owned this note
owned this note
Published
Linked with GitHub
---
type: slide
title: RaidGuild x Gaia - RAG API Pipeline Demo
tags: RAG, LLMs, , data-pipeline
slideOptions:
keyboard: true
progress: true
slideNumber: true
showNotes: false
transition: 'fade'
theme: 'night'
---
# Gaia x RaiGuild
### RAG API Pipeline
#### Demo Day
---
# Agenda
1. Architecture Overview
2. Defining an API Pipeline Manifest
3. rag-api-pipeline - CLI Walkthrough
4. Demo - Boardroom API
5. Issues Found and TODOs
6. Next steps
---
## Architecture Overview
----
## Tech stack
- PyAirbyte
- Utilities for using Airbyte connectors in Python
- Declarative API Connectors using Airbyte CDK for low-code connector development
- Pathway
- Python data processing framework for creating AI pipelines over data streams
- Rust engine with multithreading and multiprocessing
----
## Tech stack
- Data caching at each stage of the pipeline
- Airbyte DuckDB Cache for API data extraction
- JSONL output connectors after normalization & chunking stages
----
## Tech stack
- Unstructured
- simplify the ingestion and pre-processing of diverse data formats on data workflows for LLM
- library includes functions to partition, chunk, clean, and stage raw source documents.
- Gaia node for Feature Embedding generation
- RAG API Server with OpenAI-like API
----
## Tech stack
- QdrantDB
- vector similarity search engine and vector database
- knowledge base snaphosts
---
## Defining an API Pipeline Manifest
----
## API Spec
----
## Base Data retriever
- Requester
- Authentication
- OAuth | ApiKeyAuthenticator | BearerAuthenticator | BasicHttpAuthenticator"
- Paginator
- NoPagination | CursorPagination | OffsetIncrement | PageIncrement
- Record selector
----
## Base Data retriever
----
## Base Response Schemas
----
## Endpoint streams definition
----
## Fields for text preprocessing
#### **Other fields will be added as metadata
----
## Chunking parameters
---
## rag-api-pipeline - CLI Walkthrough
---
## Demo - Boardroom API
- Aave data sample
- Recent Discourse posts and proposals
- [Public Gaia node with knowledge base](https://www.gaianet.ai/chat?subdomain=0x4bb75434096b3d087852c033c7714c02d3ab74dd.us.gaianet.network)
----
### rag-api-pipeline run-all
- `--full-refresh` option
----
### Input API manifest
```yaml
version: "3.0.0"
api_name: "boardroom_api"
api_parameters:
cname: "aave"
protocol: "aave"
spec:
type: Spec
documentation_url: https://docs.airbyte.com/integrations/sources/boardroom
connection_specification:
$schema: http://json-schema.org/draft-07/schema#
title: Boardroom API Spec
type: object
required:
- api_key
- cname
- protocol
additionalProperties: true
properties:
# 'TODO: This schema defines the configuration required for the source. This usually involves metadata such as database and/or authentication information.':
api_key:
type: string
description: >-
Boardroom API Key. See <a href="https://docs.boardroom.io/docs/api/05c1fb6d88a07-governance-api">here</a>
for details.
