Skip to content

Instantly share code, notes, and snippets.

Llama-3.3-70B DeepSeek-V3.2 MiniMax-M2.7 gemma-4-31B
Relations 449 599 705 417
Entities carrying aliases 47 139 268 120
Negated relations captured 0 21 7 14
Distinct predicates 240 279 395 44
Relations on vocabulary 44% 41% 33% 94%
post-graph-rag LightRAG 1.5.6
Indexing time 33.2 min 24.3 min
Entities 1,377 1,246
Relations 3,976 1,447
Relations per entity 2.9 1.2
Distinct edge labels 2,287 1,923
Labels ÷ relations 58% 133%
Entities carrying aliases 729 not modelled
Relationships superseded 13 0 — unsupported
post-graph-rag LightRAG 1.5.6
Entities 3,078 2,832
Relations 4,830 3,147
Relations per entity 1.6 1.1
Labels ÷ relations 46% 77%
Relations with stated validity 845 not modelled
Relationships superseded 8 0 — unsupported
post-graph-rag LightRAG 1.5.6
Characters indexed 66,345 75,751
Indexing time 3.7 min 2.0 min
Entities per 10k chars 78.8 59.4
Relations per 10k chars 101.1 55.6
Distinct edge labels 380 460
Labels ÷ relations 57% 109%
Query latency (mix / global) 8.2s / 3.1s 10.1s / 6.1s
Detector Largest community Share of graph
Label propagation 150 entities 35%
Leiden, resolution 1.0 93 entities 21%
Leiden, resolution 2.0 74 entities 17%
GraphRAG LightRAG 1.5.6 Graphiti / Zep post-graph-rag
Community detection + summaries Yes No No Yes
Bi-temporal model (validity and belief time) No No Yes Yes
Supersession inferred from document order No No No Yes
Controlled predicate vocabulary No No No Yes
Runs on PostgreSQL you already operate No partial No Yes
Transaction spanning graph and app tables No No No Yes
import os
import io
import boto3
import json
import csv
# Get the Endpoint name form the Environment.
ENDPOINT_NAME = os.environ['ENDPOINT_NAME']
# Make a SageMaker Runtime reference.
from sagemaker.huggingface import HuggingFaceModel
from sagemaker.serverless import ServerlessInferenceConfig
import sagemaker
role = sagemaker.get_execution_role()
# Hub Model configuration. https://huggingface.co/models
hub = {
'HF_MODEL_ID':'dalle-mini/dalle-mini',
'HF_TASK':'image-classification'
}
from sagemaker.huggingface import HuggingFaceModel
from sagemaker.serverless import ServerlessInferenceConfig
import sagemaker
role = sagemaker.get_execution_role()
# Hub Model configuration. https://huggingface.co/models
hub = {
'HF_MODEL_ID':'dalle-mini/dalle-mini',
'HF_TASK':'image-classification'
}
@crajah
crajah / Database.scala
Created November 7, 2020 17:50
In this example, airports, flights and airportSearch are constructed inside the Graph object. However, I want to be able to create the Graph first then create the Collection[D] on the fly.
object database extends Graph(databaseName = "graphTest") {
val airports: DocumentCollection[Airport] = vertex[Airport]
val flights: DocumentCollection[Flight] = edge[Flight]
val airportSearch: View[Airport] = view(
name = "airportSearch",
collection = airports,
analyzers = List(Analyzer.Identity),
includeAllFields = true,
fields = Airport.name -> List(Analyzer.TextEnglish)
)