AutoRAG

https://marker-inc-korea.github.io/AutoRAG
Admin
AutoRAG is a RAG AutoML tool for automatically finding an optimal RAG pipeline for your data.
Tokens:98,387
Snippets:1,154
Trust Score:8.3
Update:3 months ago
Show doc for...
Tokens:
Raw
### Start vLLM API Server

Source: https://marker-inc-korea.github.io/AutoRAG/_sources/nodes/generator/vllm_api.md.txt

Command to launch the vLLM API server on a machine with vLLM installed.

```bash
vllm serve Qwen/Qwen2.5-14B-Instruct-AWQ -q awq --port 8012
```

--------------------------------

### Install all dependencies from source

Source: https://marker-inc-korea.github.io/AutoRAG/_sources/install.md.txt

Install all optional dependencies when developing from source.

```bash
uv pip install -e .[all]
```

--------------------------------

### Install development and testing dependencies

Source: https://marker-inc-korea.github.io/AutoRAG/_sources/install.md.txt

Install requirements for running tests and building documentation.

```bash
pip install -r tests/requirements.txt
pip install -r docs/requirements.txt
```

--------------------------------

### Full AutoRAG YAML Configuration

Source: https://marker-inc-korea.github.io/AutoRAG/_sources/integration/vectordb/pinecone.md.txt

Complete configuration example including vector database setup, retrieval node lines, and generation modules.

```yaml
vectordb:
  - name: openai_pinecone
    db_type: pinecone
    embedding_model: openai_embed_3_large
    index_name: openai_embed_3_large
    api_key: ${PINECONE_API_KEY}
    similarity_metric: cosine
    dimension: 1536
node_lines:
- node_line_name: retrieve_node_line  # Arbitrary node line name
  nodes:
    - node_type: semantic_retrieval
      strategy:
        metrics: [retrieval_f1, retrieval_recall, retrieval_precision]
      top_k: 3
      modules:
        - module_type: vectordb
          vectordb: openai_pinecone
- node_line_name: post_retrieve_node_line  # Arbitrary node line name
  nodes:
    - node_type: prompt_maker
      strategy:
        metrics: [bleu, meteor, rouge]
      modules:
        - module_type: fstring
          prompt: "Read the passages and answer the given question. \n Question: {query} \n Passage: {retrieved_contents} \n Answer : "
    - node_type: generator
      strategy:
        metrics: [bleu, rouge]
      modules:
        - module_type: llama_index_llm
          llm: openai
          model: [ gpt-4o-mini ]
```

--------------------------------

### Install dependencies

Source: https://marker-inc-korea.github.io/AutoRAG/_sources/deploy/api_endpoint.md.txt

Install the required requests library for Python.

```bash
pip install requests
```

--------------------------------

### Install build dependencies

Source: https://marker-inc-korea.github.io/AutoRAG/_sources/troubleshooting.md.txt

Install necessary C++ build tools for Linux or macOS environments.

```bash
sudo apt-get install build-essential # linux
brew install gcc # mac
```

--------------------------------

### Initialize Chroma with YAML

Source: https://marker-inc-korea.github.io/AutoRAG/integration/vectordb/chroma.html

Example structure for configuring the Chroma vector store using a YAML file.

```yaml
vectordb:
  - name: chroma_default
    db_type: chroma
    client_type: persistent
    embedding_model: mock
    collection_name: openai
    path: ${PROJECT_DIR}/resources/chroma
```

--------------------------------

### Install GPU support

Source: https://marker-inc-korea.github.io/AutoRAG/_sources/install.md.txt

Install dependencies required for running local models on GPU.

```bash
pip install "AutoRAG[gpu]"
```

--------------------------------

### Install full feature dependencies

Source: https://marker-inc-korea.github.io/AutoRAG/_sources/install.md.txt

Recommended additional libraries for full functionality, including NLTK data downloads.

```bash
pip install --upgrade pyOpenSSL
pip install nltk
python3 -c "import nltk; nltk.download('punkt_tab')"
python3 -c "import nltk; nltk.download('averaged_perceptron_tagger_eng')"
```

--------------------------------

### Start Ollama Server

Source: https://marker-inc-korea.github.io/AutoRAG/_sources/integration/llm/ollama.md.txt

Command to initiate the Ollama server process.

```bash
ollama serve
```

--------------------------------

### Install Japanese language support

Source: https://marker-inc-korea.github.io/AutoRAG/_sources/install.md.txt

Install dependencies for Japanese language processing.

```bash
pip install "AutoRAG[ja]"
```

