI,CAAC,EAAE,EAAE,mCAAmC,CAAC,CAAC;IACtD,CAAC;IAED,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AAC1B,CAAC"}
|
|
@@ -10,35 +10,127 @@ export declare const createPipelineInputSchema: z.ZodObject<{
|
|
|
10
10
|
promptpack_version: z.ZodDefault<z.ZodString>;
|
|
11
11
|
/** Device names to include. E.g. ["Samsung Galaxy S24 (Family)"] */
|
|
12
12
|
devices: z.ZodArray<z.ZodString, "many">;
|
|
13
|
-
gates: z.ZodArray<z.ZodObject<{
|
|
14
|
-
|
|
13
|
+
gates: z.ZodArray<z.ZodUnion<[z.ZodObject<{
|
|
14
|
+
type: z.ZodOptional<z.ZodLiteral<"metric">>;
|
|
15
|
+
metric: z.ZodEnum<["inference_time_ms", "peak_memory_mb", "npu_compute_percent", "gpu_compute_percent", "cpu_compute_percent", "ttft_ms", "tps", "response_rate_tps", "ttft_min_ms", "ttft_max_ms", "compile_time_ms", "compile_peak_memory_mb", "first_load_time_ms", "first_load_peak_memory_mb", "warm_load_time_ms", "npu_layer_count", "gpu_layer_count", "cpu_layer_count", "total_layer_count", "npu_layer_percent", "gpu_layer_percent", "cpu_layer_percent", "inference_time_ms_cv", "peak_memory_mb_cv", "npu_compute_percent_cv", "gpu_compute_percent_cv", "cpu_compute_percent_cv", "compile_time_ms_cv", "compile_peak_memory_mb_cv", "first_load_time_ms_cv", "first_load_peak_memory_mb_cv", "warm_load_time_ms_cv"]>;
|
|
15
16
|
operator: z.ZodEnum<["<=", "<", ">=", ">", "=="]>;
|
|
16
17
|
threshold: z.ZodNumber;
|
|
17
18
|
description: z.ZodOptional<z.ZodString>;
|
|
18
19
|
}, "strip", z.ZodTypeAny, {
|
|
19
|
-
metric: "inference_time_ms" | "peak_memory_mb" | "npu_compute_percent" | "gpu_compute_percent" | "cpu_compute_percent" | "ttft_ms" | "tps";
|
|
20
|
+
metric: "inference_time_ms" | "peak_memory_mb" | "npu_compute_percent" | "gpu_compute_percent" | "cpu_compute_percent" | "ttft_ms" | "tps" | "response_rate_tps" | "ttft_min_ms" | "ttft_max_ms" | "compile_time_ms" | "compile_peak_memory_mb" | "first_load_time_ms" | "first_load_peak_memory_mb" | "warm_load_time_ms" | "npu_layer_count" | "gpu_layer_count" | "cpu_layer_count" | "total_layer_count" | "npu_layer_percent" | "gpu_layer_percent" | "cpu_layer_percent" | "inference_time_ms_cv" | "peak_memory_mb_cv" | "npu_compute_percent_cv" | "gpu_compute_percent_cv" | "cpu_compute_percent_cv" | "compile_time_ms_cv" | "compile_peak_memory_mb_cv" | "first_load_time_ms_cv" | "first_load_peak_memory_mb_cv" | "warm_load_time_ms_cv";
|
|
20
21
|
operator: "<=" | "<" | ">=" | ">" | "==";
|
|
21
22
|
threshold: number;
|
|
23
|
+
type?: "metric" | undefined;
|
|
22
24
|
description?: string | undefined;
|
|
23
25
|
}, {
|
|
24
|
-
metric: "inference_time_ms" | "peak_memory_mb" | "npu_compute_percent" | "gpu_compute_percent" | "cpu_compute_percent" | "ttft_ms" | "tps";
|
|
26
|
+
metric: "inference_time_ms" | "peak_memory_mb" | "npu_compute_percent" | "gpu_compute_percent" | "cpu_compute_percent" | "ttft_ms" | "tps" | "response_rate_tps" | "ttft_min_ms" | "ttft_max_ms" | "compile_time_ms" | "compile_peak_memory_mb" | "first_load_time_ms" | "first_load_peak_memory_mb" | "warm_load_time_ms" | "npu_layer_count" | "gpu_layer_count" | "cpu_layer_count" | "total_layer_count" | "npu_layer_percent" | "gpu_layer_percent" | "cpu_layer_percent" | "inference_time_ms_cv" | "peak_memory_mb_cv" | "npu_compute_percent_cv" | "gpu_compute_percent_cv" | "cpu_compute_percent_cv" | "compile_time_ms_cv" | "compile_peak_memory_mb_cv" | "first_load_time_ms_cv" | "first_load_peak_memory_mb_cv" | "warm_load_time_ms_cv";
|
|
25
27
|
operator: "<=" | "<" | ">=" | ">" | "==";
|
|
26
28
|
threshold: number;
|
|
29
|
+
type?: "metric" | undefined;
|
|
27
30
|
description?: string | undefined;
|
|
28
|
-
}>,
|
|
31
|
+
}>, z.ZodObject<{
|
|
32
|
+
type: z.ZodLiteral<"composition">;
|
|
33
|
+
/** ONNX op type to match (case-insensitive). Use "*" to match any layer type. */
|
|
34
|
+
op_type: z.ZodDefault<z.ZodString>;
|
|
35
|
+
compute_unit: z.ZodEnum<["NPU", "GPU", "CPU"]>;
|
|
36
|
+
operator: z.ZodEnum<["<=", "<", ">=", ">", "=="]>;
|
|
37
|
+
/** Allowed matching-layer count. `0` is the headline "no fallback" rule. */
|
|
38
|
+
threshold: z.ZodNumber;
|
|
39
|
+
description: z.ZodOptional<z.ZodString>;
|
|
40
|
+
}, "strip", z.ZodTypeAny, {
|
|
41
|
+
type: "composition";
|
|
42
|
+
operator: "<=" | "<" | ">=" | ">" | "==";
|
|
43
|
+
threshold: number;
|
|
44
|
+
op_type: string;
|
|
45
|
+
compute_unit: "NPU" | "GPU" | "CPU";
|
|
46
|
+
description?: string | undefined;
|
|
47
|
+
}, {
|
|
48
|
+
type: "composition";
|
|
49
|
+
operator: "<=" | "<" | ">=" | ">" | "==";
|
|
50
|
+
threshold: number;
|
|
51
|
+
compute_unit: "NPU" | "GPU" | "CPU";
|
|
52
|
+
description?: string | undefined;
|
|
53
|
+
op_type?: string | undefined;
|
|
54
|
+
}>]>, "many">;
|
|
29
55
|
/**
|
|
30
56
|
* Optional multi-model matrix. When omitted the pipeline runs in legacy
|
|
31
57
|
* single-model mode (model_artifact_id is supplied per-run instead).
|
|
58
|
+
*
|
|
59
|
+
* Each entry MUST set EITHER `artifact_id` (pre-uploaded model, existing
|
|
60
|
+
* path) OR `llm_compile_source` (Plan B — EdgeGate compiles + links the
|
|
61
|
+
* LLM via Qualcomm AI Hub on first run, then caches the composite). The
|
|
62
|
+
* backend enforces mutual exclusion; we surface a client-side error
|
|
63
|
+
* before the network call when both are set.
|
|
32
64
|
*/
|
|
33
|
-
models: z.ZodOptional<z.ZodArray<z.ZodObject<{
|
|
65
|
+
models: z.ZodOptional<z.ZodArray<z.ZodEffects<z.ZodObject<{
|
|
34
66
|
name: z.ZodString;
|
|
35
|
-
artifact_id: z.ZodString
|
|
67
|
+
artifact_id: z.ZodOptional<z.ZodString>;
|
|
68
|
+
/**
|
|
69
|
+
* If set instead of artifact_id: EdgeGate will compile the LLM
|
|
70
|
+
* via Qualcomm AI Hub on first pipeline run, then cache the
|
|
71
|
+
* resulting composite artifact. The pipeline is created in
|
|
72
|
+
* compile_pending state and the response includes a
|
|
73
|
+
* compile_job_id alongside the pipeline_id.
