Needle-in-a-Haystack Retrieval on BABILong 2K
65.2AccuracyMemDLM (Train & Inference)
Evaluation Results
| Method | Links | |
|---|---|---|
| MemDLM (Train & Inference)Backbone=LLaDA2.12026.03 | 65.2 | |
| MemDLM (Train-Only)Backbone=LLaDA2.12026.03 | 63.5 | |
| MemDLM (Train & Inference)Backbone=LLaDA-MoE2026.03 | 61.2 | |
| MemDLM (Train-Only)Backbone=LLaDA-MoE2026.03 | 59.4 | |
| Standard MDLMBackbone=LLaDA2.12026.03 | 56 | |
| Standard MDLMBackbone=LLaDA-MoE2026.03 | 46 |