--------------------------------

### Install vLLM Dependencies

Source: https://marker-inc-korea.github.io/AutoRAG/_sources/local_model.md.txt

Commands to install the necessary extras for vLLM support in AutoRAG using pip or uv.

```bash
pip install "AutoRAG[vllm]"
```

```bash
uv pip install "AutoRAG[vllm]"
```

```bash
pip install "AutoRAG[gpu]"
```

```bash
uv pip install -e .[all]
```

--------------------------------

### Retrieve Documents Example Request

Source: https://marker-inc-korea.github.io/AutoRAG/_sources/deploy/api_endpoint.md.txt

An example of a query request sent to the /v1/retrieve endpoint.

```json
{
  "query": "latest trends in AI"
}
```

--------------------------------

### Execute Chunking Process

Source: https://marker-inc-korea.github.io/AutoRAG/_sources/data_creation/chunk/chunk.md.txt

Start the chunking process using a specified configuration file.

```python
chunker.start_chunking("your/path/to/chunk_config.yaml")
```

--------------------------------

### Parser Initialization and Execution

Source: https://marker-inc-korea.github.io/AutoRAG/data_creation/parse/parse.html

How to initialize the Parser instance and start the parsing process using a configuration file.

```APIDOC
## Parser(data_path_glob)

### Description
Initializes the Parser instance to prepare for document parsing.

### Parameters
- **data_path_glob** (string) - Required - A glob pattern specifying the path of the documents to be parsed (e.g., "path/to/files/*.pdf").

## start_parsing(config_path)

### Description
Starts the parsing process based on the provided YAML configuration file.

### Parameters
- **config_path** (string) - Required - The file path to the YAML configuration file defining the parsing modules.
```

--------------------------------

### Configure nodes with strategies

Source: https://marker-inc-korea.github.io/AutoRAG/_sources/optimization/custom_config.md.txt

Example of setting node types, top_k parameters, and evaluation strategies within a node line.

```yaml
node_lines:
  - node_line_name: node_line_1
    nodes:
      - node_type: semantic_retrieval
        top_k: 10
        strategy:
          metrics: [bleu, meteor, rouge]
          speed_threshold: 10
```

--------------------------------

### Build AutoRAG from source

Source: https://marker-inc-korea.github.io/AutoRAG/_sources/install.md.txt

Clone the repository and install in editable mode.

```bash
git clone https://github.com/Marker-Inc-Korea/AutoRAG.git
```

```bash
cd AutoRAG
pip install -e .
```

--------------------------------

### Configure Parsing for All Files

Source: https://marker-inc-korea.github.io/AutoRAG/_sources/data_creation/parse/parse.md.txt

Example YAML configuration for applying multiple parsing modules to all file types.

```yaml
modules:
  - module_type: langchain_parse
    file_type: all_files
    parse_method: [ directory, unstructured, upstagedocumentparse ]
  - module_type: clova
    file_type: all_files
    table_detection: true
  - module_type: llamaparse
    file_type: all_files
    result_type: markdown
    language: ko
    use_vendor_multimodal_model: true
    vendor_multimodal_model_name: openai-gpt-4o-mini
```

--------------------------------

### Run Weaviate via Docker

Source: https://marker-inc-korea.github.io/AutoRAG/_sources/integration/vectordb/weaviate.md.txt

Command to start a local Weaviate instance using Docker.

```bash
docker run -p 8080:8080 -p 50051:50051 cr.weaviate.io/semitechnologies/weaviate:1.27.3
```

--------------------------------

### Full Cohere Reranker Configuration Example

Source: https://marker-inc-korea.github.io/AutoRAG/nodes/passage_reranker/cohere.html

A complete example of the cohere_reranker module configuration including batch size and model selection.

```yaml
- module_type: cohere_reranker
  api_key: your_cohere_api_key
  batch: 64
  model: rerank-multilingual-v2.0
```

--------------------------------

### Configure OpenAI LLM with GPT-5

Source: https://marker-inc-korea.github.io/AutoRAG/_sources/nodes/generator/openai_llm.md.txt

Example configuration for using GPT-5 models with specific reasoning effort and verbosity settings.

```yaml
modules:
  - module_type: openai_llm
    llm: [ gpt-5.1 ]
    max_tokens: 512
    reasoning:
      effort: high
      verbosity: low
```

--------------------------------

### Full QA Creation Pipeline

Source: https://marker-inc-korea.github.io/AutoRAG/_sources/data_creation/qa_creation/qa_creation.md.txt