|
|
74
|
+
*/
|
|
75
|
+
llm_compile_source: z.ZodOptional<z.ZodObject<{
|
|
76
|
+
source_artifact_ids: z.ZodArray<z.ZodString, "many">;
|
|
77
|
+
sequence_lengths: z.ZodDefault<z.ZodArray<z.ZodNumber, "many">>;
|
|
78
|
+
context_lengths: z.ZodDefault<z.ZodArray<z.ZodNumber, "many">>;
|
|
79
|
+
roles: z.ZodOptional<z.ZodArray<z.ZodString, "many">>;
|
|
80
|
+
target_runtime: z.ZodDefault<z.ZodString>;
|
|
81
|
+
}, "strip", z.ZodTypeAny, {
|
|
82
|
+
source_artifact_ids: string[];
|
|
83
|
+
sequence_lengths: number[];
|
|
84
|
+
context_lengths: number[];
|
|
85
|
+
target_runtime: string;
|
|
86
|
+
roles?: string[] | undefined;
|
|
87
|
+
}, {
|
|
88
|
+
source_artifact_ids: string[];
|
|
89
|
+
sequence_lengths?: number[] | undefined;
|
|
90
|
+
context_lengths?: number[] | undefined;
|
|
91
|
+
roles?: string[] | undefined;
|
|
92
|
+
target_runtime?: string | undefined;
|
|
93
|
+
}>>;
|
|
36
94
|
}, "strip", z.ZodTypeAny, {
|
|
37
|
-
artifact_id: string;
|
|
38
95
|
name: string;
|
|
96
|
+
artifact_id?: string | undefined;
|
|
97
|
+
llm_compile_source?: {
|
|
98
|
+
source_artifact_ids: string[];
|
|
99
|
+
sequence_lengths: number[];
|
|
100
|
+
context_lengths: number[];
|
|
101
|
+
target_runtime: string;
|
|
102
|
+
roles?: string[] | undefined;
|
|
103
|
+
} | undefined;
|
|
39
104
|
}, {
|
|
40
|
-
artifact_id: string;
|
|
41
105
|
name: string;
|
|
106
|
+
artifact_id?: string | undefined;
|
|
107
|
+
llm_compile_source?: {
|
|
108
|
+
source_artifact_ids: string[];
|
|
109
|
+
sequence_lengths?: number[] | undefined;
|
|
110
|
+
context_lengths?: number[] | undefined;
|
|
111
|
+
roles?: string[] | undefined;
|
|
112
|
+
target_runtime?: string | undefined;
|
|
113
|
+
} | undefined;
|
|
114
|
+
}>, {
|
|
115
|
+
name: string;
|
|
116
|
+
artifact_id?: string | undefined;
|
|
117
|
+
llm_compile_source?: {
|
|
118
|
+
source_artifact_ids: string[];
|
|
119
|
+
sequence_lengths: number[];
|
|
120
|
+
context_lengths: number[];
|
|
121
|
+
target_runtime: string;
|
|
122
|
+
roles?: string[] | undefined;
|
|
123
|
+
} | undefined;
|
|
124
|
+
}, {
|
|
125
|
+
name: string;
|
|
126
|
+
artifact_id?: string | undefined;
|
|
127
|
+
llm_compile_source?: {
|
|
128
|
+
source_artifact_ids: string[];
|
|
129
|
+
sequence_lengths?: number[] | undefined;
|
|
130
|
+
context_lengths?: number[] | undefined;
|
|
131
|
+
roles?: string[] | undefined;
|
|
132
|
+
target_runtime?: string | undefined;
|
|
133
|
+
} | undefined;
|
|
42
134
|
}>, "many">>;
|
|
43
135
|
/** Measurement repeats per cell (1–5). Defaults to API default (3). */
|
|
44
136
|
repeats: z.ZodOptional<z.ZodNumber>;
|
|
@@ -70,15 +162,30 @@ export declare const createPipelineInputSchema: z.ZodObject<{
|
|
|
70
162
|
promptpack_id: string;
|
|
71
163
|
promptpack_version: string;
|
|
72
164
|
devices: string[];
|
|
73
|
-
gates: {
|
|
74
|
-
metric: "inference_time_ms" | "peak_memory_mb" | "npu_compute_percent" | "gpu_compute_percent" | "cpu_compute_percent" | "ttft_ms" | "tps";
|
|
165
|
+
gates: ({
|
|
166
|
+
metric: "inference_time_ms" | "peak_memory_mb" | "npu_compute_percent" | "gpu_compute_percent" | "cpu_compute_percent" | "ttft_ms" | "tps" | "response_rate_tps" | "ttft_min_ms" | "ttft_max_ms" | "compile_time_ms" | "compile_peak_memory_mb" | "first_load_time_ms" | "first_load_peak_memory_mb" | "warm_load_time_ms" | "npu_layer_count" | "gpu_layer_count" | "cpu_layer_count" | "total_layer_count" | "npu_layer_percent" | "gpu_layer_percent" | "cpu_layer_percent" | "inference_time_ms_cv" | "peak_memory_mb_cv" | "npu_compute_percent_cv" | "gpu_compute_percent_cv" | "cpu_compute_percent_cv" | "compile_time_ms_cv" | "compile_peak_memory_mb_cv" | "first_load_time_ms_cv" | "first_load_peak_memory_mb_cv" | "warm_load_time_ms_cv";
|
|
167
|
+
operator: "<=" | "<" | ">=" | ">" | "==";
|
|
168
|
+
threshold: number;
|
|
169
|
+
type?: "metric" | undefined;
|
|
170
|
+
description?: string | undefined;
|
|
171
|
+
} | {
|
|
172
|
+
type: "composition";
|
|
75
173
|
operator: "<=" | "<" | ">=" | ">" | "==";
|
|
76
174
|
threshold: number;
|
|
175
|
+
op_type: string;
|
|
176
|
+
compute_unit: "NPU" | "GPU" | "CPU";
|
|
77
177
|
description?: string | undefined;
|
|
78
|
-
}[];
|
|
178
|
+
})[];
|
|
79
179
|
models?: {
|
|
80
|
-
artifact_id: string;
|
|
81
180
|
name: string;
|
|
181
|
+
artifact_id?: string | undefined;
|
|
182
|
+
llm_compile_source?: {
|
|
183
|
+
source_artifact_ids: string[];
|
|
184
|
+
sequence_lengths: number[];
|
|
185
|
+
context_lengths: number[];
|
|
186
|
+
target_runtime: string;
|
|
187
|
+
roles?: string[] | undefined;
|
|
188
|
+
} | undefined;
|
|
82
189
|
}[] | undefined;
|
|
83
190
|
repeats?: number | undefined;
|
|
84
191
|
input_specs?: Record<string, {
|
|
@@ -90,16 +197,31 @@ export declare const createPipelineInputSchema: z.ZodObject<{
|
|
|
90
197
|
name: string;
|
|
91
198
|
promptpack_id: string;
|
|
92
199
|
devices: string[];
|
|
93
|
-
gates: {
|
|
94
|
-
metric: "inference_time_ms" | "peak_memory_mb" | "npu_compute_percent" | "gpu_compute_percent" | "cpu_compute_percent" | "ttft_ms" | "tps";
|
|
200
|
+
gates: ({
|
|
201
|
+
metric: "inference_time_ms" | "peak_memory_mb" | "npu_compute_percent" | "gpu_compute_percent" | "cpu_compute_percent" | "ttft_ms" | "tps" | "response_rate_tps" | "ttft_min_ms" | "ttft_max_ms" | "compile_time_ms" | "compile_peak_memory_mb" | "first_load_time_ms" | "first_load_peak_memory_mb" | "warm_load_time_ms" | "npu_layer_count" | "gpu_layer_count" | "cpu_layer_count" | "total_layer_count" | "npu_layer_percent" | "gpu_layer_percent" | "cpu_layer_percent" | "inference_time_ms_cv" | "peak_memory_mb_cv" | "npu_compute_percent_cv" | "gpu_compute_percent_cv" | "cpu_compute_percent_cv" | "compile_time_ms_cv" | "compile_peak_memory_mb_cv" | "first_load_time_ms_cv" | "first_load_peak_memory_mb_cv" | "warm_load_time_ms_cv";
|
|
202
|
+
operator: "<=" | "<" | ">=" | ">" | "==";
|
|
203
|
+
threshold: number;
|
|
204
|
+
type?: "metric" | undefined;
|
|
205
|
+
description?: string | undefined;
|
|
206
|
+
} | {
|
|
207
|
+
type: "composition";
|
|
95
208
|
operator: "<=" | "<" | ">=" | ">" | "==";
|
|
96
209
|
threshold: number;
|
|
210
|
+
compute_unit: "NPU" | "GPU" | "CPU";
|
|
97
211
|
description?: string | undefined;
|
|
98
|
-
|
|
212
|
+
op_type?: string | undefined;
|
|
213
|
+
})[];
|
|
99
214
|
promptpack_version?: string | undefined;
|
|
100
215
|
models?: {
|
|
101
|
-
artifact_id: string;
|
|
102
216
|
name: string;
|
|
217
|
+
artifact_id?: string | undefined;
|
|
218
|
+
llm_compile_source?: {
|
|
219
|
+
source_artifact_ids: string[];
|
|
220
|
+
sequence_lengths?: number[] | undefined;
|
|
221
|
+
context_lengths?: number[] | undefined;
|
|
222
|
+
roles?: string[] | undefined;
|
|
223
|
+
target_runtime?: string | undefined;
|
|
224
|
+
} | undefined;
|
|
103
225
|
}[] | undefined;
|
|
104
226
|
repeats?: number | undefined;
|
|
105
227
|
input_specs?: Record<string, {
|
|
@@ -19,34 +19,137 @@ export const createPipelineInputSchema = z.object({
|
|
|
19
19
|
/** Device names to include. E.g. ["Samsung Galaxy S24 (Family)"] */
|
|
20
20
|
devices: z.array(z.string().min(1)).min(1).max(5),
|
|
21
21
|
gates: z
|
|
22
|
-
.array(z.
|
|
23
|
-
metric
|
|
24
|
-
|
|
25
|
-
|
|
26
|
-
|
|
27
|
-
|
|
28
|
-
|
|
29
|
-
|
|
30
|
-
|
|
31
|
-
|
|
32
|
-
|
|
33
|
-
|
|
34
|
-
|
|
35
|
-
|
|
36
|
-
|
|
37
|
-
|
|
38
|
-
|
|
39
|
-
|
|
40
|
-
|
|
41
|
-
|
|
42
|
-
|
|
22
|
+
.array(z.union([
|
|
23
|
+
// ----- Metric gates (scalar threshold against a normalised metric).
|
|
24
|
+
// Existing shape preserved exactly so older callers don't have to
|
|
25
|
+
// touch their payloads. `type` is optional — when omitted, the
|
|
26
|
+
// backend interprets the gate as a metric gate.
|
|
27
|
+
z.object({
|
|
28
|
+
type: z.literal("metric").optional(),
|
|
29
|
+
metric: z.enum([
|
|
30
|
+
// Stable — populated for every model AI Hub profiles successfully.
|
|
31
|
+
"inference_time_ms",
|
|
32
|
+
"peak_memory_mb",
|
|
33
|
+
// Compute-unit breakdown — fires when AI Hub returns the per-IP
|
|
34
|
+
// split. npu_compute_percent is the killer gate for "did silent
|
|
35
|
+
// fallback to GPU/CPU happen?" — that's the EdgeGate value prop.
|
|
36
|
+
"npu_compute_percent",
|
|
37
|
+
"gpu_compute_percent",
|
|
38
|
+
"cpu_compute_percent",
|
|
39
|
+
// LLM-only — resolve to "Metric not available" on vision/audio
|
|
40
|
+
// models. Backend names them `ttft_ms` and `tps` (NOT the legacy
|
|
41
|
+
// `throughput_tps` we used to ship here).
|
|
42
|
+
"ttft_ms",
|
|
43
|
+
"tps",
|
|
44
|
+
// Plan B additions (2026-06-15) — surfaced from `llm_metrics`
|
|
45
|
+
// block in the worker normalizer. response_rate_tps is the
|
|
46
|
+
// sustained token-rate (1000/token-inference). ttft_min_ms /
|
|
47
|
+
// ttft_max_ms are the per-prompt-component spread.
|
|
48
|
+
"response_rate_tps",
|
|
49
|
+
"ttft_min_ms",
|
|
50
|
+
"ttft_max_ms",
|
|
51
|
+
// Phase 1 (2026-06-11) — schema-verified AI Hub fields previously
|
|
52
|
+
// dropped. Compile-time + cold/warm load metrics. See
|
|
53
|
+
// docs/superpowers/plans/2026-06-11-gate-expansion.md.