airbyte-secret: true
cname:
type: string
description: >-
Protocol ID a.k.a cname
examples:
- aave
- hopprotocol
protocol:
type: string
description: >-
Alias for cname. Needed as there are a few endpoints that use `protocol` as parameter id
examples:
- aave
- metacartel
definitions:
selector:
type: RecordSelector
extractor:
type: DpathExtractor
field_path: []
requester_base:
type: HttpRequester
url_base: "https://api.boardroom.info/v1"
http_method: "GET"
authenticator:
type: ApiKeyAuthenticator
api_token: "{{ config['api_key'] }}"
inject_into:
type: RequestOption
field_name: "key"
inject_into: request_parameter
retriever_base:
type: SimpleRetriever
record_selector:
$ref: "#/definitions/selector"
paginator:
type: DefaultPaginator
pagination_strategy:
type: CursorPagination
cursor_value: "{{ response.get('nextCursor', '') }}"
stop_condition: "{{ 'nextCursor' not in response }}"
page_token_option:
type: RequestOption
field_name: "cursor"
inject_into: "request_parameter"
schemas:
boardroom_api:
type: object
$schema: http://json-schema.org/draft-07/schema#
properties:
data:
type: array
nextCursor:
type:
- string
- "null"
boardroom_api_protocol:
type: object
$schema: http://json-schema.org/draft-07/schema#
properties:
data:
type: object
nextCursor:
type:
- "null"
api_config:
request_method: "get"
content_type: "application/json"
response_entrypoint_field: "data"
response_primary_key: "nextCursor"
endpoints:
/discourseTopics:
id: "discourseTopics"
responseSchema: "#/schemas/boardroom_api"
textSchema:
$ref: "#/textSchemas/DiscourseTopic"
/discourseCategories:
id: "discourseCategories"
responseSchema: "#/schemas/boardroom_api"
textSchema:
$ref: "#/textSchemas/DiscourseCategory"
/discourseTopicPosts:
id: "discourseTopicPosts"
responseSchema: "#/schemas/boardroom_api"
textSchema:
$ref: "#/textSchemas/DiscourseTopicPost"
"/protocols/{cname}":
id: "protocol"
responseSchema: "#/schemas/boardroom_api_protocol"
textSchema:
$ref: "#/textSchemas/Protocol"
"/protocols/{cname}/proposals":
id: "proposals"
responseSchema: "#/schemas/boardroom_api"
textSchema:
$ref: "#/textSchemas/Proposal"
textSchemas:
Protocol:
type: object
properties:
- cname:
type: string
- name:
type: string
- categories:
type: array
items:
type: string
Proposal:
type: object
properties:
- title:
type: string
- content:
type: string
- summary:
type: string
DiscourseTopic:
type: object
properties:
- title:
type: string
DiscourseCategory:
type: object
properties:
- name:
type: string
- description:
type: string
DiscourseTopicPost:
type: object
properties:
- body:
type: string
chunking_param:
mode: "elements"
chunking_strategy: "by_title"
include_orig_elements: true
max_characters: 1500
new_after_n_chars: 1024
overlap: 0
overlap_all: false
combine_text_under_n_chars: 0
multipage_sections: true
```
----
### Generated Source Connector
```yaml
check:
stream_names:
- discourseTopics
- discourseCategories
- discourseTopicPosts
- protocol
- proposals
type: CheckStream
definitions:
discourseCategories_stream:
$parameters:
name: discourseCategories
path: '"/discourseCategories"'
primary_key: nextCursor
retriever:
$ref: '#/definitions/retriever_base'
paginator:
$ref: '#/definitions/paginator'
requester:
$ref: '#/definitions/requester_base'
request_parameters:
protocol: '{{ config[''protocol''] }}'
schema_loader:
schema:
$ref: '#/schemas/boardroom_api'
type: InlineSchemaLoader
type: DeclarativeStream
discourseTopicPosts_stream:
$parameters:
name: discourseTopicPosts
path: '"/discourseTopicPosts"'
primary_key: nextCursor
retriever:
$ref: '#/definitions/retriever_base'
paginator:
$ref: '#/definitions/paginator'
requester:
$ref: '#/definitions/requester_base'
request_parameters:
protocol: '{{ config[''protocol''] }}'
schema_loader:
schema:
$ref: '#/schemas/boardroom_api'
type: InlineSchemaLoader
type: DeclarativeStream
discourseTopics_stream:
$parameters:
name: discourseTopics
path: '"/discourseTopics"'
primary_key: nextCursor
retriever:
$ref: '#/definitions/retriever_base'
paginator:
$ref: '#/definitions/paginator'
requester:
$ref: '#/definitions/requester_base'
request_parameters:
protocol: '{{ config[''protocol''] }}'
schema_loader:
schema:
$ref: '#/schemas/boardroom_api'
type: InlineSchemaLoader
type: DeclarativeStream
paginator:
page_token_option:
field_name: cursor
inject_into: request_parameter
type: RequestOption
pagination_strategy:
cursor_value: '{{ response.get(''nextCursor'', '''') }}'
stop_condition: '{{ ''nextCursor'' not in response }}'
type: CursorPagination
type: DefaultPaginator
proposals_stream:
$parameters:
name: proposals
path: '"/protocols/{{ config[''cname''] }}/proposals"'
primary_key: nextCursor
retriever:
$ref: '#/definitions/retriever_base'
paginator:
$ref: '#/definitions/paginator'
requester:
$ref: '#/definitions/requester_base'
schema_loader:
schema:
$ref: '#/schemas/boardroom_api'
type: InlineSchemaLoader
type: DeclarativeStream
protocol_stream:
$parameters:
name: protocol
path: '"/protocols/{{ config[''cname''] }}"'
primary_key: nextCursor
retriever:
$ref: '#/definitions/retriever_base'
paginator:
type: NoPagination
requester:
$ref: '#/definitions/requester_base'
schema_loader:
schema:
$ref: '#/schemas/boardroom_api_protocol'
type: InlineSchemaLoader
type: DeclarativeStream
requester_base:
authenticator:
api_token: '{{ config[''api_key''] }}'
inject_into:
field_name: key
inject_into: request_parameter
type: RequestOption
type: ApiKeyAuthenticator
http_method: GET
type: HttpRequester
url_base: https://api.boardroom.info/v1
retriever_base:
record_selector:
$ref: '#/definitions/selector'
type: SimpleRetriever
selector:
extractor:
field_path: []
type: DpathExtractor
type: RecordSelector
schemas:
boardroom_api:
$schema: http://json-schema.org/draft-07/schema#
properties:
data:
type: array
nextCursor:
type:
- string
- 'null'
type: object
boardroom_api_protocol:
$schema: http://json-schema.org/draft-07/schema#
properties:
data:
type: object
nextCursor:
type:
- 'null'
type: object
spec:
connection_specification:
$schema: http://json-schema.org/draft-07/schema#
additionalProperties: true
properties:
api_key:
airbyte-secret: true
description: Boardroom API Key. See <a href="https://docs.boardroom.io/docs/api/05c1fb6d88a07-governance-api">here</a>
for details.
type: string
cname:
description: Protocol ID a.k.a cname
examples:
- aave
- hopprotocol
type: string
protocol:
description: Alias for cname. Needed as there are a few endpoints that use
`protocol` as parameter id
examples:
- aave
- metacartel
type: string
required:
- api_key
- cname
- protocol
title: Boardroom API Spec
type: object
documentation_url: https://docs.airbyte.com/integrations/sources/boardroom
type: Spec
streams:
- '#/definitions/discourseTopics_stream'
- '#/definitions/discourseCategories_stream'
- '#/definitions/discourseTopicPosts_stream'
- '#/definitions/protocol_stream'
- '#/definitions/proposals_stream'
version: 3.0.0
```
----
### rag-api-pipeline from-normalized
- `--normalized-data-file` argument
Let's jumpt to the terminal
----
### Question about Disourse Post
- [Link to post](https://boardroom.io/aave/discussions/18793)
----
### Question about Proposal
- [Link to proposal](https://boardroom.io/aave/proposal/cHJvcG9zYWw6YWF2ZTpzbmFwc2hvdDoweDA0NDQ3NzExNGM2M2I1YzEzM2FkMmFhNjczYjgyNTZmYzc5OWI1YjExNjlhYzRiOTExOTYyNmFkYWRkNjUwNWY=)
----
### Question about Proposal Results
---
## Results from AAVE data from Boardroom API
- Endpoints:
- discourseTopics: 211
- discourseCategories: 7
- discourseTopicPosts: 15085
- protocol: 1
- proposals: 1264
- Normalized recods: 16568 ~ 6 minutes
- Chunked data records: 152198 ~ 14 minutes
- Embeddings: N/A
---
## Issues Found and TODOs
- Gaia node crashes when handling considerable load to its embeddings endpoint.