A complete example demonstrating the end-to-end process of generating QA data from an initial corpus.

```python
from llama_index.llms.openai import OpenAI

from autorag.data.qa.filter.dontknow import dontknow_filter_rule_based
from autorag.data.qa.generation_gt.llama_index_gen_gt import (
    make_basic_gen_gt,
    make_concise_gen_gt,
)
from autorag.data.qa.query.llama_gen_query import factoid_query_gen
from autorag.data.qa.sample import random_single_hop

llm = OpenAI()
initial_corpus = initial_raw.chunk(
    "llama_index_chunk", chunk_method="token", chunk_size=128, chunk_overlap=5
)
initial_qa = (
    initial_corpus.sample(random_single_hop, n=3)
    .map(
        lambda df: df.reset_index(drop=True),
    )
    .make_retrieval_gt_contents()
    .batch_apply(
        factoid_query_gen,  # query generation
        llm=llm,
    )
    .batch_apply(
        make_basic_gen_gt,  # answer generation (basic)
        llm=llm,
    )
    .batch_apply(
        make_concise_gen_gt,  # answer generation (concise)
        llm=llm,
    )
    .filter(
        dontknow_filter_rule_based,  # filter don't know
        lang="en",
    )
)

initial_qa.to_parquet('./qa.parquet', './corpus.parquet')
```

--------------------------------

### Install AutoRAG via pip

Source: https://marker-inc-korea.github.io/AutoRAG/_sources/install.md.txt

Standard installation command for the AutoRAG package.

```bash
pip install AutoRAG
```

--------------------------------

### Install Korean language support

Source: https://marker-inc-korea.github.io/AutoRAG/_sources/install.md.txt

Install dependencies for Korean language processing. Requires JDK 17 for konlpy.

```bash
pip install "AutoRAG[ko]"
```

--------------------------------

### Initialize and run API server with Python

Source: https://marker-inc-korea.github.io/AutoRAG/_sources/deploy/api_endpoint.md.txt

Use ApiRunner to start the server from a YAML configuration or a trial directory. Requires nest_asyncio to handle event loops in certain environments.

```python
from autorag.deploy import ApiRunner
import nest_asyncio

nest_asyncio.apply()

runner = ApiRunner.from_yaml('your/path/to/pipeline.yaml', project_dir='your/project/directory')
runner.run_api_server()
```

```python
from autorag.deploy import ApiRunner
import nest_asyncio

nest_asyncio.apply()

runner = ApiRunner.from_trial_folder('/your/path/to/trial_dir')
runner.run_api_server()
```

--------------------------------

### Example AutoRAG Configuration

Source: https://marker-inc-korea.github.io/AutoRAG/_sources/nodes/passage_reranker/mixedbreadai_reranker.md.txt

A basic configuration snippet for the Mixedbread AI reranker module.

```yaml
modules:
  - module_type: mixedbreadai_reranker
```

--------------------------------

### Install vLLM dependencies

Source: https://marker-inc-korea.github.io/AutoRAG/_sources/install.md.txt

Optional dependencies for using vLLM as a generator or embedding backend.

```bash
# pip
pip install "AutoRAG[vllm]"

# uv (recommended)
uv pip install "AutoRAG[vllm]"
```

--------------------------------

### FlashRank Reranker Configuration

Source: https://marker-inc-korea.github.io/AutoRAG/_sources/nodes/passage_reranker/flashrank_reranker.md.txt

Example configuration for the flashrank_reranker module in a YAML file.

```yaml
- module_type: flashrank_reranker
  batch: 32
  model: "ms-marco-MiniLM-L-12-v2"
```

--------------------------------

### Install parsing dependencies

Source: https://marker-inc-korea.github.io/AutoRAG/_sources/install.md.txt

Install dependencies for document parsing. Requires system-level packages like libmagic, tesseract, and poppler.

```bash
pip install "AutoRAG[parse]"
```

--------------------------------

### Full Pipeline Configuration with Milvus

Source: https://marker-inc-korea.github.io/AutoRAG/_sources/integration/vectordb/milvus.md.txt

Example of a complete YAML configuration integrating Milvus with retrieval and generation nodes.

```yaml
vectordb:
  - name: openai_milvus
    db_type: milvus
    embedding_model: openai_embed_3_large
    collection_name: openai_embed_3_large
    uri: ${MILVUS_URI}
    token: ${MILVUS_TOKEN}
    embedding_batch: 50
    similarity_metric: cosine
    index_type: IVF_FLAT
    params:
      nlist: 16384
node_lines:
- node_line_name: retrieve_node_line  # Arbitrary node line name
  nodes:
    - node_type: semantic_retrieval
      strategy:
        metrics: [retrieval_f1, retrieval_recall, retrieval_precision]
      top_k: 3
      modules:
        - module_type: vectordb
          vectordb: openai_milvus
- node_line_name: post_retrieve_node_line  # Arbitrary node line name
  nodes:
    - node_type: prompt_maker
      strategy:
        metrics: [bleu, meteor, rouge]
      modules:
        - module_type: fstring
          prompt: "Read the passages and answer the given question. \n Question: {query} \n Passage: {retrieved_contents} \n Answer : "
    - node_type: generator
      strategy:
        metrics: [bleu, rouge]
      modules:
        - module_type: llama_index_llm
          llm: openai
          model: [ gpt-4o-mini ]
```

