RHEL AI Pro Suite APIs

RHEL AI Pro Suite에서 제공하는 REST API에 대해 설명합니다.

POST `/v1/documents/texts`

Text를 Vector DB에 색인하는 API

Request

Body

content: (Required) 청크 및 임베딩 변환이 될 원본 텍스트
metadata: (Required) 청크 관련 메타 데이터
- source: (Optional) 텍스트 출처
- name: (Optional) document 이름
- domain: (Required) taxonomy 분야
- author: (Optional) 작성자
index_name: (Optional) 색인 대상 인덱스 이름. 만약 정의되어 있지 않으면, 설정 파일내 plugins.vector_store.[vector_db명].index_name을 참조함.

Example

curl -X POST http://localhost:8888/v1/documents/texts -H "Content-Type: application/json" -d '{
    "content": "Zerobaseone (RR: Jerobeiseuwon; stylized in all caps; abbreviated as ZB1) is a South Korean boy band ...",
    "metadata": {
        "source": "https://test-example.com/10",
        "name": "k-pop singer",
        "domain": "k-pop",
        "author" : "Michael"
    },
    "index_name": "ai_odyssey_demo_documents-000001"
}'

Response

Success (200)

{
    "message": "success",
    "index_name": "ai_odyssey_demo_documents-000001"
}

Failure (500)

{
    "TID": "8ca96e7d-9d05-477c-81bb-9aae2d91a1d2",
    "error code": 11999,
    "message": "Chunker Fail",
    "detail": "chunk exception"
}

POST `/v1/documents/files`

Text 형식의 document file을 Vector DB에 색인하는 API

Request

Parameter

index_name: (Optional) 색인 대상 인덱스 이름. 만약 정의되어 있지 않으면, 설정 파일내 plugins.vector_store.[vector_db명].index_name을 참조함.

Body

file: (Required) text 형태의 파일
source: (Optional) 텍스트 출처
name: (Optional) document 이름
domain: (Required) taxonomy 분야
author: (Optional) 작성자

Example

curl -X POST http://localhost:8888/v1/documents/files?index_name=ai_odyssey_demo_documents-000001 \
    -H "accept: application/json" \
    -H "Content-Type: multipart/form-data" \
    -F "file=@/path/to/your/file.txt" \
    -F "source=https://test-example.com/10" \
    -F "name=k-pop singer" \
    -F "domain=k-pop" \
    -F "author=Michael"

Response

Success (200)

{
    "message": "success",
    "index_name": "ai_odyssey_demo_documents-000001"
}

Failure (500)

{
    "TID": "8ca96e7d-9d05-477c-81bb-9aae2d91a1d2",
    "error code": 11999,
    "message": "Chunker Fail",
    "detail": "chunk exception"
}

POST `/v1/qna/generate`

QnA 테스트 셋 생성 및 qna.yaml 파일 내용을 전달하는 API

Request

Body

target_model: (Optional) QnA 테스트 셋을 생성할 LLM. 만약 정의되어 있지 않으면, 설정 파일내 plugins.evaluation.rhel-ai.model을 참조함.
critic_llm: (Optional) 비평가 LLM 접속 정보 설정. 만약 정의되어 있지 않으면, 설정 파일내 plugins.evaluation.rhel-ai.critic_llm을 참조함.
- host: LLM 서버 접속 정보
- model: 검증 대상 LLM 모델명
- headers: (Optional) LLM과 HTTP 통신 시 header 로 전달하는 값들
testset_size: (Optional) 조회된 Content 1개당 생성할 QnA 테스트 셋 개수. Default: 3
domain: (Required) taxonomy 분야
document_index_name: (Optional) 컨텐츠 조회 대상 인덱스 이름. Default: 설정 파일내 vector_store의 index_name
testset_index_name: (Optional) QA 테스트 셋 색인 대상 인덱스 이름. 만약 정의되어 있지 않으면, 설정 파일내 plugins.vector_store.[vector_db명].index_name을 참조함.
qna_yaml: (Required) QnA yaml 파일 설정
- version: (Optional) 버전. Default: 3
- created_by: (Required) Git 사용자 이름
- document_outline: (Optional) 문서의 개요. 만약 정의되어 있지 않으면, LLM에 본문을 전달하여 요약문을 생성함.
- repo: (Required) 마크다운 파일을 보관하는 repository의 URL
- commit: (Required) 마크다운 파일과 함께 리포지토리에 있는 커밋의 SHA
- patterns: (Required) 리포지토리에 있는 마크다운 파일을 지정