|
|
54
|
+
"compile_time_ms",
|
|
55
|
+
"compile_peak_memory_mb",
|
|
56
|
+
"first_load_time_ms",
|
|
57
|
+
"first_load_peak_memory_mb",
|
|
58
|
+
"warm_load_time_ms",
|
|
59
|
+
// Phase 2a — layer-count composition. Gate on
|
|
60
|
+
// `cpu_layer_count == 0` to catch silent NPU-fallback that
|
|
61
|
+
// aggregate `npu_compute_percent` can miss.
|
|
62
|
+
"npu_layer_count",
|
|
63
|
+
"gpu_layer_count",
|
|
64
|
+
"cpu_layer_count",
|
|
65
|
+
"total_layer_count",
|
|
66
|
+
"npu_layer_percent",
|
|
67
|
+
"gpu_layer_percent",
|
|
68
|
+
"cpu_layer_percent",
|
|
69
|
+
// Phase 2b — CV (coefficient of variation) variants. Computed by
|
|
70
|
+
// backend aggregator as `{source_metric}_cv`, preserving the unit
|
|
71
|
+
// suffix. Gate on `inference_time_ms_cv < 0.05` for stricter
|
|
72
|
+
// determinism than our default 10% flake threshold.
|
|
73
|
+
"inference_time_ms_cv",
|
|
74
|
+
"peak_memory_mb_cv",
|
|
75
|
+
"npu_compute_percent_cv",
|
|
76
|
+
"gpu_compute_percent_cv",
|
|
77
|
+
"cpu_compute_percent_cv",
|
|
78
|
+
"compile_time_ms_cv",
|
|
79
|
+
"compile_peak_memory_mb_cv",
|
|
80
|
+
"first_load_time_ms_cv",
|
|
81
|
+
"first_load_peak_memory_mb_cv",
|
|
82
|
+
"warm_load_time_ms_cv",
|
|
83
|
+
]),
|
|
84
|
+
operator: z.enum(["<=", "<", ">=", ">", "=="]),
|
|
85
|
+
threshold: z.number().positive(),
|
|
86
|
+
description: z.string().optional(),
|
|
87
|
+
}),
|
|
88
|
+
// ----- Composition gates (Phase 2a). Count per-layer
|
|
89
|
+
// (op_type, compute_unit) matches in AI Hub's execution_detail
|
|
90
|
+
// and gate the count. The killer use case: catch silent NPU→CPU
|
|
91
|
+
// fallback that aggregate `npu_compute_percent` can mask when
|
|
92
|
+
// one heavy op dominates time-weighted percentages.
|
|
93
|
+
//
|
|
94
|
+
// Example (no Conv layer allowed on CPU):
|
|
95
|
+
// { type: "composition", op_type: "Conv", compute_unit: "CPU",
|
|
96
|
+
// operator: "==", threshold: 0 }
|
|
97
|
+
//
|
|
98
|
+
// Example (no layer of ANY type on CPU — i.e. "no fallback"):
|
|
99
|
+
// { type: "composition", op_type: "*", compute_unit: "CPU",
|
|
100
|
+
// operator: "==", threshold: 0 }
|
|
101
|
+
z.object({
|
|
102
|
+
type: z.literal("composition"),
|
|
103
|
+
/** ONNX op type to match (case-insensitive). Use "*" to match any layer type. */
|
|
104
|
+
op_type: z.string().min(1).default("*"),
|
|
105
|
+
compute_unit: z.enum(["NPU", "GPU", "CPU"]),
|
|
106
|
+
operator: z.enum(["<=", "<", ">=", ">", "=="]),
|
|
107
|
+
/** Allowed matching-layer count. `0` is the headline "no fallback" rule. */
|
|
108
|
+
threshold: z.number().int().min(0),
|
|
109
|
+
description: z.string().optional(),
|
|
110
|
+
}),
|
|
111
|
+
]))
|
|
43
112
|
.min(1),
|
|
44
113
|
/**
|
|
45
114
|
* Optional multi-model matrix. When omitted the pipeline runs in legacy
|
|
46
115
|
* single-model mode (model_artifact_id is supplied per-run instead).
|
|
116
|
+
*
|
|
117
|
+
* Each entry MUST set EITHER `artifact_id` (pre-uploaded model, existing
|
|
118
|
+
* path) OR `llm_compile_source` (Plan B — EdgeGate compiles + links the
|
|
119
|
+
* LLM via Qualcomm AI Hub on first run, then caches the composite). The
|
|
120
|
+
* backend enforces mutual exclusion; we surface a client-side error
|
|
121
|
+
* before the network call when both are set.
|
|
47
122
|
*/
|
|
48
123
|
models: z
|
|
49
|
-
.array(z
|
|
124
|
+
.array(z
|
|
125
|
+
.object({
|
|
126
|
+
name: z.string().min(1),
|
|
127
|
+
artifact_id: z.string().min(1).optional(),
|
|
128
|
+
/**
|
|
129
|
+
* If set instead of artifact_id: EdgeGate will compile the LLM
|
|
130
|
+
* via Qualcomm AI Hub on first pipeline run, then cache the
|
|
131
|
+
* resulting composite artifact. The pipeline is created in
|
|
132
|
+
* compile_pending state and the response includes a
|
|
133
|
+
* compile_job_id alongside the pipeline_id.
|
|
134
|
+
*/
|
|
135
|
+
llm_compile_source: z
|
|
136
|
+
.object({
|
|
137
|
+
source_artifact_ids: z.array(z.string().uuid()).min(1).max(8),
|
|
138
|
+
sequence_lengths: z
|
|
139
|
+
.array(z.number().int().positive())
|
|
140
|
+
.default([128, 1]),
|
|
141
|
+
context_lengths: z
|
|
142
|
+
.array(z.number().int().positive())
|
|
143
|
+
.default([4096]),
|
|
144
|
+
roles: z.array(z.string().min(1)).optional(),
|
|
145
|
+
target_runtime: z.string().min(1).default("genie"),
|
|
146
|
+
})
|
|
147
|
+
.optional(),
|
|
148
|
+
})
|
|
149
|
+
.refine((m) => (m.artifact_id !== undefined) !== (m.llm_compile_source !== undefined), {
|
|
150
|
+
message: "Each model must set EXACTLY ONE of `artifact_id` (pre-uploaded model) " +
|
|
151
|
+
"OR `llm_compile_source` (auto-compile via AI Hub). Both set OR neither set is invalid.",
|
|
152
|
+
}))
|
|
50
153
|
.min(1)
|
|
51
154
|
.max(10)
|
|
52
155
|
.optional(),
|
|
@@ -106,13 +209,36 @@ export async function createPipelineHandler(client, input) {
|
|
|
106
209
|
promptpack_id: input.promptpack_id,
|
|
107
210
|
version: input.promptpack_version,
|
|
108
211
|
};
|
|
109
|
-
const gates = input.gates.map((g) =>
|
|
110
|
-
|
|
111
|
-
|
|
112
|
-
|
|
113
|
-
|
|
114
|
-
|
|
115
|
-
|
|
212
|
+
const gates = input.gates.map((g) => {
|
|
213
|
+
const baseOperator = OPERATOR_MAP[g.operator] ?? g.operator;
|
|
214
|
+
if (g.type === "composition") {
|
|
215
|
+
return {
|
|
216
|
+
type: "composition",
|
|
217
|
+
op_type: g.op_type ?? "*",
|
|
218
|
+
compute_unit: g.compute_unit,
|
|
219
|
+
operator: baseOperator,
|
|
220
|
+
threshold: g.threshold,
|
|
221
|
+
...(g.description !== undefined ? { description: g.description } : {}),
|
|
222
|
+
};
|
|
223
|
+
}
|
|
224
|
+
return {
|
|
225
|
+
// type defaults to "metric" on the backend when absent — keep payload
|
|
226
|
+
// minimal for the common case.
