---
license: cc-by-4.0
task_categories:
  - question-answering
tags:
  - persistent-memory
  - event-stream
  - synthetic
---

# PNS-WORLD

Data for *Learning Persistent Referential Identity for Two-Hop Retrieval in Event-Stream Models*.

| Experiment | Assets |
|---|---|
| E1 / E3 | `data/`, `manifests/`, `audit/` |
| E8-FULL | `e8_full/` ordinary and 512-write challenge evaluation shards, training samples, split manifests |
| Reviewer-requested tests | `rr2/original/`, `rr2/compose/` evaluation shards, complete training manifests, disjoint-name registration and hashes |

```bash
pip install huggingface_hub
hf download nur-dev/pns-world --repo-type dataset --revision paper-v2.1 --local-dir pns-data
hf download nur-dev/pns-bind-25m --revision paper-v2.1 --local-dir pns-model
cd pns-model
pip install -r e8_full/requirements.txt
python rr2/reproduce.py verify --data ../pns-data/rr2
python rr2/reproduce.py prepare-training --data ../pns-data/rr2 --output rr2_train --workers 24
```

`rr2/REGISTRATION.json` lists the 66-name inventories and tokenization checks.
Full training corpora are regenerated from the deterministic released generator.
Evaluation commands are in the model repository. `paper-v1.0` and `paper-v2.0`
retain the original releases.