Webpage: praxis-research.org/grafting | Paper link: https://arxiv.org/abs/2610.00767

Work done during Peter and Dani's MATS Fellowship (10.0) under the mentorship of Shi, with Jinghua Ou as Research Manager.

TLDR;

We share a simple modification to synthetic document fine-tuning (SDF) that works surprisingly well: do fine-tuning on a pre-training checkpoint and add the resulting weight difference onto the post-trained version that you want to deploy. We call this grafting.

We empirically validate grafting for implanting false facts, training model organisms, and constitutional mid-training on five model families up to 284 billion parameters. We find:

Native SDF (fine-tuning on the post-trained model) can significantly damage the coherence and capabilities of a base model, especially for alignment-related properties like comprehension of reality or sharp preferences. Grafting significantly alleviates this.
Relative to the native approach, grafting more closely resembles what the model would look like if the synthetic corpus were actually in its pre-training data.
One graft is good for many steps of post-training. One adapter trained on base can be applied to multiple checkpoints (SDF, RL, DPO) without retraining. (such as adversarial concealment for model organisms).
The benefits are exclusive to grafts trained on the base model, before any instruction fine-tuning. For [...]

---

Outline:

(00:29) TLDR;

(03:08) Introduction

(04:14) Why can't we mix synthetic documents into pre-training?

(05:28) Ok, why not just do it after?

(08:12) How to Graft

(09:16) Results

(09:28) AuditBench Organisms

(13:29) False Facts

(13:33) We try grafting on synthetic documents that describe incorrect facts, specifically those used in the Negation Neglect paper. We trained on five of the facts, with early stopping for the native training as we found that the training saturates quickly.