|
|
227
|
+
metric: g.metric,
|
|
228
|
+
operator: baseOperator,
|
|
229
|
+
threshold: g.threshold,
|
|
230
|
+
...(g.description !== undefined ? { description: g.description } : {}),
|
|
231
|
+
};
|
|
232
|
+
});
|
|
233
|
+
const model_matrix = input.models?.map((m) => {
|
|
234
|
+
if (m.llm_compile_source !== undefined) {
|
|
235
|
+
return {
|
|
236
|
+
label: m.name,
|
|
237
|
+
llm_compile_source: m.llm_compile_source,
|
|
238
|
+
};
|
|
239
|
+
}
|
|
240
|
+
return { artifact_id: m.artifact_id, label: m.name };
|
|
241
|
+
});
|
|
116
242
|
const run_policy = input.repeats !== undefined ? { measurement_repeats: input.repeats } : undefined;
|
|
117
243
|
try {
|
|
118
244
|
const pipeline = await client.createPipeline(input.workspace_id, {
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"create_pipeline.js","sourceRoot":"","sources":["../../src/tools/create_pipeline.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AACxB,OAAO,EAAkB,aAAa,EAAE,MAAM,cAAc,CAAC;AAG7D,MAAM,SAAS,GAAG,EAAE,CAAC;AAErB,4DAA4D;AAC5D,MAAM,YAAY,GAA2B;IAC3C,IAAI,EAAE,KAAK;IACX,GAAG,EAAE,IAAI;IACT,IAAI,EAAE,KAAK;IACX,GAAG,EAAE,IAAI;IACT,IAAI,EAAE,IAAI;CACX,CAAC;AAEF,MAAM,CAAC,MAAM,yBAAyB,GAAG,CAAC,CAAC,MAAM,CAAC;IAChD,YAAY,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE;IAC/B,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,GAAG,CAAC;IAChC,0GAA0G;IAC1G,aAAa,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;IAChC,8CAA8C;IAC9C,kBAAkB,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,OAAO,CAAC;IACtD,oEAAoE;IACpE,OAAO,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;IACjD,KAAK,EAAE,CAAC;SACL,KAAK,CACJ,CAAC,CAAC,MAAM,CAAC;
|
|
1
|
+
{"version":3,"file":"create_pipeline.js","sourceRoot":"","sources":["../../src/tools/create_pipeline.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AACxB,OAAO,EAAkB,aAAa,EAAE,MAAM,cAAc,CAAC;AAG7D,MAAM,SAAS,GAAG,EAAE,CAAC;AAErB,4DAA4D;AAC5D,MAAM,YAAY,GAA2B;IAC3C,IAAI,EAAE,KAAK;IACX,GAAG,EAAE,IAAI;IACT,IAAI,EAAE,KAAK;IACX,GAAG,EAAE,IAAI;IACT,IAAI,EAAE,IAAI;CACX,CAAC;AAEF,MAAM,CAAC,MAAM,yBAAyB,GAAG,CAAC,CAAC,MAAM,CAAC;IAChD,YAAY,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE;IAC/B,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,GAAG,CAAC;IAChC,0GAA0G;IAC1G,aAAa,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;IAChC,8CAA8C;IAC9C,kBAAkB,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,OAAO,CAAC;IACtD,oEAAoE;IACpE,OAAO,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;IACjD,KAAK,EAAE,CAAC;SACL,KAAK,CACJ,CAAC,CAAC,KAAK,CAAC;QACN,qEAAqE;QACrE,kEAAkE;QAClE,+DAA+D;QAC/D,gDAAgD;QAChD,CAAC,CAAC,MAAM,CAAC;YACP,IAAI,EAAE,CAAC,CAAC,OAAO,CAAC,QAAQ,CAAC,CAAC,QAAQ,EAAE;YACtC,MAAM,EAAE,CAAC,CAAC,IAAI,CAAC;gBACb,mEAAmE;gBACnE,mBAAmB;gBACnB,gBAAgB;gBAChB,gEAAgE;gBAChE,gEAAgE;gBAChE,iEAAiE;gBACjE,qBAAqB;gBACrB,qBAAqB;gBACrB,qBAAqB;gBACrB,+DAA+D;gBAC/D,iEAAiE;gBACjE,0CAA0C;gBAC1C,SAAS;gBACT,KAAK;gBACL,8DAA8D;gBAC9D,2DAA2D;gBAC3D,6DAA6D;gBAC7D,mDAAmD;gBACnD,mBAAmB;gBACnB,aAAa;gBACb,aAAa;gBACb,kEAAkE;gBAClE,sDAAsD;gBACtD,uDAAuD;gBACvD,iBAAiB;gBACjB,wBAAwB;gBACxB,oBAAoB;gBACpB,2BAA2B;gBAC3B,mBAAmB;gBACnB,8CAA8C;gBAC9C,2DAA2D;gBAC3D,4CAA4C;gBAC5C,iBAAiB;gBACjB,iBAAiB;gBACjB,iBAAiB;gBACjB,mBAAmB;gBACnB,mBAAmB;gBACnB,mBAAmB;gBACnB,mBAAmB;gBACnB,iEAAiE;gBACjE,kEAAkE;gBAClE,6DAA6D;gBAC7D,oDAAoD;gBACpD,sBAAsB;gBACtB,mBAAmB;gBACnB,wBAAwB;gBACxB,wBAAwB;gBACxB,wBAAwB;gBACxB,oBAAoB;gBACpB,2BAA2B;gBAC3B,uBAAuB;gBACvB,8BAA8B;gBAC9B,sBAAsB;aACvB,CAAC;YACF,QAAQ,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,GAAG,EAAE,IAAI,EAAE,GAAG,EAAE,IAAI,CAAC,CAAC;YAC9C,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;YAChC,WAAW,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;SACjC,CAAC;QACF,sDAAsD;QACtD,+DAA+D;QAC/D,gEAAgE;QAChE,8DAA8D;QAC9D,oDAAoD;QACpD,EAAE;QACF,0CAA0C;QAC1C,iEAAiE;QACjE,qCAAqC;QACrC,EAAE;QACF,8DAA8D;QAC9D,8DAA8D;QAC9D,qCAAqC;QACrC,CAAC,CAAC,MAAM,CAAC;YACP,IAAI,EAAE,CAAC,CAAC,OAAO,CAAC,aAAa,CAAC;YAC9B,iFAAiF;YACjF,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,GAAG,CAAC;YACvC,YAAY,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,KAAK,EAAE,KAAK,CAAC,CAAC;YAC3C,QAAQ,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,GAAG,EAAE,IAAI,EAAE,GAAG,EAAE,IAAI,CAAC,CAAC;YAC9C,4EAA4E;YAC5E,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;YAClC,WAAW,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;SACnC,CAAC;KACH,CAAC,CACH;SACA,GAAG,CAAC,CAAC,CAAC;IACT;;;;;;;;;OASG;IACH,MAAM,EAAE,CAAC;SACN,KAAK,CACJ,CAAC;SACE,MAAM,CAAC;QACN,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;QACvB,WAAW,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,QAAQ,EAAE;QACzC;;;;;;WAMG;QACH,kBAAkB,EAAE,CAAC;aAClB,MAAM,CAAC;YACN,mBAAmB,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;YAC7D,gBAAgB,EAAE,CAAC;iBAChB,KAAK,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE,CAAC,QAAQ,EAAE,CAAC;iBAClC,OAAO,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC,CAAC;YACpB,eAAe,EAAE,CAAC;iBACf,KAAK,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE,CAAC,QAAQ,EAAE,CAAC;iBAClC,OAAO,CAAC,CAAC,IAAI,CAAC,CAAC;YAClB,KAAK,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,QAAQ,EAAE;YAC5C,cAAc,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,OAAO,CAAC;SACnD,CAAC;aACD,QAAQ,EAAE;KACd,CAAC;SACD,MAAM,CACL,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,WAAW,KAAK,SAAS,CAAC,KAAK,CAAC,CAAC,CAAC,kBAAkB,KAAK,SAAS,CAAC,EAC7E;QACE,OAAO,EACL,wEAAwE;YACxE,wFAAwF;KAC3F,CACF,CACJ;SACA,GAAG,CAAC,CAAC,CAAC;SACN,GAAG,CAAC,EAAE,CAAC;SACP,QAAQ,EAAE;IACb,uEAAuE;IACvE,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,QAAQ,EAAE;IAClD;;;;;;;;;;;OAWG;IACH,WAAW,EAAE,CAAC;SACX,MAAM,CACL,CAAC,CAAC,MAAM,EAAE,EACV,CAAC,CAAC,MAAM,CAAC;QACP,KAAK,EAAE,CAAC;aACL,KAAK,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE,CAAC,QAAQ,EAAE,CAAC;aAClC,GAAG,CAAC,CAAC,CAAC;aACN,GAAG,CAAC,CAAC,CAAC;aACN,QAAQ,CAAC,sCAAsC,CAAC;QACnD,KAAK,EAAE,CAAC;aACL,IAAI,CAAC,CAAC,SAAS,EAAE,SAAS,EAAE,OAAO,EAAE,OAAO,EAAE,MAAM,CAAC,CAAC;aACtD,OAAO,CAAC,SAAS,CAAC;aAClB,QAAQ,CAAC,eAAe,CAAC;KAC7B,CAAC,CACH;SACA,QAAQ,EAAE;SACV,QAAQ,CACP,0DAA0D;QACxD,mFAAmF;QACnF,uDAAuD;QACvD,6EAA6E;QAC7E,uDAAuD,CAC1D;CACJ,CAAC,CAAC;AAIH,MAAM,CAAC,KAAK,UAAU,qBAAqB,CACzC,MAAsB,EACtB,KAA0B;IAE1B,gEAAgE;IAChE,MAAM,UAAU,GAAG,KAAK,CAAC,MAAM,EAAE,MAAM,IAAI,CAAC,CAAC;IAC7C,MAAM,SAAS,GAAG,UAAU,GAAG,KAAK,CAAC,OAAO,CAAC,MAAM,CAAC;IACpD,IAAI,KAAK,CAAC,MAAM,IAAI,SAAS,GAAG,SAAS,EAAE,CAAC;QAC1C,OAAO;YACL,OAAO,EAAE,IAAI;YACb,OAAO,EAAE;gBACP;oBACE,IAAI,EAAE,MAAM;oBACZ,IAAI,EACF,GAAG,KAAK,CAAC,MAAM,CAAC,MAAM,aAAa,KAAK,CAAC,OAAO,CAAC,MAAM,cAAc,SAAS,UAAU;wBACxF,sCAAsC,SAAS,+BAA+B;wBAC9E,0BAA0B;iBAC7B;aACF;SACF,CAAC;IACJ,CAAC;IAED,0CAA0C;IAC1C,MAAM,aAAa,GAAG,KAAK,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,CAAC,EAAE,IAAI,EAAE,OAAO,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC;IAC7E,MAAM,cAAc,GAAG;QACrB,aAAa,EAAE,KAAK,CAAC,aAAa;QAClC,OAAO,EAAE,KAAK,CAAC,kBAAkB;KAClC,CAAC;IACF,MAAM,KAAK,GAAG,KAAK,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE;QAClC,MAAM,YAAY,GAAG,YAAY,CAAC,CAAC,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC,QAAQ,CAAC;QAC5D,IAAI,CAAC,CAAC,IAAI,KAAK,aAAa,EAAE,CAAC;YAC7B,OAAO;gBACL,IAAI,EAAE,aAAsB;gBAC5B,OAAO,EAAE,CAAC,CAAC,OAAO,IAAI,GAAG;gBACzB,YAAY,EAAE,CAAC,CAAC,YAAY;gBAC5B,QAAQ,EAAE,YAAY;gBACtB,SAAS,EAAE,CAAC,CAAC,SAAS;gBACtB,GAAG,CAAC,CAAC,CAAC,WAAW,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,WAAW,EAAE,CAAC,CAAC,WAAW,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;aACvE,CAAC;QACJ,CAAC;QACD,OAAO;YACL,sEAAsE;YACtE,+BAA+B;YAC/B,MAAM,EAAE,CAAC,CAAC,MAAM;YAChB,QAAQ,EAAE,YAAY;YACtB,SAAS,EAAE,CAAC,CAAC,SAAS;YACtB,GAAG,CAAC,CAAC,CAAC,WAAW,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,WAAW,EAAE,CAAC,CAAC,WAAW,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;SACvE,CAAC;IACJ,CAAC,CAAC,CAAC;IACH,MAAM,YAAY,GAAG,KAAK,CAAC,MAAM,EAAE,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE;QAC3C,IAAI,CAAC,CAAC,kBAAkB,KAAK,SAAS,EAAE,CAAC;YACvC,OAAO;gBACL,KAAK,EAAE,CAAC,CAAC,IAAI;gBACb,kBAAkB,EAAE,CAAC,CAAC,kBAAkB;aACzC,CAAC;QACJ,CAAC;QACD,OAAO,EAAE,WAAW,EAAE,CAAC,CAAC,WAAqB,EAAE,KAAK,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC;IACjE,CAAC,CAAC,CAAC;IACH,MAAM,UAAU,GAAG,KAAK,CAAC,OAAO,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,mBAAmB,EAAE,KAAK,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC,SAAS,CAAC;IAEpG,IAAI,CAAC;QACH,MAAM,QAAQ,GAAG,MAAM,MAAM,CAAC,cAAc,CAAC,KAAK,CAAC,YAAY,EAAE;YAC/D,IAAI,EAAE,KAAK,CAAC,IAAI;YAChB,aAAa;YACb,cAAc;YACd,KAAK;YACL,GAAG,CAAC,YAAY,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,YAAY,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;YACvD,GAAG,CAAC,UAAU,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,UAAU,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;YACnD,GAAG,CAAC,KAAK,CAAC,WAAW,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,WAAW,EAAE,KAAK,CAAC,WAAW,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;SAC/E,CAAC,CAAC;QACH,OAAO;YACL,OAAO,EAAE;gBACP;oBACE,IAAI,EAAE,MAAM;oBACZ,IAAI,EAAE;wBACJ,sBAAsB,QAAQ,CAAC,IAAI,UAAU,QAAQ,CAAC,EAAE,IAAI;wBAC5D,EAAE;wBACF,KAAK,QAAQ,CAAC,YAAY,gBAAgB,IAAI,CAAC,GAAG,CAAC,QAAQ,CAAC,WAAW,EAAE,CAAC,CAAC,eAAe,QAAQ,CAAC,UAAU,kBAAkB;wBAC/H,KAAK,QAAQ,CAAC,UAAU,UAAU;wBAClC,iBAAiB,QAAQ,CAAC,aAAa,IAAI,QAAQ,CAAC,kBAAkB,EAAE;wBACxE,EAAE;wBACF,6FAA6F;qBAC9F,CAAC,IAAI,CAAC,IAAI,CAAC;iBACb;aACF;SACF,CAAC;IACJ,CAAC;IAAC,OAAO,GAAG,EAAE,CAAC;QACb,IAAI,GAAG,YAAY,aAAa,EAAE,CAAC;YACjC,OAAO;gBACL,OAAO,EAAE,IAAI;gBACb,OAAO,EAAE,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,8BAA8B,GAAG,CAAC,MAAM,EAAE,EAAE,CAAC;aAC9E,CAAC;QACJ,CAAC;QACD,MAAM,GAAG,CAAC;IACZ,CAAC;AACH,CAAC"}
|
|
@@ -97,22 +97,30 @@ export async function importHuggingfaceModelHandler(client, input) {
|
|
|
97
97
|
const sizeLabel = latest.size_bytes !== null
|
|
98
98
|
? `${(latest.size_bytes / 1_048_576).toFixed(1)} MB`
|
|
99
99
|
: "unknown size";
|
|
100
|
+
// Plan B / F-T6.2 — multi-part .bin imports surface every component.
