events list |
|---|
[
{
"event_id": "CSE-001",
"date": "2021-03-23T00:00:00",
"headline": "Ever Given blocks Suez Canal",
"affected_commodities": [
"oil",
"shipping",
"grain"
],
"price_direction": {
"oil": "UP",
"shipping": "UP",
"grain": "UP",
"lng": null
},
"gro... |
LMNOP Evaluation Benchmark
39-event hand-labelled benchmark for evaluating hallucination and factual grounding in LLM-generated commodity intelligence reports.
Built for CS-552 Modern NLP, EPFL — Group g26.
Dataset
Each event contains:
- date, headline, affected domains (oil/LNG/grain/shipping)
- ground-truth severity label
- expected cross-commodity causal effects
- correct_detection_means field for event detection scoring
Usage
import json
from huggingface_hub import hf_hub_download
path = hf_hub_download(repo_id="NicolasKhamis/lmnop-benchmark", filename="benchmark_events.json", repo_type="dataset")
with open(path) as f:
data = json.load(f)
events = data["events"] # list of 39 dicts
Paper
LMNOP: Suppressing Hallucination in LLM-Generated Analytical Text via Multi-Agent Structured Data Grounding
https://github.com/CS-552/open-project-m3-lmnop
- Downloads last month
- 54