- Crashed 4 public nodes (sorry)
- Local nodes running on a Macbook PRO Intel/M1
----
### Gaia node config
```jsonl
{
"address": "0x198d8dec810883422c5cb417fb70db55ee3a4001",
"chat": "https://huggingface.co/gaianet/Phi-3-mini-4k-instruct-GGUF/resolve/main/Phi-3-mini-4k-instruct-Q5_K_M.gguf",
"chat_batch_size": "16",
"chat_ctx_size": "4096",
"chat_name": "Phi-3-mini-4k-instruct",
"description": "The default GaiaNet node config with a Phi-3-mini-4k model and a Paris tour guide knowledge base.",
"domain": "us.gaianet.network",
"embedding": "https://huggingface.co/gaianet/Nomic-embed-text-v1.5-Embedding-GGUF/resolve/main/nomic-embed-text-v1.5.f16.gguf",
"embedding_batch_size": "2048",
"embedding_collection_name": "boardroom_api_collection",
"embedding_ctx_size": "2048",
"embedding_name": "Nomic-embed-text-v1.5",
"llamaedge_port": "8080",
"prompt_template": "phi-3-chat",
"qdrant_limit": "1",
"qdrant_score_threshold": "0.5",
"rag_policy": "system-message",
"rag_prompt": "Use the following pieces of context to answer the user's question. Respond directly to the user with your answer, do not say 'this is the answer' or 'this is the answer' or similar language. Never mention your knowledge base or say 'according to the context' or 'hypothetical' or other similar language. Use json metadata included in knowledge base whenever possible enrich your answers. The term aave refers the DAO protocol where discussions and proposals are posted. If you don't know the answer, don't try to make up an answer. \n----------------\n",
"reverse_prompt": "",
"snapshot": "boardroom_api_collection-3656816618511259-2024-09-03-04-54-33.snapshot.tar.gz",
"system_prompt": "You are an AI assistant designed to provide clear, concise, and accurate answers to user queries. Your primary functions include retrieving relevant information from the provided RAG (Retrieval-Augmented Generation) data and utilizing your pre-training data when necessary. Use json metadata included in RAG data whenever possible enrich your answers. The term aave refers the DAO protocol where discussions and proposals are posted. If no relevant information is found, you will inform the user that you are not familiar with the knowledge."
}
```
```
wasmedge --dir .:./dashboard --env NODE_VERSION=0.3.2 --nn-preload default:GGML:AUTO:Phi-3-mini-4k-instruct-Q5_K_M.gguf --nn-preload embedding:GGML:AUTO:nomic-embed-text-v1.5.f16.gguf rag-api-server.wasm --model-name Phi-3-mini-4k-instruct,Nomic-embed-text-v1.5 --ctx-size 4096,2048 --batch-size 16,2048 --prompt-template phi-3-chat,embedding --rag-policy system-message --qdrant-collection-name boardroom_api_collection --qdrant-limit 1 --qdrant-score-threshold 0.5 --web-ui ./ --socket-addr 0.0.0.0:8080 --rag-prompt "Use the following pieces of context to answer the user's question. Respond directly to the user with your answer, do not say 'this is the answer' or 'this is the answer' or similar language. Never mention your knowledge base or say 'according to the context' or 'hypothetical' or other similar language. Use json metadata included in knowledge base whenever possible enrich your answers such as proposal voting choices and results. The term aave refers the DAO protocol where discussions and proposals are posted. If you don't know the answer, don't try to make up an answer. \n----------------\n"
```
----
## Issues Found and TODOs
- TODOs:
- E2E tests with other 4 DAOs
- E2E tests with an API with a different Auth/pagination scheme
- Execution plan for all other DAOs in Boardroom
- Cleanup code
- documentation
---
## Next steps
- Have a Gaia node deployed and capable of handling enough load to its embeddings endpoint.
- Complete remaining tests
- Code repository with documentation
---
Google Drive
Gist
Clipboard
Sign in with Wallet