|
|
101
|
+
// The legacy `artifact_id` still points at part 1 for backward compat;
|
|
102
|
+
// `artifacts` (when length > 1) is the full set.
|
|
103
|
+
const isMultiPart = Array.isArray(latest.artifacts) && latest.artifacts.length > 1;
|
|
104
|
+
const lines = [
|
|
105
|
+
`Import complete: **${input.hf_repo_id}**`,
|
|
106
|
+
``,
|
|
107
|
+
];
|
|
108
|
+
if (isMultiPart) {
|
|
109
|
+
const parts = latest.artifacts;
|
|
110
|
+
lines.push(`Imported ${parts.length} component artifacts (multi-part LLM):`, ``);
|
|
111
|
+
for (const part of parts) {
|
|
112
|
+
const partSize = part.size_bytes > 0
|
|
113
|
+
? `${(part.size_bytes / 1_048_576).toFixed(1)} MB`
|
|
114
|
+
: "unknown size";
|
|
115
|
+
lines.push(`- [${part.role}] ${part.filename} (${partSize}) — artifact_id: ${part.id}`);
|
|
116
|
+
}
|
|
117
|
+
lines.push(``, `Wire these into a pipeline via Plan B's llm_compile_source on a model entry:`, ` models: [{`, ` name: "${input.hf_repo_id}",`, ` llm_compile_source: {`, ` source_artifact_ids: [${parts.map((p) => `"${p.id}"`).join(", ")}],`, ` roles: [${parts.map((p) => `"${p.role}"`).join(", ")}],`, ` }`, ` }]`, ``, `Or compile them up-front with \`edgegate_llm_compile\` and then reference the resulting composite_artifact_id.`);
|
|
118
|
+
}
|
|
119
|
+
else {
|
|
120
|
+
lines.push(`- artifact_id: ${latest.artifact_id}`, `- filename: ${latest.filename ?? "(unknown)"}`, `- size: ${sizeLabel}`, ``, `Use this artifact_id in \`edgegate_create_pipeline\` to gate this model:`, ` models: [{ name: "${input.hf_repo_id}", artifact_id: "${latest.artifact_id}" }]`);
|
|
121
|
+
}
|
|
100
122
|
return {
|
|
101
|
-
content: [
|
|
102
|
-
{
|
|
103
|
-
type: "text",
|
|
104
|
-
text: [
|
|
105
|
-
`Import complete: **${input.hf_repo_id}**`,
|
|
106
|
-
``,
|
|
107
|
-
`- artifact_id: ${latest.artifact_id}`,
|
|
108
|
-
`- filename: ${latest.filename ?? "(unknown)"}`,
|
|
109
|
-
`- size: ${sizeLabel}`,
|
|
110
|
-
``,
|
|
111
|
-
`Use this artifact_id in \`edgegate_create_pipeline\` to gate this model:`,
|
|
112
|
-
` models: [{ name: "${input.hf_repo_id}", artifact_id: "${latest.artifact_id}" }]`,
|
|
113
|
-
].join("\n"),
|
|
114
|
-
},
|
|
115
|
-
],
|
|
123
|
+
content: [{ type: "text", text: lines.join("\n") }],
|
|
116
124
|
};
|
|
117
125
|
}
|
|
118
126
|
catch (err) {
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"import_huggingface_model.js","sourceRoot":"","sources":["../../src/tools/import_huggingface_model.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AACxB,OAAO,EAAkB,aAAa,EAAE,MAAM,cAAc,CAAC;AAI7D,MAAM,CAAC,MAAM,iCAAiC,GAAG,CAAC,CAAC,MAAM,CAAC;IACxD,YAAY,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE;IAC/B,UAAU,EAAE,CAAC;SACV,MAAM,EAAE;SACR,KAAK,CAAC,gBAAgB,EAAE,2EAA2E,CAAC;IACvG,QAAQ,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,OAAO,CAAC,MAAM,CAAC;IAC/C,QAAQ,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;IAC/B,mBAAmB,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,QAAQ,EAAE,CAAC,OAAO,CAAC,IAAI,CAAC;IACzD,gBAAgB,EAAE,CAAC;SAChB,MAAM,EAAE;SACR,GAAG,EAAE;SACL,GAAG,CAAC,CAAC,CAAC;SACN,GAAG,CAAC,GAAG,CAAC;SACR,QAAQ,EAAE;SACV,OAAO,CAAC,GAAG,CAAC;CAChB,CAAC,CAAC;AAIH,MAAM,gBAAgB,GAAG,KAAK,CAAC;AAE/B,MAAM,CAAC,KAAK,UAAU,6BAA6B,CACjD,MAAsB,EACtB,KAAkC;IAElC,IAAI,CAAC;QACH,MAAM,IAAI,GAAiE;YACzE,UAAU,EAAE,KAAK,CAAC,UAAU;YAC5B,QAAQ,EAAE,KAAK,CAAC,QAAQ;SACzB,CAAC;QACF,IAAI,KAAK,CAAC,QAAQ,KAAK,SAAS,EAAE,CAAC;YACjC,IAAI,CAAC,QAAQ,GAAG,KAAK,CAAC,QAAQ,CAAC;QACjC,CAAC;QAED,MAAM,GAAG,GAAG,MAAM,MAAM,CAAC,sBAAsB,CAAC,KAAK,CAAC,YAAY,EAAE,IAAI,CAAC,CAAC;QAE1E,IAAI,CAAC,KAAK,CAAC,mBAAmB,EAAE,CAAC;YAC/B,OAAO;gBACL,OAAO,EAAE;oBACP;wBACE,IAAI,EAAE,MAAM;wBACZ,IAAI,EAAE;4BACJ,wBAAwB,KAAK,CAAC,UAAU,KAAK;4BAC7C,EAAE;4BACF,oBAAoB,GAAG,CAAC,aAAa,EAAE;4BACvC,aAAa,GAAG,CAAC,MAAM,EAAE;4BACzB,EAAE;4BACF,0BAA0B;4BAC1B,0DAA0D,KAAK,CAAC,YAAY,mBAAmB,KAAK,CAAC,UAAU,mCAAmC;4BAClJ,EAAE;4BACF,sFAAsF;yBACvF,CAAC,IAAI,CAAC,IAAI,CAAC;qBACb;iBACF;aACF,CAAC;QACJ,CAAC;QAED,qCAAqC;QACrC,MAAM,UAAU,GAAG,IAAI,CAAC,GAAG,EAAE,GAAG,KAAK,CAAC,gBAAgB,GAAG,KAAK,CAAC;QAC/D,IAAI,MAAM,GAAgB,GAAG,CAAC;QAE9B,OAAO,MAAM,CAAC,MAAM,KAAK,MAAM,IAAI,MAAM,CAAC,MAAM,KAAK,QAAQ,EAAE,CAAC;YAC9D,IAAI,IAAI,CAAC,GAAG,EAAE,IAAI,UAAU,EAAE,CAAC;gBAC7B,OAAO;oBACL,OAAO,EAAE;wBACP;4BACE,IAAI,EAAE,MAAM;4BACZ,IAAI,EAAE;gCACJ,eAAe,KAAK,CAAC,UAAU,6BAA6B,KAAK,CAAC,gBAAgB,IAAI;gCACtF,EAAE;gCACF,oBAAoB,MAAM,CAAC,aAAa,EAAE;gCAC1C,qBAAqB,MAAM,CAAC,MAAM,EAAE;gCACpC,EAAE;gCACF,oFAAoF;gCACpF,iEAAiE;6BAClE,CAAC,IAAI,CAAC,IAAI,CAAC;yBACb;qBACF;iBACF,CAAC;YACJ,CAAC;YAED,MAAM,KAAK,CAAC,gBAAgB,CAAC,CAAC;YAC9B,MAAM,GAAG,MAAM,MAAM,CAAC,uBAAuB,CAAC,KAAK,CAAC,YAAY,EAAE,MAAM,CAAC,aAAa,CAAC,CAAC;QAC1F,CAAC;QAED,IAAI,MAAM,CAAC,MAAM,KAAK,QAAQ,EAAE,CAAC;YAC/B,OAAO;gBACL,OAAO,EAAE,IAAI;gBACb,OAAO,EAAE;oBACP;wBACE,IAAI,EAAE,MAAM;wBACZ,IAAI,EAAE;4BACJ,eAAe,KAAK,CAAC,UAAU,YAAY;4BAC3C,EAAE;4BACF,oBAAoB,MAAM,CAAC,aAAa,EAAE;4BAC1C,YAAY,MAAM,CAAC,YAAY,IAAI,eAAe,EAAE;4BACpD,EAAE;4BACF,gBAAgB;4BAChB,sFAAsF;4BACtF,qEAAqE;4BACrE,yDAAyD;yBAC1D,CAAC,IAAI,CAAC,IAAI,CAAC;qBACb;iBACF;aACF,CAAC;QACJ,CAAC;QAED,oBAAoB;QACpB,MAAM,SAAS,GACb,MAAM,CAAC,UAAU,KAAK,IAAI;YACxB,CAAC,CAAC,GAAG,CAAC,MAAM,CAAC,UAAU,GAAG,SAAS,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,KAAK;YACpD,CAAC,CAAC,cAAc,CAAC;QAErB,
|
|
1
|
+