--------------------------------

### API curl commands

Source: https://marker-inc-korea.github.io/AutoRAG/_sources/deploy/api_endpoint.md.txt

Command line interface examples for interacting with various API endpoints.

```bash
curl -X POST "http://example.com/v1/run" \
     -H "Content-Type: application/json" \
     -d '{"query": "example query", "result_column": "generated_texts"}'
```

```bash
curl -X POST "http://example.com/v1/stream" \
     -H "Content-Type: application/json" \
     -d '{"query": "example query", "result_column": "generated_texts"}' \
     --no-buffer
```

```bash
curl -X POST "http://example.com/v1/retrieve" \
     -H "Content-Type: application/json" \
     -d '{"query": "example query"}'
```

```bash
curl -X GET "http://example.com/version"
```

--------------------------------

### Configure v0.3.17 Retrieval Nodes

Source: https://marker-inc-korea.github.io/AutoRAG/_sources/migration.md.txt

Example YAML configuration demonstrating the split of retrieval nodes into lexical, semantic, and hybrid types with required strategy and top_k parameters.

```yaml
vectordb:
  - name: openai_embed_3_small
    db_type: chroma
    client_type: persistent
    embedding_model: openai_embed_3_small
    collection_name: openai_embed_3_small
    path: ${PROJECT_DIR}/resources/chroma
  - name: openai_embed_3_large
    db_type: chroma
    client_type: persistent
    embedding_model: openai_embed_3_large
    collection_name: openai_embed_3_large
    path: ${PROJECT_DIR}/resources/chroma
    embedding_batch: 50
node_lines:
- node_line_name: retrieve_node_line
  nodes:
    - node_type: lexical_retrieval # Lexical retrieval node type
      strategy:
        metrics: [retrieval_f1, retrieval_recall]
      top_k: 10
      modules:
        - module_type: bm25
          bm25_tokenizer: [ facebook/opt-125m, porter_stemmer ]
    - node_type: semantic_retrieval # semantic retrieval node type
      strategy:
        metrics: [retrieval_f1, retrieval_recall]
      top_k: 10
      modules:
        - module_type: vectordb
          vectordb: [openai_embed_3_large, openai_embed_3_small]
    - node_type: hybrid_retrieval # Hybrid retrieval node type
      strategy:
        metrics: [retrieval_f1, retrieval_recall]
      top_k: 10
      modules:
        - module_type: hybrid_rrf
          weight_range: (4, 30)
        - module_type: hybrid_cc
          normalize_method: [ mm, tmm, z, dbsf ]
          weight_range: (0.0, 1.0)
          test_weight_size: 51
```

--------------------------------

### Execute parsing pipeline

Source: https://marker-inc-korea.github.io/AutoRAG/_sources/data_creation/parse/parse.md.txt

Start the parsing process by passing the path of the configuration YAML file to the start_parsing method.

```python
parser.start_parsing("your/path/to/parse_config.yaml")
```

--------------------------------

### Configure OpenAI LLM in config.yaml

Source: https://marker-inc-korea.github.io/AutoRAG/_sources/nodes/generator/openai_llm.md.txt

Standard configuration example for defining multiple OpenAI models and temperature settings in an AutoRAG config file.

```yaml
modules:
  - module_type: openai_llm
    llm: [ gpt-3.5-turbo, gpt-4-turbo-2024-04-09 ]
    temperature: [ 0.1, 1.0 ]
    max_tokens: 512
```

--------------------------------

### Upgrade pip for installation

Source: https://marker-inc-korea.github.io/AutoRAG/_sources/troubleshooting.md.txt

Use this command to ensure pip is up to date when encountering build errors.

```bash
pip install --upgrade pip
```

--------------------------------

### Full AutoRAG Pipeline Configuration with Qdrant

Source: https://marker-inc-korea.github.io/AutoRAG/_sources/integration/vectordb/qdrant.md.txt

