File size: 2,229 Bytes
b844c48 032c521 b844c48 e88c706 032c521 e88c706 032c521 e88c706 032c521 f718fde 032c521 e88c706 032c521 e88c706 032c521 e88c706 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 | ---
title: README
emoji: 🧭
colorFrom: green
colorTo: indigo
sdk: static
pinned: false
---
<p align="center">
<img src="https://huggingface.co/spaces/dlab-spp/README/resolve/main/overview.png" alt="SPP: annotate pretraining data with normative reflections, inject at different pretraining stages, evaluate alignment and safety" width="100%">
</p>
Alignment — and the assistant identity itself — is normally introduced only *after*
pretraining, once behavioral priors are already set. **SPP** installs the desired persona from
token zero instead: we define it through normative values in a constitution, generate
first-person moral reflections grounded in that constitution, and insert them throughout the
pretraining corpus behind an `<assistant>` token. Post-training then binds the chat assistant
identity to the installed persona. Pretraining up to **3B on 500B tokens**, SPP improves
constitution following and jailbreak robustness while preserving capabilities — and *when* the
data arrives matters: models trained with reflections from token zero prioritize values
differently and take fewer risky actions in out-of-distribution moral dilemmas than models
given the exact same data only at the end of pretraining, an advantage that **grows with scale**.
## Collections
📦 **[Pretraining Datasets](https://huggingface.co/collections/dlab-spp/pretraining-datasets-6a6b3b62c372179e342a1de6)**
— the reflection data, the corpus selection manifest, safety scores, and verification files.
🤖 **[Models — 3B](https://huggingface.co/collections/dlab-spp/models-3b-6a6b3cbc2bf4be9aba1c4a16)** ·
**[Models — 1.7B](https://huggingface.co/collections/dlab-spp/models-17b-6a6b3cbf48d091f505d412aa)**
— all five recipes, at both scales. We release all pretraining checkpoints, base, and instruct models, at both scales
💬 **[Post-training Dataset](https://huggingface.co/collections/dlab-spp/post-training-dataset-6a6c801cd9d46e4ff5d2ac25)**
— SP-SFT, the mixture that performs persona binding.
📊 **[Evals](https://huggingface.co/collections/dlab-spp/evals-6a6c810619bb9009709286da)**
— ConstitutionEval and an audited AIRiskDilemmas.
<sub>From [EPFL DLAB](https://huggingface.co/epfl-dlab).</sub>
|