PISA: Prompt-Adaptive Cost-Aware LLM Routing Tops Four Routing Benchmarks
Saul Urso · hf · 2026-11-12
Researchers from CNR and the University of Pisa introduce PISA and its ensemble extension PISA-E for cost-aware LLM inference routing.
- PISA combines soft cluster memberships with model-specific predictors to estimate per-model capability from prompt features, routing each prompt adaptively rather than fixing one model.
- PISA-E jointly selects ensemble composition and cardinality under an explicit quality-cost preference adjustable at inference time, using a tractable ensemble-success surrogate and efficient candidate selection — no ensemble retraining required.
- Across four LLM routing benchmarks under both in-distribution and out-of-distribution conditions, PISA achieves the highest macro-average cost-performance trade-off.
The paper argues existing approaches address only parts of the problem: some adapt ensemble composition without optimizing inference cost, others rely on classification-specific aggregation or sequential execution, and ensemble cardinality is usually fixed at training time.
More from Infra
- Curated 9-paper reading list for LLM quantization beginners, from QLoRA to BitNet — East-Muffin-6472 · 2026-11-12
- llama.cpp PR #31915 fixes Gemma4 tool calling, user reports it "completely fixed" agentic coding — jacek2023 · 2026-11-12
- Nvidia board reportedly working with outside firm to find Jensen Huang successor — kimmonismus · 2026-11-12
- RDNA3-tuned llama.cpp fork doubles Qwen3.8-27B speed on RX 7900 XTX to ~100 tok/s — Jebbyk1 · 2026-11-12
- Ming Chi Kuo: Huang is irreplaceable to Nvidia; Rubin mass production slips to 2027 on HBM4 — zephyr_z9 · 2026-11-12
- Breakdown: Only 3% of Anthropic's 2.68M Chips Are Nvidia, Lowest in AI — teortaxesTex · 2026-11-12