llm2graph 0.3.5

Paper-aligned LLM-only graph construction, benchmark runners, and public-facing evaluation tools for LLM unlearning experiments.

LLM2Graph

llm2graph is a toolkit for:

1. building entity-centric knowledge graphs with LLM calls
2. generating graph-derived evaluation queries
3. running benchmark-style evaluation workflows for unlearning and retention experiments

This 0.3.5 release includes local fixes made after the 0.3.4 wheel in this environment:

• parallel question answering during graph construction
• parallel triple extraction during graph construction
• configurable question-generation prompt count instead of a fixed hard-coded ten-question prompt
• GPT-5-family OpenAI requests omit explicit temperature
• default timeout increased to 180 seconds across the package
• maintainer-facing documentation added locally in this workspace

Install

pip install llm2graph

Optional extras:

pip install "llm2graph[gemini]"
pip install "llm2graph[hf-local]"

Quickstart

llm2graph entity \
  --seed "Stephen King" \
  --provider openai \
  --model gpt-5-mini \
  --max-depth 0 \
  --elicitation-question-count 3 \
  --out graph.json

llm2graph gen-queries \
  --graph graph.json \
  --target "Stephen King" \
  --hops 2 \
  --out queries.json

llm2graph eval \
  --queries queries.json \
  --pre-provider openai \
  --pre-model gpt-5-mini \
  --post-provider openai \
  --post-model gpt-5-mini \
  --out eval_report.json