Example

curl -X POST http://localhost:8888/v1/qna/generate -H "Content-Type: application/json" -d '{
    "target_model": "/home/[user-id]/.cache/instructlab/models/merlinite-7b-lab-Q4_K_M.gguf",
    "critic_llm": {
        "model": "/home/[user-id]/.cache/instructlab/models/mistral-7b-instruct-v0.2.Q4_K_M.gguf",
        "host": "http://127.0.0.1:8000",
        "headers": {
            "Content-Type": "application/json"
        }
    },
    "testset_size": 3,
    "domain": "k-pop",
    "document_index_name": "ai_odyssey_demo_documents-000001",
    "testset_index_name": "ai_odyssey_demo_testset-000001",
    "qna_yaml": {
        "version": 3,
        "created_by": "test_author",
        "repo": "https://github.com/juliadenham/Summit_knowledge",
        "commit": "5f7158a5ce83c4ff493bfe341fe31ecad64ff697",
        "patterns": ["chickadee.md"]
    }
}'

Response:

Success (200)

version: 3
created_by: test_author
domain: k-pop
seed_examples:
- context: 'Zerobaseone (RR: Jerobeiseuwon; stylized in all caps; abbreviated as ZB1) is a South Korean
    boy band formed through Mnet''s reality competition program Boys Planet and managed by WakeOne. The
    group consists of nine members: Kim Ji-woong, Zhang Hao, Sung Han-bin, Seok Matthew, Kim Tae-rae,
    Ricky, Kim Gyu-vin, Park Gun-wook, and Han Yu-jin. ...'
  questions_and_answers:
  - question: ' "Which South Korean boy band was formed through Mnet''s reality competition program Boys
      Planet and achieved commercial success with their debut EP?'
    answer: Zerobaseone
  - question: ' "Which South Korean boy band, formed through Mnet''s Boys Planet competition and managed
      by WakeOne, debuted in 2023 with the EP ''Youth in the Shade'' and achieved commercial success?'
    answer: Zerobaseone
  - question: ' Who are the nine members of Zerobaseone, a South Korean boy band formed through Mnet''s
      Boys Planet competition?'
    answer: 'The nine members of Zerobaseone are: Kim Ji-woong, Zhang Hao, Sung Han-bin, Seok Matthew,
      Kim Tae-rae, Ricky, Kim Gyu-vin, Park Gun-wook, and Han Yu-jin.'
- context: ...
document_outline: ' South Korean boy band Zerobaseone (ZB1), formed through Mnet''s reality competition
  Boys Planet, debuted with EP "Youth in the Shade" in July 2023. The group of nine members, including
  Kim Ji-woong and previous contestants on other shows, sold over two million units, topped South Korean
  charts, and earned global recognition with Billboard Global 200 hit "In Bloom."'
document:
  repo: https://github.com/juliadenham/Summit_knowledge
  commit: 5f7158a5ce83c4ff493bfe341fe31ecad64ff697
  patterns:
  - chickadee.md

Failure (500)

{
    "TID": "8ca96e7d-9d05-477c-81bb-9aae2d91a1d2",
    "error code": 40005,
    "message": "Testset Result Generation Fail!",
    "detail": "testset result exception"
}

PATCH `/v1/documents`

Document 메타 데이터 업데이트

Request

Parameter

index_name: (Optional) 색인 대상 인덱스 이름. 만약 정의되어 있지 않으면, 설정 파일내 plugins.vector_store.[vector_db명].index_name을 참조함.

Body

filter:
- domain: (Required) document 유형
update:
- status: (Required) document 상태값. 현재는 "trained"만 지원

Example

curl -X PATCH http://localhost:8888/v1/documents -H "Content-Type: application/json" -d '{
    "filter": {
        "domain": "k-pop"
    },
    "update": {
        "status": "trained"
    }
}'

Response

Success (200)

{
    "messsage": "success",
    "updated_count": 5
}

Failure (500)

{
    "TID": "6b36673d-ff2e-493c-8e00-ebed52dc5721",
    "error code": 20999,
    "message": "VectorStore Process Fail",
    "detail": "NotFoundError(404, 'index_not_found_exception', 'no such index [aaaa]', aaaa, index_or_alias)"
}

POST `/v1/qna/evaluate`

QnA 테스트 셋 평가 및 평가 결과를 Vector DB에 색인하는 API

Request

Body

target_model: (Optional) 답변(answer)을 생성할 LLM.
critic_llm: (Optional) 비평가 LLM 접속 정보 설정. 만약 정의되어 있지 않으면, 설정 파일내 plugins.evaluation.rhel-ai.critic_llm을 참조함.
- host: LLM 서버 접속 정보
- model: 검증 대상 LLM 모델명
- headers: (Optional) LLM과 HTTP 통신 시 header 로 전달하는 값들
k: (Optional) RAG 파이프라인의 context 사이즈. Default: 3
domain: (Required) taxonomy 분야
testset_index_name: (Optional) QnA 테스트 셋 조회 대상 인덱스 이름. 만약 정의되어 있지 않으면, 설정 파일내 plugins.evaluation.rhel-ai.testset_index_name을 참조함.
evaluation_index_name: (Optional) 생성된 평가 결과 색인 대상 인덱스 이름. 만약 정의되어 있지 않으면, 설정 파일내 plugins.evaluation.rhel-ai.testset_index_name을 참조함.

