| --- |
| license: apache-2.0 |
| library_name: xllm |
| pipeline_tag: text-generation |
| tags: |
| - xllm |
| - looped-language-model |
| - paper-part2 |
| - dense |
| --- |
| |
| # d4-m |
|
|
| Table 4, M, D4 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](https://github.com/ifm-ai/xllm-loop) and load the directory with `xllm.paper_part2.artifacts.load_artifact`. |
|
|
| ## Model details |
|
|
| | | | |
| | --- | --- | |
| | Model | dense Transformer of 4 blocks, untied output layer; width 3,072, 48 heads (12 KV heads), SwiGLU FFN 8,704 | |
| | Parameters | 810,052,608, including the embedding and output layers | |
| | Evaluation | full KV cache | |
| | Training | recipe `m_d4`: 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 |
|
|
| ```bash |
| hf download IFM/LoopedLM-P2-d4-m --local-dir d4-m |
| ``` |
|
|
| 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: |
|
|
| ```bash |
| ENABLE_FLASH_ATTENTION_3=true python eval_paper_part2.py --artifact d4-m \ |
| --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("d4-m")` 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 |
|
|
| ```bibtex |
| @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. |
|
|
