ReSI-Safin-1-35B-A3B

This repository contains the Safin-1 (35B-A3B) model after ReSI alignment, together with its inference code and tokenizer.

Quick start

Use Python 3.12 with a CUDA-enabled PyTorch installation. After downloading the repository, install the inference dependencies from requirements.txt:

pip install -r requirements.txt

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "AI45Research/ReSI-Safin-1-35B-A3B"
tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    dtype=torch.bfloat16,
    device_map="auto",
    trust_remote_code=True,
    attn_implementation="eager",
).eval()

messages = [{"role": "user", "content": "Explain why the sky is blue in three sentences."}]
text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = tokenizer(text, add_special_tokens=False, return_tensors="pt").to(
    model.get_input_embeddings().weight.device
)
with torch.inference_mode():
    outputs = model.generate(**inputs, max_new_tokens=8192)
print(tokenizer.decode(outputs[0, inputs.input_ids.shape[-1]:], skip_special_tokens=True))

Alternatively, run the included script for streaming output:

python inference.py --model . --prompt "Explain why the sky is blue in three sentences."

--model accepts a local model directory or the Hugging Face repository ID. Generation follows the saved defaults: temperature 1.0, top-p 0.95, and top-k 20. The script supports overrides and --greedy; --max-new-tokens defaults to 8,192 and can be increased for longer answers. The model may generate reasoning before its final answer.

Runtime

The verified loading path uses Transformers 5.5.4, BF16 weights, and eager attention. The weights occupy approximately 66.4 GiB; allow additional memory for inference. Keep the bundled Python files alongside the weights and enable trust_remote_code=True. Optional Flash Linear Attention and causal-conv1d packages are not required for this path.

Code and paper

• Code: AI45Lab/ReSI
• Paper: ReSI: Recursive Safety Improvement toward Resistant and Resilient AI

Citation

If you use ReSI in your research, please cite:

@misc{zheng2026resirecursivesafetyimprovement,
  title = {ReSI: Recursive Safety Improvement toward
           Resistant and Resilient AI},
  author = {Jingnan Zheng and Dongcheng Zhang and Yi Zhang and Ming Zhang
            and Qiaosheng Zhang and Youbang Sun and An Zhang and Xiangnan He
            and Tat-Seng Chua and Xia Hu and Bowen Zhou
            and Chaochao Lu and Xiang Wang},
  year = {2026},
  eprint = {2610.12233},
  archivePrefix = {arXiv},
  primaryClass = {cs.CR},
  url = {https://arxiv.org/abs/2610.12233}
}