A complete example of a YAML configuration file integrating Qdrant with retrieval and generation nodes.

```yaml
vectordb:
  - name: openai_qdrant
    db_type: qdrant
    embedding_model: openai_embed_3_large
    collection_name: openai_embed_3_large
    client_type: docker
    embedding_batch: 50
    similarity_metric: cosine
    dimension: 1536
node_lines:
- node_line_name: retrieve_node_line  # Arbitrary node line name
  nodes:
    - node_type: semantic_retrieval
      strategy:
        metrics: [retrieval_f1, retrieval_recall, retrieval_precision]
      top_k: 3
      modules:
        - module_type: vectordb
          vectordb: openai_qdrant
- node_line_name: post_retrieve_node_line  # Arbitrary node line name
  nodes:
    - node_type: prompt_maker
      strategy:
        metrics: [bleu, meteor, rouge]
      modules:
        - module_type: fstring
          prompt: "Read the passages and answer the given question. \n Question: {query} \n Passage: {retrieved_contents} \n Answer : "
    - node_type: generator
      strategy:
        metrics: [bleu, rouge]
      modules:
        - module_type: llama_index_llm
          llm: openai
          model: [ gpt-4o-mini ]
```

--------------------------------

### Configure Passage Compressor in config.yaml

Source: https://marker-inc-korea.github.io/AutoRAG/_sources/nodes/passage_compressor/passage_compressor.md.txt

Example configuration for a passage compressor node, including strategy metrics and module selection.

```yaml
node_lines:
- node_line_name: retrieve_node_line  # Arbitrary node line name
  nodes:
  - node_type: passage_compressor
    strategy:
      metrics: [retrieval_token_f1, retrieval_token_recall, retrieval_token_precision]
      speed_threshold: 10
    modules:
      - module_type: pass_compressor
      - module_type: tree_summarize
        llm: openai
        model: gpt-3.5-turbo-16k
```

--------------------------------

### Use Default Vector Database Configuration

Source: https://marker-inc-korea.github.io/AutoRAG/_sources/integration/vectordb/vectordb.md.txt

Specify the default Chroma configuration in the YAML modules section for quick setup.

```yaml
modules:
  - module_type: vectordb
    vectordb: default
```

--------------------------------

### AutoRAG Configuration YAML

Source: https://marker-inc-korea.github.io/AutoRAG/_sources/integration/llm/ollama.md.txt

Example configuration file for an AutoRAG experiment using vector retrieval and an Ollama-based generator.

```yaml
vectordb:
  - name: mpnet_base_chroma
    db_type: chroma
    client_type: persistent
    embedding_model: huggingface_all_mpnet_base_v2
    collection_name: huggingface_all_mpnet_base_v2
    path: ${PROJECT_DIR}/data/chroma
node_lines:
  - node_line_name: retrieve_node_line
    nodes:
      - node_type: semantic_retrieval
        strategy:
          metrics: [ retrieval_f1, retrieval_recall, retrieval_precision ]
        top_k: 3
        modules:
          - module_type: vectordb
            vectordb: mpnet_base_chroma
  - node_line_name: post_retrieve_node_line
    nodes:
      - node_type: prompt_maker
        strategy:
          metrics: [ meteor, rouge, bert_score ]
        modules:
          - module_type: fstring
            prompt: "Read the passages and answer the given question. \n Question: {query} \n Passage: {retrieved_contents} \n Answer : "
      - node_type: generator
        strategy:
          metrics: [ meteor, rouge, bert_score ]
        modules:
          - module_type: llama_index_llm
            llm: ollama
            model: llama3
            temperature: [ 0.1, 0.5, 1.0 ]
            batch: 1
```

--------------------------------

### Configure Parsing for Specific File Types

Source: https://marker-inc-korea.github.io/AutoRAG/_sources/data_creation/parse/parse.md.txt

Example YAML configuration defining explicit parsing methods for different file extensions.

```yaml
modules:
  # PDF
  - module_type: langchain_parse
    file_type: pdf
    parse_method: pdfminer
  # CSV
  - module_type: langchain_parse
    file_type: csv
    parse_method: csv
  # JSON
  - module_type: langchain_parse
    file_type: json
    parse_method: json
    jq_schema: .content
  # Markdown
  - module_type: langchain_parse
    file_type: md
    parse_method: unstructuredmarkdown
  # HTML
  - module_type: langchain_parse
    file_type: html
    parse_method: bshtml
  # XML
  - module_type: langchain_parse
    file_type: xml
    parse_method: unstructuredxml
```

--------------------------------

### Define Modular RAG Pipeline in YAML

Source: https://marker-inc-korea.github.io/AutoRAG/_sources/roadmap/modular_rag.md.txt

