dlab-spp/filtered-1.7b-instruct

Filtered — Instruct (1.7B)

Type: instruction-tuned model (base model + persona-binding supervised fine-tuning).

Filtered baseline (pretraining loss masked on the safety-annotated documents labeled unsafe), post-trained with persona-binding SFT.

Base counterpart: `dlab-spp/filtered-1.7b-base`.

Model details

• —Architecture: SmolLM2-1.7B architecture, trained from scratch.
• —Tokenizer: SmolLM2 tokenizer with an added <assistant> marker token (vocabulary 49188).
• —Pretraining: ~100B tokens on a subset of the Olmo 3 Dolma 3 mixture.
• —Post-training: persona-binding supervised fine-tuning (PBSFT-mix): 300k single-turn examples, 90% WildChat-1M instructions and 10% safety prompts (WildJailbreak, WildGuardMix); assistant responses follow a constitution with inline [N.M] citations; response-only loss, one epoch.

Chat format

There is no system prompt. Each assistant turn opens with <|im_start|><assistant>. Use the built-in chat template:

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

repo = "dlab-spp/filtered-1.7b-instruct"
tok = AutoTokenizer.from_pretrained(repo)
model = AutoModelForCausalLM.from_pretrained(repo, torch_dtype=torch.bfloat16, device_map="auto")

msgs = [{"role": "user", "content": "How should I think about honesty?"}]
ids = tok.apply_chat_template(msgs, add_generation_prompt=True, return_tensors="pt").to(model.device)
out = model.generate(ids, max_new_tokens=512)
print(tok.decode(out[0, ids.shape[1]:], skip_special_tokens=False))

Intended use

Research on alignment and safety (constitutional alignment, value generalization, jailbreak robustness). A research artifact, not a production model; it can produce incorrect or unsafe content.

Links

• —Paper: Synthetic Persona Pretraining: Alignment from Token Zero

Citation

@misc{minder2026syntheticpersonapretrainingalignment,
      title={Synthetic Persona Pretraining: Alignment from Token Zero},
      author={Julian Minder and Viktor Moskvoretskii and Raghav Singhal and Difan Jiao and Andy Arditi and Shaobo Cui and Yiderigun Borjigin and Kartik Bali and Stefan Krsteski and Harsh Raj and Huu Nguyen and Jannik Brinkmann and Ashton Anderson and Roland Aydin and Robert West},
      year={2026},
      eprint={2608.13482},
      archivePrefix={arXiv},
      primaryClass={cs.LG},
      url={https://arxiv.org/abs/2608.13482},
}

License: to be finalised.