llada-8b-dllm-registers-mix60k-r4-corrected
This model is described in the paper Register Tokens for Bounded-State Reasoning in Diffusion Language Models.
Corrected matched-protocol LLaDA-8B R4 baseline trained on MI355X with 4 continuous register slots.
This checkpoint is part of the dLLM Registers project — register tokens as a bounded, trained, continuous channel for carrying decoding state across denoising windows in diffusion language models.
- Paper section: ICLR 2027 matched register-versus-memory validation
- Carry channel: continuous register tokens (
channel_mode=registers,num_registers=4,tail_length=0) - Base model: GSAI-ML/LLaDA-8B-Base
- Training data: mix60k (OpenMathInstruct-2 + OpenCodeInstruct, 60K examples; source SHA-256 58115f6be4bad1c635f3cdc28a57ee83305d38bb82a10a240030cdd9c32fc7f0)
- Training config:
SFT/slurm/p4_memory_pilot_rocm.sbatch - Chunk size: C = 128 tokens
- Prompt dropout rate (Bernoulli per-trace CSG mask): 0.3
- Training semantics: Primary register carry is not detached, auxiliary reconstruction is disabled, and all four slots are trained only by the next-chunk task objective. The saved config records
d1_detach_primary_register_bridge=falseandd1_aux_recon_loss=false. - Date uploaded: 2026-09-02
How to load
from transformers import AutoModel, AutoTokenizer
model = AutoModel.from_pretrained("albertge/llada-8b-dllm-registers-mix60k-r4-corrected", trust_remote_code=True)
tok = AutoTokenizer.from_pretrained("albertge/llada-8b-dllm-registers-mix60k-r4-corrected", trust_remote_code=True)
To use the carry channel correctly at inference, see the evaluator at
eval/eval.py
and the wrapper eval/run_mix60k_full_eval.sh.
Key flags for this checkpoint:
--num_registers 4 --channel_mode registers --tail_length 0 --use_mask_token_for_registers
Repository
Training and eval code: https://github.com/lbertge/d1-registers
Citation
If you use this checkpoint, please cite the dLLM Registers paper:
@misc{dllm-registers-2026,
title = {Register Tokens for Unbounded Reasoning in Diffusion Language Models},
author = {Albert Ge and collaborators},
year = {2026},
note = {Preprint},
url = {https://github.com/lbertge/d1-registers}
}