SILVER
/
NODEDC_1C
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
NODEDC_1C/llm_normalizer/docs/API.md

2.0 KiB
Raw Blame History

API Contract

Base URL: http://localhost:8787

POST /api/normalize

Core request fields:

  • promptVersion (e.g. normalizer_v2_0_2)
  • schemaVersion (e.g. v2_0_2)
  • userQuestion
  • model transport fields (apiKey, model, baseUrl, temperature, maxOutputTokens)

For v2.0.2, backend returns:

  • schema_version: "v2_0_2"
  • normalized payload with normalized_query_v2_0_2
  • deterministic route_hint_summary

Schema selection:

  • promptVersion=normalizer_v2_0_2 or schemaVersion=v2_0_2 -> normalized_query_v2_0_2
  • promptVersion=normalizer_v2_0_1 or schemaVersion=v2_0_1 -> normalized_query_v2_0_1
  • promptVersion=normalizer_v2 or schemaVersion=v2 -> normalized_query_v2
  • otherwise -> normalized_query_v1

POST /api/eval/run

Supports v2 family (v2, v2_0_1, v2_0_2) with inline batch via rawQuestions.

Assistant Stage 1 eval target is additive and enabled only when eval_target=assistant_stage1. Legacy normalizer eval remains default when eval_target is omitted.

Example:

{
  "mode": "single-pass-strict",
  "rawQuestions": "вопрос 1; вопрос 2; вопрос 3",
  "useMock": false,
  "normalizeConfig": {
    "promptVersion": "normalizer_v2_0_2",
    "schemaVersion": "v2_0_2",
    "model": "gpt-4o-mini",
    "temperature": 0,
    "maxOutputTokens": 900
  }
}

Assistant Stage 1 example:

{
  "eval_target": "assistant_stage1",
  "mode": "single-pass-strict",
  "useMock": true,
  "caseSetFile": "assistant_stage1_canonical_v0_1.json",
  "compare_with_report_file": "assistant-stage1-baseline.json",
  "normalizeConfig": {
    "promptVersion": "normalizer_v2_0_2"
  }
}

v2.0.2 eval metrics include:

  • schema_validation_pass_rate
  • scope_detection_accuracy
  • route_resolution_accuracy
  • no_route_precision
  • false_no_route_rate
  • execution_state_consistency_rate
  • clarification_precision
  • clarification_recall

Presets and History

  • GET /api/presets
  • POST /api/presets/save
  • GET /api/history
  • GET /api/history/:trace_id