{"version":3,"file":"import_huggingface_model.js","sourceRoot":"","sources":["../../src/tools/import_huggingface_model.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AACxB,OAAO,EAAkB,aAAa,EAAE,MAAM,cAAc,CAAC;AAI7D,MAAM,CAAC,MAAM,iCAAiC,GAAG,CAAC,CAAC,MAAM,CAAC;IACxD,YAAY,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE;IAC/B,UAAU,EAAE,CAAC;SACV,MAAM,EAAE;SACR,KAAK,CAAC,gBAAgB,EAAE,2EAA2E,CAAC;IACvG,QAAQ,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,OAAO,CAAC,MAAM,CAAC;IAC/C,QAAQ,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;IAC/B,mBAAmB,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,QAAQ,EAAE,CAAC,OAAO,CAAC,IAAI,CAAC;IACzD,gBAAgB,EAAE,CAAC;SAChB,MAAM,EAAE;SACR,GAAG,EAAE;SACL,GAAG,CAAC,CAAC,CAAC;SACN,GAAG,CAAC,GAAG,CAAC;SACR,QAAQ,EAAE;SACV,OAAO,CAAC,GAAG,CAAC;CAChB,CAAC,CAAC;AAIH,MAAM,gBAAgB,GAAG,KAAK,CAAC;AAE/B,MAAM,CAAC,KAAK,UAAU,6BAA6B,CACjD,MAAsB,EACtB,KAAkC;IAElC,IAAI,CAAC;QACH,MAAM,IAAI,GAAiE;YACzE,UAAU,EAAE,KAAK,CAAC,UAAU;YAC5B,QAAQ,EAAE,KAAK,CAAC,QAAQ;SACzB,CAAC;QACF,IAAI,KAAK,CAAC,QAAQ,KAAK,SAAS,EAAE,CAAC;YACjC,IAAI,CAAC,QAAQ,GAAG,KAAK,CAAC,QAAQ,CAAC;QACjC,CAAC;QAED,MAAM,GAAG,GAAG,MAAM,MAAM,CAAC,sBAAsB,CAAC,KAAK,CAAC,YAAY,EAAE,IAAI,CAAC,CAAC;QAE1E,IAAI,CAAC,KAAK,CAAC,mBAAmB,EAAE,CAAC;YAC/B,OAAO;gBACL,OAAO,EAAE;oBACP;wBACE,IAAI,EAAE,MAAM;wBACZ,IAAI,EAAE;4BACJ,wBAAwB,KAAK,CAAC,UAAU,KAAK;4BAC7C,EAAE;4BACF,oBAAoB,GAAG,CAAC,aAAa,EAAE;4BACvC,aAAa,GAAG,CAAC,MAAM,EAAE;4BACzB,EAAE;4BACF,0BAA0B;4BAC1B,0DAA0D,KAAK,CAAC,YAAY,mBAAmB,KAAK,CAAC,UAAU,mCAAmC;4BAClJ,EAAE;4BACF,sFAAsF;yBACvF,CAAC,IAAI,CAAC,IAAI,CAAC;qBACb;iBACF;aACF,CAAC;QACJ,CAAC;QAED,qCAAqC;QACrC,MAAM,UAAU,GAAG,IAAI,CAAC,GAAG,EAAE,GAAG,KAAK,CAAC,gBAAgB,GAAG,KAAK,CAAC;QAC/D,IAAI,MAAM,GAAgB,GAAG,CAAC;QAE9B,OAAO,MAAM,CAAC,MAAM,KAAK,MAAM,IAAI,MAAM,CAAC,MAAM,KAAK,QAAQ,EAAE,CAAC;YAC9D,IAAI,IAAI,CAAC,GAAG,EAAE,IAAI,UAAU,EAAE,CAAC;gBAC7B,OAAO;oBACL,OAAO,EAAE;wBACP;4BACE,IAAI,EAAE,MAAM;4BACZ,IAAI,EAAE;gCACJ,eAAe,KAAK,CAAC,UAAU,6BAA6B,KAAK,CAAC,gBAAgB,IAAI;gCACtF,EAAE;gCACF,oBAAoB,MAAM,CAAC,aAAa,EAAE;gCAC1C,qBAAqB,MAAM,CAAC,MAAM,EAAE;gCACpC,EAAE;gCACF,oFAAoF;gCACpF,iEAAiE;6BAClE,CAAC,IAAI,CAAC,IAAI,CAAC;yBACb;qBACF;iBACF,CAAC;YACJ,CAAC;YAED,MAAM,KAAK,CAAC,gBAAgB,CAAC,CAAC;YAC9B,MAAM,GAAG,MAAM,MAAM,CAAC,uBAAuB,CAAC,KAAK,CAAC,YAAY,EAAE,MAAM,CAAC,aAAa,CAAC,CAAC;QAC1F,CAAC;QAED,IAAI,MAAM,CAAC,MAAM,KAAK,QAAQ,EAAE,CAAC;YAC/B,OAAO;gBACL,OAAO,EAAE,IAAI;gBACb,OAAO,EAAE;oBACP;wBACE,IAAI,EAAE,MAAM;wBACZ,IAAI,EAAE;4BACJ,eAAe,KAAK,CAAC,UAAU,YAAY;4BAC3C,EAAE;4BACF,oBAAoB,MAAM,CAAC,aAAa,EAAE;4BAC1C,YAAY,MAAM,CAAC,YAAY,IAAI,eAAe,EAAE;4BACpD,EAAE;4BACF,gBAAgB;4BAChB,sFAAsF;4BACtF,qEAAqE;4BACrE,yDAAyD;yBAC1D,CAAC,IAAI,CAAC,IAAI,CAAC;qBACb;iBACF;aACF,CAAC;QACJ,CAAC;QAED,oBAAoB;QACpB,MAAM,SAAS,GACb,MAAM,CAAC,UAAU,KAAK,IAAI;YACxB,CAAC,CAAC,GAAG,CAAC,MAAM,CAAC,UAAU,GAAG,SAAS,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,KAAK;YACpD,CAAC,CAAC,cAAc,CAAC;QAErB,qEAAqE;QACrE,uEAAuE;QACvE,iDAAiD;QACjD,MAAM,WAAW,GAAG,KAAK,CAAC,OAAO,CAAC,MAAM,CAAC,SAAS,CAAC,IAAI,MAAM,CAAC,SAAS,CAAC,MAAM,GAAG,CAAC,CAAC;QACnF,MAAM,KAAK,GAAa;YACtB,sBAAsB,KAAK,CAAC,UAAU,IAAI;YAC1C,EAAE;SACH,CAAC;QAEF,IAAI,WAAW,EAAE,CAAC;YAChB,MAAM,KAAK,GAAG,MAAM,CAAC,SAAiD,CAAC;YACvE,KAAK,CAAC,IAAI,CACR,YAAY,KAAK,CAAC,MAAM,wCAAwC,EAChE,EAAE,CACH,CAAC;YACF,KAAK,MAAM,IAAI,IAAI,KAAK,EAAE,CAAC;gBACzB,MAAM,QAAQ,GACZ,IAAI,CAAC,UAAU,GAAG,CAAC;oBACjB,CAAC,CAAC,GAAG,CAAC,IAAI,CAAC,UAAU,GAAG,SAAS,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,KAAK;oBAClD,CAAC,CAAC,cAAc,CAAC;gBACrB,KAAK,CAAC,IAAI,CACR,MAAM,IAAI,CAAC,IAAI,KAAK,IAAI,CAAC,QAAQ,KAAK,QAAQ,oBAAoB,IAAI,CAAC,EAAE,EAAE,CAC5E,CAAC;YACJ,CAAC;YACD,KAAK,CAAC,IAAI,CACR,EAAE,EACF,8EAA8E,EAC9E,cAAc,EACd,cAAc,KAAK,CAAC,UAAU,IAAI,EAClC,2BAA2B,EAC3B,+BAA+B,KAAK,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,IAAI,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,EAC3E,iBAAiB,KAAK,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,IAAI,CAAC,CAAC,IAAI,GAAG,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,EAC/D,OAAO,EACP,MAAM,EACN,EAAE,EACF,gHAAgH,CACjH,CAAC;QACJ,CAAC;aAAM,CAAC;YACN,KAAK,CAAC,IAAI,CACR,kBAAkB,MAAM,CAAC,WAAW,EAAE,EACtC,eAAe,MAAM,CAAC,QAAQ,IAAI,WAAW,EAAE,EAC/C,WAAW,SAAS,EAAE,EACtB,EAAE,EACF,0EAA0E,EAC1E,uBAAuB,KAAK,CAAC,UAAU,oBAAoB,MAAM,CAAC,WAAW,MAAM,CACpF,CAAC;QACJ,CAAC;QAED,OAAO;YACL,OAAO,EAAE,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC;SACpD,CAAC;IACJ,CAAC;IAAC,OAAO,GAAG,EAAE,CAAC;QACb,IAAI,GAAG,YAAY,aAAa,EAAE,CAAC;YACjC,OAAO;gBACL,OAAO,EAAE,IAAI;gBACb,OAAO,EAAE;oBACP;wBACE,IAAI,EAAE,MAAM;wBACZ,IAAI,EACF,GAAG,CAAC,MAAM,KAAK,GAAG;4BAChB,CAAC,CAAC,2GAA2G,GAAG,CAAC,MAAM,EAAE;4BACzH,CAAC,CAAC,GAAG,CAAC,MAAM,KAAK,GAAG;gCAClB,CAAC,CAAC,4EAA4E;oCAC5E,uEAAuE;gCACzE,CAAC,CAAC,qBAAqB,GAAG,CAAC,MAAM,KAAK,GAAG,CAAC,MAAM,EAAE;qBACzD;iBACF;aACF,CAAC;QACJ,CAAC;QACD,MAAM,GAAG,CAAC;IACZ,CAAC;AACH,CAAC;AAED,SAAS,KAAK,CAAC,EAAU;IACvB,OAAO,IAAI,OAAO,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,UAAU,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC;AAC/C,CAAC"}
|
|
@@ -0,0 +1,47 @@
|
|
|
1
|
+
import { z } from "zod";
|
|
2
|
+
import { EdgeGateClient } from "../client.js";
|
|
3
|
+
import type { ToolResult } from "./setup_workspace.js";
|
|
4
|
+
export declare const llmCompileInputSchema: z.ZodObject<{
|
|
5
|
+
workspace_id: z.ZodString;
|
|
6
|
+
/**
|
|
7
|
+
* Pre-uploaded source artifacts (1 per component). Typical Llama-class:
|
|
8
|
+
* 3 artifacts (prompt + token + kv_cache). For other architectures pass
|
|
9
|
+
* M artifacts matching the number of components.
