huginn-m-fixed-r5

Table 2, M, Fixed R=5 of Towards Looped Models Done Right. Part II: Rethinking at Fixed Points.

[!IMPORTANT] Loading this checkpoint requires the xLLM code. The weights are stored in xLLM's native format (BF16 Safetensors with an xLLM config.json and artifact_manifest.json). This is not a Hugging Face transformers checkpoint: AutoModel.from_pretrained cannot load it. Get the code at https://github.com/ifm-ai/xllm-loop and load the directory with xllm.paper_part2.artifacts.load_artifact.

Model details


Model	looped Transformer: 1 prelude, 2 recurrent and 1 coda block; width 3,072, 48 heads (12 KV heads), SwiGLU FFN 8,704
Parameters	819,495,936, including the embedding and output layers
Input injection	diagonal
Recurrent depth in training	fixed R = 5
Evaluation	R = 5 with terminal KV; the recurrent state is seeded by example identity (seed 42)
Training	recipe `m_fixed_r5`: 20,480 updates of 512 x 8,192 tokens (85.90B), AdamW, warmup-stable-decay
Weights	BF16 Safetensors: the training checkpoint's FP32 weights rounded to BF16
Tokenizer	`tokenizer/`, the Jais64k tokenizer of training

Download

hf download IFM/LoopedLM-P2-huginn-m-fixed-r5 --local-dir huginn-m-fixed-r5

The xLLM loader checks the directory against artifact_manifest.json: it rejects symbolic links and files the manifest does not list, apart from the .gitattributes file and the .cache/huggingface/ folder that hf download --local-dir adds. Download into a directory as above, not into the Hub cache (~/.cache/huggingface/hub), whose files are symbolic links.

Use

Evaluate with eval_paper_part2.py from the xLLM repository:

ENABLE_FLASH_ATTENTION_3=true python eval_paper_part2.py --artifact huginn-m-fixed-r5 \
    --data /path/to/eval-data/data.json --out out ppl

data.json and the evaluation inputs come from release/paper-part2/prepare-eval-data.py --output /path/to/eval-data.

xllm.paper_part2.artifacts.load_artifact("huginn-m-fixed-r5") verifies the directory and returns the model config fields and the BF16 state dict on CPU.

artifact_manifest.json records the size and SHA-256 of every file in this repository.

Paper and citation

Towards Looped Models Done Right. Part II: Rethinking at Fixed Points: https://arxiv.org/abs/2610.06833

@article{huang2026fixedpoints,
  title   = {Towards Looped Models Done Right, Part II: Rethinking at Fixed Points},
  author  = {Benhao Huang and Chufan Shi and Junlin Chen and Shicheng Wen and Zhengzhong Liu and Eric Xing and Xuezhe Ma},
  journal = {arXiv preprint arXiv:2610.06833},
  year    = {2026}
}

License

The weights are released under the Apache License 2.0 (LICENSE); NOTICE records the tokenizer's attribution and how the artifact was prepared. The xLLM code is distributed under its own license.