Example configuration showing how to structure node lines, policy nodes, and merger nodes to create a cyclic RAG pipeline.

```yaml
node_lines:
  - node_line_name: node_line_1
    nodes:
      - node_type: merger
        pass_value: query
      - node_type: retrieval
        strategy:
          metrics: [retrieval_recall]
        top_k: 10
        modules:
          - module_type: bm25
      - node_type: passage_reranker
        strategy:
          metrics: [reranker_recall]
        top_k: 3
        modules:
          - module_type: upr
      - node_type: policy
        target_node_lines:
          - node_line_1
          - node_line_2
        modules:
          - module_type: kf1_policy
          - module_type: rl_policy
  - node_line_name: node_line_2
    nodes:
      - node_type: prompt_maker
        modules:
          - module_type: fstring
            prompt: "Answer the question: {query} \n\n Passage: {retrieved_contents}"
      - node_type: generator
        strategy:
          metrics: [bleu, rouge, meteor]
        modules:
          - module_type: llama_index_llm
            llm: openai
```

--------------------------------

### Get Search String Indices

Source: https://marker-inc-korea.github.io/AutoRAG/_modules/autorag/data/utils/util.html

Finds the start and end indices of a search string within a text. Returns (0, 0) if not found.

```python
def get_start_end_idx(original_text: str, search_str: str) -> Tuple[int, int]:
	start_idx = original_text.find(search_str)
	if start_idx == -1:
		return 0, 0
	end_idx = start_idx + len(search_str)
	return start_idx, end_idx - 1
```

--------------------------------

### Initialize and Run Parser

Source: https://marker-inc-korea.github.io/AutoRAG/_sources/data_creation/parse/parse.md.txt

Instantiate the Parser with a data path glob and execute the parsing process using a configuration file.

```python
from autorag.parser import Parser

parser = Parser(data_path_glob="your/data/path/*")
parser.start_parsing("your/path/to/parse_config.yaml")
```

--------------------------------

### GET /version

Source: https://marker-inc-korea.github.io/AutoRAG/_sources/deploy/api_endpoint.md.txt

Get the current version of the API.

```APIDOC
## GET /version

### Description
Returns the current version of the API as a string.

### Method
GET

### Endpoint
/version

### Response
#### Success Response (200)
- **version** (string) - The version of the API.
```

--------------------------------

### Chroma.__init__

Source: https://marker-inc-korea.github.io/AutoRAG/_sources/integration/vectordb/chroma.md.txt

Initializes the Chroma vector store client with specific configuration parameters.

```APIDOC
## Chroma.__init__

### Description
Initializes a new instance of the Chroma vector store. This class handles the connection to the database and manages embedding storage and retrieval.

### Signature
`Chroma(embedding_model: str, collection_name: str, embedding_batch: int = 100, client_type: str = "persistent", similarity_metric: str = "cosine", path: str = None, host: str = "localhost", port: int = 8000, ssl: bool = False, headers: Optional[Dict[str, str]] = None, api_key: Optional[str] = None, tenant: str = DEFAULT_TENANT, database: str = DEFAULT_DATABASE)`

### Parameters
- **embedding_model** (str) - Required - The name or identifier of the embedding model to use.
- **collection_name** (str) - Required - The name of the collection to store embeddings.
- **embedding_batch** (int) - Optional - Batch size for embedding operations (default: 100).
- **client_type** (str) - Optional - The type of client to use: "ephemeral", "persistent", "http", or "cloud" (default: "persistent").
- **similarity_metric** (str) - Optional - The metric used for similarity calculations: "cosine", "ip", or "l2" (default: "cosine").
- **path** (str) - Optional - The path for persistent storage (required for persistent client).
- **host** (str) - Optional - Host address for HTTP client (default: "localhost").
- **port** (int) - Optional - Port number for HTTP client (default: 8000).
- **ssl** (bool) - Optional - Enable SSL for HTTP client (default: False).
- **headers** (Dict[str, str]) - Optional - Custom headers for HTTP client.
- **api_key** (str) - Optional - API key for cloud client.
- **tenant** (str) - Optional - Tenant identifier.
- **database** (str) - Optional - Database identifier.
```

--------------------------------

### Install AutoRAG

Source: https://marker-inc-korea.github.io/AutoRAG/_sources/integration/llm/nvidia_nim.md.txt

Install the AutoRAG package via pip.

```bash
pip install autorag
```

--------------------------------

### Run Web Interface via CLI

Source: https://marker-inc-korea.github.io/AutoRAG/_sources/deploy/web.md.txt

Commands to launch the web interface using a YAML configuration file or a trial path, with optional project directory specification.

```bash
autorag run_web --yaml_path your/path/to/pipeline.yaml
```

```bash
autorag run_web --yaml_path your/path/to/pipeline.yaml --project_dir your/project/directory
```

```bash
autorag run_web --trial_path your/path/to/trial
```

