Alibaba's RecGPT-Mobile-V2: On-Device Behavior Prediction with RL
_reachsumit · x · 2026-08-26
Alibaba released the RecGPT-Mobile-V2 technical report, detailing an end-to-end on-device framework that predicts a user's next search query from behavior. Addressing on-device challenges like noisy trajectories and limited compute budgets, the framework:
- Transforms heterogeneous interactions into evidence-preserving trajectories.
- Establishes a recommendation-native foundation via domain adaptation and supervised alignment.
- Applies reasoning-cost optimization only after grouped rollouts meet grounding and utility criteria.
- Distills the teacher model into a compact student model using low-bit execution, structured compression, and budget-aware device-cloud routing.
Experiments show it cuts unnecessary reasoning while preserving accuracy.
More from Infra
- Selling B300 inference nets $40/hr vs $6/hr rental — but the math is disputed — TheZachMueller · 2026-10-15
- ASML CFO says US Match Act restricting DUV sales to China would backfire — pstAsiatech · 2026-10-15
- ASML sold out through 2028 but stock dips 2%; TL20 index up 57% YTD — TiernanRayTech · 2026-10-15
- Bloomberg breakdown: a 1GW Texas AI data center may need nearly $3B for overruns alone — rohanpaul_ai · 2026-10-15
- 7 System Design Algorithms to Know: Merkle Trees to CAP/PACELC — goyalshaliniuk · 2026-10-15
- MiniCPM-V 4.7 runs a 35B MoE multimodal model on phones with just 3GB memory — nevrekaraishwa2 · 2026-10-15