|
|
10
|
+
*/
|
|
11
|
+
source_artifact_ids: z.ZodArray<z.ZodString, "many">;
|
|
12
|
+
/** AI Hub device id (e.g. "x_elite", "x2_elite", "sa8295p"). */
|
|
13
|
+
device_id: z.ZodString;
|
|
14
|
+
/**
|
|
15
|
+
* Label only — backend always compiles to QNN_DLC under the hood. The
|
|
16
|
+
* label is used downstream for profile dispatch hints.
|
|
17
|
+
*/
|
|
18
|
+
target_runtime: z.ZodDefault<z.ZodString>;
|
|
19
|
+
/** Default [128, 1] — typical prompt/token split for Llama-class. */
|
|
20
|
+
sequence_lengths: z.ZodDefault<z.ZodArray<z.ZodNumber, "many">>;
|
|
21
|
+
/** Default [4096]. */
|
|
22
|
+
context_lengths: z.ZodDefault<z.ZodArray<z.ZodNumber, "many">>;
|
|
23
|
+
/**
|
|
24
|
+
* Per-component role labels (e.g. ["prompt", "token", "kv_cache"]).
|
|
25
|
+
* Required when source_artifact_ids has >1 entry — the AI Hub SDK
|
|
26
|
+
* mandates graph_names for multi-component compiles.
|
|
27
|
+
*/
|
|
28
|
+
roles: z.ZodOptional<z.ZodArray<z.ZodString, "many">>;
|
|
29
|
+
}, "strip", z.ZodTypeAny, {
|
|
30
|
+
workspace_id: string;
|
|
31
|
+
source_artifact_ids: string[];
|
|
32
|
+
sequence_lengths: number[];
|
|
33
|
+
context_lengths: number[];
|
|
34
|
+
target_runtime: string;
|
|
35
|
+
device_id: string;
|
|
36
|
+
roles?: string[] | undefined;
|
|
37
|
+
}, {
|
|
38
|
+
workspace_id: string;
|
|
39
|
+
source_artifact_ids: string[];
|
|
40
|
+
device_id: string;
|
|
41
|
+
sequence_lengths?: number[] | undefined;
|
|
42
|
+
context_lengths?: number[] | undefined;
|
|
43
|
+
roles?: string[] | undefined;
|
|
44
|
+
target_runtime?: string | undefined;
|
|
45
|
+
}>;
|
|
46
|
+
export type LLMCompileInput = z.infer<typeof llmCompileInputSchema>;
|
|
47
|
+
export declare function llmCompileHandler(client: EdgeGateClient, input: LLMCompileInput): Promise<ToolResult>;
|
|
@@ -0,0 +1,111 @@
|
|
|
1
|
+
import { z } from "zod";
|
|
2
|
+
import { EdgeGateError } from "../client.js";
|
|
3
|
+
export const llmCompileInputSchema = z.object({
|
|
4
|
+
workspace_id: z.string().uuid(),
|
|
5
|
+
/**
|
|
6
|
+
* Pre-uploaded source artifacts (1 per component). Typical Llama-class:
|
|
7
|
+
* 3 artifacts (prompt + token + kv_cache). For other architectures pass
|
|
8
|
+
* M artifacts matching the number of components.
|
|
9
|
+
*/
|
|
10
|
+
source_artifact_ids: z.array(z.string().uuid()).min(1).max(8),
|
|
11
|
+
/** AI Hub device id (e.g. "x_elite", "x2_elite", "sa8295p"). */
|
|
12
|
+
device_id: z.string().min(1),
|
|
13
|
+
/**
|
|
14
|
+
* Label only — backend always compiles to QNN_DLC under the hood. The
|
|
15
|
+
* label is used downstream for profile dispatch hints.
|
|
16
|
+
*/
|
|
17
|
+
target_runtime: z.string().min(1).default("genie"),
|
|
18
|
+
/** Default [128, 1] — typical prompt/token split for Llama-class. */
|
|
19
|
+
sequence_lengths: z.array(z.number().int().positive()).default([128, 1]),
|
|
20
|
+
/** Default [4096]. */
|
|
21
|
+
context_lengths: z.array(z.number().int().positive()).default([4096]),
|
|
22
|
+
/**
|
|
23
|
+
* Per-component role labels (e.g. ["prompt", "token", "kv_cache"]).
|
|
24
|
+
* Required when source_artifact_ids has >1 entry — the AI Hub SDK
|
|
25
|
+
* mandates graph_names for multi-component compiles.
|
|
26
|
+
*/
|
|
27
|
+
roles: z.array(z.string().min(1)).optional(),
|
|
28
|
+
});
|
|
29
|
+
export async function llmCompileHandler(client, input) {
|
|
30
|
+
// Client-side guard: roles required when multi-component. Surface the
|
|
31
|
+
// same constraint the backend enforces so the agent gets a clean error
|
|
32
|
+
// without a network round-trip.
|
|
33
|
+
if (input.source_artifact_ids.length > 1 && (!input.roles || input.roles.length === 0)) {
|
|
34
|
+
return {
|
|
35
|
+
isError: true,
|
|
36
|
+
content: [
|
|
37
|
+
{
|
|
38
|
+
type: "text",
|
|
39
|
+
text: `'roles' is required when source_artifact_ids has more than 1 entry ` +
|
|
40
|
+
`(the AI Hub SDK requires graph_names for multi-component compiles). ` +
|
|
41
|
+
`Typical Llama-class: roles=["prompt", "token", "kv_cache"].`,
|
|
42
|
+
},
|
|
43
|
+
],
|
|
44
|
+
};
|
|
45
|
+
}
|
|
46
|
+
if (input.roles && input.roles.length !== input.source_artifact_ids.length) {
|
|
47
|
+
return {
|
|
48
|
+
isError: true,
|
|
49
|
+
content: [
|
|
50
|
+
{
|
|
51
|
+
type: "text",
|
|
52
|
+
text: `'roles' length (${input.roles.length}) must match source_artifact_ids ` +
|
|
53
|
+
`length (${input.source_artifact_ids.length}).`,
|
|
54
|
+
},
|
|
55
|
+
],
|
|
56
|
+
};
|
|
57
|
+
}
|
|
58
|
+
try {
|
|
59
|
+
const job = await client.submitLLMCompile(input.workspace_id, {
|
|
60
|
+
source_artifact_ids: input.source_artifact_ids,
|
|
61
|
+
device_id: input.device_id,
|
|
62
|
+
target_runtime: input.target_runtime,
|
|
63
|
+
sequence_lengths: input.sequence_lengths,
|
|
64
|
+
context_lengths: input.context_lengths,
|
|
65
|
+
...(input.roles !== undefined ? { roles: input.roles } : {}),
|
|
66
|
+
});
|
|
67
|
+
return {
|
|
68
|
+
content: [
|
|
69
|
+
{
|
|
70
|
+
type: "text",
|
|
71
|
+
text: [
|
|
72
|
+
`LLM compile job submitted.`,
|
|
73
|
+
``,
|
|
74
|
+
`- compile_job_id: ${job.compile_job_id}`,
|
|
75
|
+
`- status: ${job.status}`,
|
|
76
|
+
`- components: ${input.source_artifact_ids.length}`,
|
|
77
|
+
`- device: ${input.device_id}`,
|
|
78
|
+
`- target_runtime: ${input.target_runtime} (compiled to QNN_DLC under the hood)`,
|
|
79
|
+
``,
|
|
80
|
+
`Poll status with:`,
|
|
81
|
+
` edgegate_check_llm_compile_status({ workspace_id: "${input.workspace_id}", compile_job_id: "${job.compile_job_id}" })`,
|
|
82
|
+
``,
|
|
83
|
+
`When status=completed, composite_artifact_id is the QNN_DLC linked model ready to use in a pipeline.`,
|
|
84
|
+
].join("\n"),
|
|
85
|
+
},
|
|
86
|
+
],
|
|
87
|
+
};
|
|
88
|
+
}
|
|
89
|
+
catch (err) {
|
|
90
|
+
if (err instanceof EdgeGateError) {
|
|
91
|
+
return {
|
|
92
|
+
isError: true,
|
|
93
|
+
content: [
|
|
94
|
+
{
|
|
95
|
+
type: "text",
|
|
96
|
+
text: err.status === 402
|
|
97
|
+
? `Your workspace has hit its monthly LLM compile cap (default 100/mo on Pro). ` +
|
|
98
|
+
`Upgrade at https://edgegate.frozo.ai/pricing or wait until next billing cycle.\n\n` +
|
|
99
|
+
`Detail: ${err.detail}`
|
|
100
|
+
: err.status === 401
|
|
101
|
+
? "EDGEGATE_API_KEY is missing, expired, or revoked. Generate a fresh key at " +
|
|
102
|
+
"https://edgegate.frozo.ai/workspace/<id>/settings#api-keys and retry."