Example

curl -X POST http://localhost:8888/v1/qna/evaluate -H "Content-Type: application/json" -d '{
    "target_model": "/home/[user-id]/.cache/instructlab/models/merlinite-7b-lab-Q4_K_M.gguf",
    "critic_llm": {
        "model": "/home/[user-id]/.cache/instructlab/models/mistral-7b-instruct-v0.2.Q4_K_M.gguf",
        "host": "http://127.0.0.1:8000",
        "headers": {
            "Content-Type": "application/json"
        }
    },
    "k": 1,
    "domain": "k-pop",
    "testset_index_name": "ai_odyssey_demo_testset-000001",
    "evaluation_index_name": "ai_odyssey_demo_evaluation-000001"
}'

Response

Success (200)

{
    "status": "ok",
    "message": {
        "data": [
            {
                "question": " Which South Korean boy band was formed through Mnet's reality competition program Boys Planet and debuted with commercial success in 2023?",
                "ground_truth": "Zerobaseone",
                "answer": " Zerobaseone is a South Korean boy band formed through Mnet's reality competition program Boys Planet and debuted with commercial success in 2023, selling over two million units and peaking number one on South Korean Circle Album Chart. The group consists of nine members: Kim Ji-woong, Zhang Hao, Sung Han-bin, Seok Matthew, Kim Tae-rae, Ricky, Kim Gyu-vin, Park Gun-wook, and Han Yu-jin.",
                "contexts": [
                    "Zerobaseone (RR: Jerobeiseuwon; stylized in all caps; abbreviated as ZB1) is a South Korean boy band formed through Mnet's reality competition program Boys Planet and managed by WakeOne. The group consists of nine members: Kim Ji-woong, Zhang Hao, Sung Han-bin, Seok Matthew, Kim Tae-rae, Ricky, Kim Gyu-vin, Park Gun-wook, and Han Yu-jin. ..."
                ],
                "metadata": {
                    "source": "https://test-example.com/10",
                    "name": "k-pop singer",
                    "domain": "k-pop",
                    "author": "Michael",
                    "referenceContexts": [
                        "Zerobaseone (RR: Jerobeiseuwon; stylized in all caps; abbreviated as ZB1) is a South Korean boy band formed through Mnet's reality competition program Boys Planet and managed by WakeOne. The group consists of nine members: Kim Ji-woong, Zhang Hao, Sung Han-bin, Seok Matthew, Kim Tae-rae, Ricky, Kim Gyu-vin, Park Gun-wook, and Han Yu-jin. ..."
                    ],
                    "generatorLLM": "/home/[user-id]/.cache/instructlab/models/mistral-7b-instruct-v0.2.Q4_K_M.gguf",
                    "criticLLM": "/home/[user-id]/.cache/instructlab/models/mistral-7b-instruct-v0.2.Q4_K_M.gguf",
                    "timestamp": 1732264019.01063
                },
                "scores": {
                    "faithfulness": 0.9163240208552347,
                    "answer_relevancy": 0.8363240208552298,
                    "context_precision": -1.0,
                    "context_recall": 0.7393248506554236,
                    "context_entity_recall": 0.9999999900000002,
                    "noise_sensitivity_relevant": 0.0,
                    "answer_similarity": 0.8493240208552298
                }
            },
            ...
        ],
        "target_model": "/home/[user-id]/.cache/instructlab/models/mistral-7b-instruct-v0.2.Q4_K_M.gguf",
        "critic_model": "/home/[user-id]/.cache/instructlab/models/mistral-7b-instruct-v0.2.Q4_K_M.gguf"
    }
}

Failure (500)

{
    "TID":"175ef959-d97f-4641-a4b0-5b78fb1c021b",
    "error code":41999,
    "message":"Evaluation Fail!",
    "detail":"evaluation exception"
}

Files

rest_api.md

Latest commit

History

rest_api.md

File metadata and controls

RHEL AI Pro Suite APIs

POST /v1/documents/texts

Request

Body

Example

Response

Success (200)

Failure (500)

POST /v1/documents/files

Request

Parameter

Body

Example

Response

Success (200)

Failure (500)

POST /v1/qna/generate

Request

Body

Example

Response:

Success (200)

Failure (500)

PATCH /v1/documents

Request

Parameter

Body

Example

Response

Success (200)

Failure (500)

POST /v1/qna/evaluate

Request

Body

Example

Response

Success (200)

Failure (500)

POST `/v1/documents/texts`

POST `/v1/documents/files`

POST `/v1/qna/generate`

PATCH `/v1/documents`

POST `/v1/qna/evaluate`