{
  "schema_version": "1.1",
  "id": "s13:https://arxiv.org/abs/2609.05363v1",
  "slug": "2609-05363v1-0aftwt3",
  "url": "https://feed7.dev/p/2609-05363v1-0aftwt3",
  "title": "Distill Globally, Adapt Locally: Reasoning Distillation and Product-Type Test-Time Training for Scalable Trade-Up Recommendation",
  "why_included": "A compact classifier distilled from LLM rationales handled product-pair decisions without inference-time LLM calls. Category adapters improved accuracy further while retaining large speed and cost gains.",
  "summary": "A **15.5M-parameter** student uses two precomputed 768-dimensional product embeddings at inference. Reasoning distillation reached **0.924 AUC**, versus 0.912 for label-only training, on 8,352 annotated pairs.",
  "practical_implication": "For high-volume classification, use an LLM to produce structured supervision offline, then serve a compact non-generative model. Product-type adapters raised AUC to **0.941** without unfreezing the shared student.",
  "agent_context": "A **15.5M-parameter** student uses two precomputed 768-dimensional product embeddings at inference. Reasoning distillation reached **0.924 AUC**, versus 0.912 for label-only training, on 8,352 annotated pairs.\n\nFor high-volume classification, use an LLM to produce structured supervision offline, then serve a compact non-generative model. Product-type adapters raised AUC to **0.941** without unfreezing the shared student.\n\nOn a 100K-pair proxy catalog, the student was about **5,000x faster and 10,000x cheaper** than direct LLM inference. Those estimates used one eight-GPU machine, and the evidence is specific to trade-up recommendations rather than coding-agent workloads.",
  "source": {
    "name": "arXiv",
    "url": "https://arxiv.org/abs/2609.05363v1",
    "published_at": "2026-09-04T17:08:50.000Z"
  },
  "source_class": "blog_post",
  "content_type": "Paper",
  "layer": "model",
  "domains": [
    "data"
  ],
  "topics": [
    "reasoning",
    "model-selection"
  ],
  "verification": {
    "status": "needs_review",
    "label": "Needs Review",
    "method": "unverified",
    "verified_at": null
  },
  "uncertainty": [
    "On a 100K-pair proxy catalog, the student was about **5,000x faster and 10,000x cheaper** than direct LLM inference. Those estimates used one eight-GPU machine, and the evidence is specific to trade-up recommendations rather than coding-agent workloads."
  ],
  "connected_context": {
    "meaning": "This provides workload-specific evidence for moving LLM reasoning out of the serving path: generate richer supervision offline, deploy a compact embedding-based classifier, then adapt only small product-type components. It complements data-quality and distillation evidence but narrows the lesson to repetitive trade-up classification; the reported speed and cost advantages do not establish transfer to generative or coding-agent tasks.",
    "corpus_size": 703,
    "generated_at": "2026-09-08T10:04:26.072Z",
    "connections": [
      {
        "title": "Data Quality Is the Compute Multiplier — Ari Morcos, DatologyAI",
        "source_name": "AI Engineer",
        "source_url": "https://www.youtube.com/watch?v=_PdK6x7PQNM",
        "feed7_url": "https://feed7.dev/p/data-quality-is-the-compute-multiplier-ari-morcos-datologyai-0x7k2ve",
        "reason": "The offline structured rationales are a form of task-matched synthetic data, providing a concrete instance where supervision quality improves a compact model beyond label-only training."
      },
      {
        "title": "Rethinking On-Policy Distillation of Large Language Models II: One Training Example",
        "source_name": "arXiv",
        "source_url": "https://arxiv.org/abs/2609.04172v1",
        "feed7_url": "https://feed7.dev/p/2609-04172v1-0kofdqa",
        "reason": "Both shift attention from raw dataset size to the information carried by distillation supervision, but this Signal uses structured offline reasoning for a non-generative student rather than diverse on-policy rollouts."
      },
      {
        "title": "Consolidating RLVR Capabilities Across Domains: A Deep Dive into Fusion Paradigms",
        "source_name": "arXiv",
        "source_url": "https://arxiv.org/abs/2608.27409v1",
        "feed7_url": "https://feed7.dev/p/2608-27409v1-1bivffd",
        "reason": "Product-type adapters mirror the broader finding that specialization choices should follow deployment constraints: local adaptation preserves a shared student while improving task-specific performance."
      }
    ]
  },
  "lifecycle": "Current",
  "published_at": "2026-09-04T17:08:50.000Z",
  "modified_at": "2026-09-04T17:08:50.000Z",
  "supersedes": [],
  "expires_at": null,
  "formats": {
    "html": "https://feed7.dev/p/2609-05363v1-0aftwt3",
    "json": "https://feed7.dev/p/2609-05363v1-0aftwt3.json",
    "markdown": "https://feed7.dev/p/2609-05363v1-0aftwt3.md"
  }
}