|
|
103
|
+
: `EdgeGate returned ${err.status}: ${err.detail}`,
|
|
104
|
+
},
|
|
105
|
+
],
|
|
106
|
+
};
|
|
107
|
+
}
|
|
108
|
+
throw err;
|
|
109
|
+
}
|
|
110
|
+
}
|
|
111
|
+
//# sourceMappingURL=llm_compile.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"llm_compile.js","sourceRoot":"","sources":["../../src/tools/llm_compile.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AACxB,OAAO,EAAkB,aAAa,EAAE,MAAM,cAAc,CAAC;AAG7D,MAAM,CAAC,MAAM,qBAAqB,GAAG,CAAC,CAAC,MAAM,CAAC;IAC5C,YAAY,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE;IAC/B;;;;OAIG;IACH,mBAAmB,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;IAC7D,gEAAgE;IAChE,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;IAC5B;;;OAGG;IACH,cAAc,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,OAAO,CAAC;IAClD,qEAAqE;IACrE,gBAAgB,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE,CAAC,QAAQ,EAAE,CAAC,CAAC,OAAO,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC,CAAC;IACxE,sBAAsB;IACtB,eAAe,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE,CAAC,QAAQ,EAAE,CAAC,CAAC,OAAO,CAAC,CAAC,IAAI,CAAC,CAAC;IACrE;;;;OAIG;IACH,KAAK,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,QAAQ,EAAE;CAC7C,CAAC,CAAC;AAIH,MAAM,CAAC,KAAK,UAAU,iBAAiB,CACrC,MAAsB,EACtB,KAAsB;IAEtB,sEAAsE;IACtE,uEAAuE;IACvE,gCAAgC;IAChC,IAAI,KAAK,CAAC,mBAAmB,CAAC,MAAM,GAAG,CAAC,IAAI,CAAC,CAAC,KAAK,CAAC,KAAK,IAAI,KAAK,CAAC,KAAK,CAAC,MAAM,KAAK,CAAC,CAAC,EAAE,CAAC;QACvF,OAAO;YACL,OAAO,EAAE,IAAI;YACb,OAAO,EAAE;gBACP;oBACE,IAAI,EAAE,MAAM;oBACZ,IAAI,EACF,qEAAqE;wBACrE,sEAAsE;wBACtE,6DAA6D;iBAChE;aACF;SACF,CAAC;IACJ,CAAC;IACD,IAAI,KAAK,CAAC,KAAK,IAAI,KAAK,CAAC,KAAK,CAAC,MAAM,KAAK,KAAK,CAAC,mBAAmB,CAAC,MAAM,EAAE,CAAC;QAC3E,OAAO;YACL,OAAO,EAAE,IAAI;YACb,OAAO,EAAE;gBACP;oBACE,IAAI,EAAE,MAAM;oBACZ,IAAI,EACF,mBAAmB,KAAK,CAAC,KAAK,CAAC,MAAM,mCAAmC;wBACxE,WAAW,KAAK,CAAC,mBAAmB,CAAC,MAAM,IAAI;iBAClD;aACF;SACF,CAAC;IACJ,CAAC;IAED,IAAI,CAAC;QACH,MAAM,GAAG,GAAG,MAAM,MAAM,CAAC,gBAAgB,CAAC,KAAK,CAAC,YAAY,EAAE;YAC5D,mBAAmB,EAAE,KAAK,CAAC,mBAAmB;YAC9C,SAAS,EAAE,KAAK,CAAC,SAAS;YAC1B,cAAc,EAAE,KAAK,CAAC,cAAc;YACpC,gBAAgB,EAAE,KAAK,CAAC,gBAAgB;YACxC,eAAe,EAAE,KAAK,CAAC,eAAe;YACtC,GAAG,CAAC,KAAK,CAAC,KAAK,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,KAAK,EAAE,KAAK,CAAC,KAAK,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;SAC7D,CAAC,CAAC;QAEH,OAAO;YACL,OAAO,EAAE;gBACP;oBACE,IAAI,EAAE,MAAM;oBACZ,IAAI,EAAE;wBACJ,4BAA4B;wBAC5B,EAAE;wBACF,qBAAqB,GAAG,CAAC,cAAc,EAAE;wBACzC,aAAa,GAAG,CAAC,MAAM,EAAE;wBACzB,iBAAiB,KAAK,CAAC,mBAAmB,CAAC,MAAM,EAAE;wBACnD,aAAa,KAAK,CAAC,SAAS,EAAE;wBAC9B,qBAAqB,KAAK,CAAC,cAAc,uCAAuC;wBAChF,EAAE;wBACF,mBAAmB;wBACnB,wDAAwD,KAAK,CAAC,YAAY,uBAAuB,GAAG,CAAC,cAAc,MAAM;wBACzH,EAAE;wBACF,sGAAsG;qBACvG,CAAC,IAAI,CAAC,IAAI,CAAC;iBACb;aACF;SACF,CAAC;IACJ,CAAC;IAAC,OAAO,GAAG,EAAE,CAAC;QACb,IAAI,GAAG,YAAY,aAAa,EAAE,CAAC;YACjC,OAAO;gBACL,OAAO,EAAE,IAAI;gBACb,OAAO,EAAE;oBACP;wBACE,IAAI,EAAE,MAAM;wBACZ,IAAI,EACF,GAAG,CAAC,MAAM,KAAK,GAAG;4BAChB,CAAC,CAAC,8EAA8E;gCAC9E,oFAAoF;gCACpF,WAAW,GAAG,CAAC,MAAM,EAAE;4BACzB,CAAC,CAAC,GAAG,CAAC,MAAM,KAAK,GAAG;gCAClB,CAAC,CAAC,4EAA4E;oCAC5E,uEAAuE;gCACzE,CAAC,CAAC,qBAAqB,GAAG,CAAC,MAAM,KAAK,GAAG,CAAC,MAAM,EAAE;qBACzD;iBACF;aACF,CAAC;QACJ,CAAC;QACD,MAAM,GAAG,CAAC;IACZ,CAAC;AACH,CAAC"}
|
package/dist/types.d.ts
CHANGED
|
@@ -187,6 +187,22 @@ export interface AuditReport {
|
|
|
187
187
|
generated_at?: string;
|
|
188
188
|
}
|
|
189
189
|
export type HFImportStatus = "queued" | "downloading" | "uploading_to_hub" | "done" | "failed";
|
|
190
|
+
/**
|
|
191
|
+
* One entry in a multi-part HF import. F-T6.2 (commit 3636c20) added
|
|
192
|
+
* multi-part `.bin` auto-discovery: a Volko76-style repo with 3 parts
|
|
193
|
+
* (prompt / token / kv_cache) now unpacks into 3 ArtifactRefs, in addition
|
|
194
|
+
* to the legacy single `artifact_id` field (which still points at part 1
|
|
195
|
+
* for backward compat with older MCP clients).
|
|
196
|
+
*/
|
|
197
|
+
export interface ArtifactRef {
|
|
198
|
+
id: UUID;
|
|
199
|
+
filename: string;
|
|
200
|
+
size_bytes: number;
|
|
201
|
+
/** "prompt" | "token" | "kv_cache" for M=3 multi-part imports; "part_N" otherwise. */
|
|
202
|
+
role: string;
|
|
203
|
+
/** 1-based. */
|
|
204
|
+
part_index: number;
|
|
205
|
+
}
|
|
190
206
|
export interface HFImportJob {
|
|
191
207
|
import_job_id: string;
|
|
192
208
|
status: HFImportStatus;
|
|
@@ -196,6 +212,36 @@ export interface HFImportJob {
|
|
|
196
212
|
size_bytes: number | null;
|
|
197
213
|
filename: string | null;
|
|
198
214
|
error_detail: string | null;
|
|
215
|
+
/**
|
|
216
|
+
* Plan B / F-T6.2 — when the HF repo contains multiple .bin parts, this
|
|
217
|
+
* array surfaces every part (size > 1 for multi-part imports). Omitted
|
|
218
|
+
* or single-element for legacy single-file imports.
|
|
219
|
+
*/
|
|
220
|
+
artifacts?: ArtifactRef[];
|
|
221
|
+
}
|
|
222
|
+
export type LLMCompileStatus = "queued" | "running" | "completed" | "failed";
|
|
223
|
+
export interface LLMCompileProgress {
|
|
224
|
+
compile_jobs_done: number;
|
|
225
|
+
compile_jobs_total: number;
|
|
226
|
+
}
|
|
227
|
+
/**
|
|
228
|
+
* Returned by POST /v1/workspaces/{ws}/llm-compile and GET
|
|
229
|
+
* /v1/workspaces/{ws}/llm-compile/{compile_job_id}. POST returns the
|
|
230
|
+
* initial queued state; GET reflects current state. On success,
|
|
231
|
+
* `composite_artifact_id` is the QNN_DLC linked composite ready to be
|
|
232
|
+
* referenced from a pipeline. On failure, `error_detail` is set.
|
|
233
|
+
*/
|
|
234
|
+
export interface LLMCompileJob {
|
|
235
|
+
compile_job_id: UUID;
|
|
236
|
+
status: LLMCompileStatus;
|
|
237
|
+
composite_artifact_id: UUID | null;
|
|
238
|
+
error_detail: string | null;
|
|
239
|
+
progress: LLMCompileProgress;
|
|
240
|
+
/** Echoed for caller convenience. */
|
|
241
|
+
device_id?: string;
|
|
242
|
+
target_runtime?: string;
|
|
243
|
+
created_at?: string;
|
|
244
|
+
updated_at?: string;
|
|
199
245
|
}
|
|
200
246
|
export interface MetricDelta {
|
|
201
247
|
current: number | null;
|
package/dist/version.d.ts
CHANGED
|
@@ -1,2 +1,2 @@
|
|
|
1
|
-
export declare const VERSION = "0.
|
|
2
|
-
export declare const USER_AGENT = "edgegate-mcp/0.
|
|
1
|
+
export declare const VERSION = "0.11.0";
|
|
2
|
+
export declare const USER_AGENT = "edgegate-mcp/0.11.0";
|
package/dist/version.js
CHANGED
package/dist/version.js.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"version.js","sourceRoot":"","sources":["../src/version.ts"],"names":[],"mappings":"AAAA,MAAM,CAAC,MAAM,OAAO,GAAG,
|
|
1
|
+
{"version":3,"file":"version.js","sourceRoot":"","sources":["../src/version.ts"],"names":[],"mappings":"AAAA,MAAM,CAAC,MAAM,OAAO,GAAG,QAAQ,CAAC;AAChC,MAAM,CAAC,MAAM,UAAU,GAAG,gBAAgB,OAAO,EAAE,CAAC"}
|