--------------------------------

### Initialize Parser instance

Source: https://marker-inc-korea.github.io/AutoRAG/_sources/data_creation/parse/parse.md.txt

Create a Parser instance by providing a glob pattern to locate the documents for parsing.

```python
from autorag.parser import Parser

parser = Parser(data_path_glob="your/data/path/*")
```

--------------------------------

### GET /version

Source: https://marker-inc-korea.github.io/AutoRAG/_sources/deploy/api_endpoint.md.txt

Retrieves the current API version.

```APIDOC
## GET /version

### Description
Returns the current version of the API.

### Method
GET

### Endpoint
/version
```

--------------------------------

### Execute Document Parsing

Source: https://marker-inc-korea.github.io/AutoRAG/_sources/data_creation/tutorial.md.txt

Initialize the Parser and execute the parsing process using a configuration file.

```python
from autorag.parser import Parser

parser = Parser(data_path_glob="your/data/path/*", project_dir="./parse_project_dir")
parser.start_parsing("your/path/to/parse_config.yaml")
```

--------------------------------

### POST /v1/run

Source: https://marker-inc-korea.github.io/AutoRAG/_sources/deploy/api_endpoint.md.txt

Run a query and get generated text with retrieved passages.

```APIDOC
## POST /v1/run

### Description
Run a query and get generated text with retrieved passages.

### Method
POST

### Endpoint
/v1/run

### Request Body
- **query** (string) - Required - The query string.
- **result_column** (string) - Optional - The result column name. Default is generated_texts.

### Response
#### Success Response (200)
- **result** (string or array of strings) - The result text or list of texts.
- **retrieved_passage** (array of objects) - List of retrieved passages.
```

--------------------------------

### Configure UPR in config.yaml

Source: https://marker-inc-korea.github.io/AutoRAG/_sources/nodes/passage_reranker/upr.md.txt

Example configuration for the UPR module within an AutoRAG pipeline.

```yaml
modules:
  - module_type: upr
    use_bf16: False
    prefix_prompt: "Passage: "
    suffix_prompt: "Please write a question based on this passage."
```

--------------------------------

### Generate QA with Existing Queries

Source: https://marker-inc-korea.github.io/AutoRAG/_modules/autorag/data/legacy/qacreation/base.html

Initializes the creation of a single-hop QA dataset by utilizing existing query data and an optional answer creation function.

```python
def make_qa_with_existing_qa(
	corpus_df: pd.DataFrame,
	existing_query_df: pd.DataFrame,
	content_size: int,
	answer_creation_func: Optional[Callable] = None,
	exist_gen_gt: Optional[bool] = False,
	output_filepath: Optional[str] = None,
	embedding_model: str = "openai_embed_3_large",
	collection: Optional[chromadb.Collection] = None,
	upsert: bool = False,
	random_state: int = 42,
	cache_batch: int = 32,
	top_k: int = 3,
	**kwargs,
) -> pd.DataFrame:
	"""
	Make single-hop QA dataset using given qa_creation_func and existing queries.


```

--------------------------------

### Get Parameter Combinations

Source: https://marker-inc-korea.github.io/AutoRAG/_modules/autorag/data/utils/util.html

Generates combinations of module parameters based on a list of module definitions.

```python
def get_param_combinations(modules: List[Dict]) -> Tuple[List[Callable], List[Dict]]:
	module_callable_list, module_params_list = [], []
	for module in modules:
		module_instance = Module.from_dict(module)
		module_params_list.append(module_instance.module_param)
		module_callable_list.append(module_instance.module)

	combinations = list(map(make_combinations, module_params_list))
	module_list, combination_list = explode(module_callable_list, combinations)
	return module_list, combination_list
```

--------------------------------

### Execute Batch Processing

Source: https://marker-inc-korea.github.io/AutoRAG/_modules/autorag/data/legacy/qacreation/llama_index.html

Example of running asynchronous tasks in a batch using the event loop.

```python
	loops = get_event_loop()
	results = loops.run_until_complete(process_batch(tasks, batch))

	return results
```

--------------------------------

### Configure BM25 in AutoRAG

Source: https://marker-inc-korea.github.io/AutoRAG/_sources/nodes/retrieval/bm25.md.txt

Example configuration for the BM25 module showing multiple supported tokenizer options.

```yaml
modules:
  - module_type: bm25
    bm25_tokenizer: [ porter_stemmer, ko_kiwi, space, gpt2, ko_kkma, ko_okt, sudachipy ]
```

