cd /news/artificial-intelligence/better-starts-better-ends-bootstrapp… · home topics artificial-intelligence article
[ARTICLE · art-65395] src=arxiv.org ↗ pub= topic=artificial-intelligence verified=true sentiment=↑ positive

Better Starts, Better Ends: Bootstrapped Iterative Self-Reasoning Distillation for Compressed Reasoning

Researchers propose BIRD (Bootstrapped Iterative Self-Reasoning Distillation), a two-stage method that improves compressed reasoning by first generating concise, correct traces from a base model under a brevity instruction and then applying on-policy distillation on cleaner prefixes. On Qwen3-8B, BIRD raises MATH-500 accuracy from 86.2% to 92.0% while cutting average response length from 3,099 to 1,115 tokens, demonstrating a stronger accuracy-efficiency trade-off than prompting or cold-start distillation.

read1 min views2 publishedJul 20, 2026

arXiv:2607.15736v1 Announce Type: new Abstract: Large reasoning models often solve problems through long chain-of-thought (CoT) traces, yet much of this computation is spent on redundant derivations, repeated self-verification, and detours that do not improve the final answer. Existing on-policy self-distillation methods reduce this cost by matching a student model to a concise copy of itself on prefixes sampled from the student's own rollouts. We show that this objective has an initialization bottleneck. Since supervision is applied only to visited prefixes, training from a verbose base model places the KL loss on contexts that are often noisy, redundant, or already off track. In such regions, a concise teacher can provide only local corrections, while the student continues to explore trajectories that an efficient reasoner should avoid. In this paper, we propose BIRD(Bootstrapped Iterative Self-Reasoning Distillation), a two-stage self-reasoning distillation method that improves the rollout distribution before on-policy training. BIRD first samples concise solutions from the base model under a brevity instruction, keeps only answer-correct traces, and performs a lightweight prompt-switch SFT step. The traces are generated with the brevity instruction but learned under the original task prompt, turning instruction-induced conciseness into a default reasoning behavior. Starting from this warm model, BIRD then applies on-policy reverse-KL distillation with a concise self-teacher, now on cleaner and more informative prefixes. Across Qwen3 series models, BIRD achieves a stronger accuracy-efficiency trade-off than prompting and cold-start on-policy distillation on MATH-500 and AIME benchmarks. On Qwen3-8B, it improves MATH-500 accuracy from 86.2% to 92.0% while reducing the average response length from 3,099 to 1,115 tokens. These results highlight prefix support as a central factor in efficient reasoning distillation.

── more in #artificial-intelligence 4 stories · sorted by recency
── more on @bird 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
Live at https://your-agent.zahid.host
Get free account → Pricing
from €0/mo · no card required
LIVE [news/better-starts-better…] indexed:0 read:1min 2026-07-20 ·