Progressive RLVR paper says TinyLoRA and distillation are both necessary
ddkang · x · 2026-07-21
The thread reports an ablation study on Progressive RLVR and says every ingredient matters.
Key findings:
- Removing distillation and training directly with TinyLoRA makes the bound 15% looser.
- Replacing TinyLoRA with standard LoRA makes the bounds vacuous on the 4B model and meaningless on the 2B model.
- On Qwen3.5-4B across Math, Code, General Knowledge, and Text-to-SQL, the bounds are tight: they land within 8–13% of training accuracy and 17–51% above the base model.
The plot in the image compares Progressive RLVR and TinyLoRA/LoRA variants across quantization levels, showing Progressive w/ TinyLoRA tracking best among the bound methods.
More from Research
- LightMTP: lightweight latent multi-token prediction boosts LLM planning in pretraining — gdm3000 · 2026-10-27
- EMNLP study: AI literary translation reads fine but varies more, even within one book — tuvllms · 2026-10-27
- Lab monkeys at $30-80K each: a 4-year breeding cycle is the real preclinical drug bottleneck — zebird0 · 2026-10-27
- OpenAI Foundation funds UNC tumor antigen project, hiring senior experimental scientist — iskander · 2026-10-27
- Steering attention heads toward honesty: up moral patienthood, down psychopathy — edelwax · 2026-10-27
- Are transcoders overkill? New paper argues plain neurons suffice for circuit analysis — burny_tech · 2026-10-27