--------------------------------

### Deploy RAG Pipeline as Web Interface

Source: https://marker-inc-korea.github.io/AutoRAG/_sources/tutorial.md.txt

Launch a web interface for the pipeline using Gradio or the CLI.

```python
from autorag.deploy import GradioRunner

runner = GradioRunner.from_yaml('your/path/to/pipeline.yaml')
runner.run_web()
```

```python
from autorag.deploy import GradioRunner

runner = GradioRunner.from_trial_folder('/your/path/to/trial_dir')
runner.run_web()
```

```bash
autorag run_web --yaml_path your/path/to/pipeline.yaml
```

```bash
autorag run_web --trial_path your/path/to/trial_dir
```

--------------------------------

### Configure Recency Filter in YAML

Source: https://marker-inc-korea.github.io/AutoRAG/_sources/nodes/passage_filter/recency_filter.md.txt

Example configuration for the recency_filter module within an AutoRAG config.yaml file.

```yaml
modules:
  - module_type: recency_filter
    threshold_datetime: 2015-01-01
```

--------------------------------

### Configure Weaviate Docker in YAML

Source: https://marker-inc-korea.github.io/AutoRAG/_sources/integration/vectordb/weaviate.md.txt

YAML configuration snippets for using a local Docker-based Weaviate instance.

```yaml
- name: openai_weaviate
  db_type: weaviate
  embedding_model: openai_embed_3_large
  collection_name: openai_embed_3_large
  client_type: docker
  host: localhost
  port: 8080
  grpc_port: 50051
  embedding_batch: 50
  similarity_metric: cosine
  text_key: content
```

```yaml
vectordb:
  - name: openai_weaviate
    db_type: weaviate
    embedding_model: openai_embed_3_large
    collection_name: openai_embed_3_large
    client_type: docker
    host: localhost
    port: 8080
    grpc_port: 50051
    embedding_batch: 50
    similarity_metric: cosine
    text_key: content
node_lines:
- node_line_name: retrieve_node_line  # Arbitrary node line name
  nodes:
    - node_type: semantic_retrieval
      strategy:
        metrics: [retrieval_f1, retrieval_recall, retrieval_precision]
      top_k: 3
      modules:
        - module_type: vectordb
          vectordb: openai_weaviate
- node_line_name: post_retrieve_node_line  # Arbitrary node line name
  nodes:
    - node_type: prompt_maker
      strategy:
        metrics: [bleu, meteor, rouge]
      modules:
        - module_type: fstring
          prompt: "Read the passages and answer the given question. \n Question: {query} \n Passage: {retrieved_contents} \n Answer : "
    - node_type: generator
      strategy:
        metrics: [bleu, rouge]
      modules:
        - module_type: llama_index_llm
          llm: openai
          model: [ gpt-4o-mini ]
```

--------------------------------

### Configure vLLM API in config.yaml

Source: https://marker-inc-korea.github.io/AutoRAG/_sources/nodes/generator/vllm_api.md.txt

Example configuration for the vllm_api module within an AutoRAG config.yaml file.

```yaml
 - module_type: vllm_api
   uri: http://localhost:8012
   llm: Qwen/Qwen2.5-14B-Instruct-AWQ
   temperature: [0, 0.5]
   max_tokens: 400
```

--------------------------------

### Use environment variables in YAML

Source: https://marker-inc-korea.github.io/AutoRAG/_sources/optimization/custom_config.md.txt

Demonstrates how to inject environment variables into configuration values using the ${VAR} syntax.

```yaml
node_lines:
  - node_line_name: node_line_1
    nodes:
      - node_type: semantic_retrieval
        top_k: ${TOP-K}
        strategy:
          metrics: [sem_score]
        modules:
          - module_type: vectordb
            vectordb: default
```

--------------------------------

### Configure MiniMax LLM in AutoRAG

Source: https://marker-inc-korea.github.io/AutoRAG/_sources/nodes/generator/minimax_llm.md.txt

Example configuration for using MiniMax models within an AutoRAG pipeline.

```yaml
modules:
  - module_type: minimax_llm
    llm: [MiniMax-M2.7, MiniMax-M2.5-highspeed]
    temperature: [0.1, 0.5]
    max_tokens: 512
    api_key: ${MINIMAX_API_KEY}
```

--------------------------------

### Run API server via CLI

Source: https://marker-inc-korea.github.io/AutoRAG/_sources/deploy/api_endpoint.md.txt

Execute the API server directly from the terminal by specifying the trial directory, host, and port.

```bash
autorag run_api --trial_dir /trial/dir/0 --host 0.0.0.0 --port 8000
```
Tokens:
Raw