TRACE：面向企业 LLM 的业务规则驱动推理课程，实现知识保留的参数化工具检索

View PDF HTML (experimental)

Abstract:Parametric retrieval enables LLMs to retrieve tools implicitly by assigning each API a unique virtual token and training the model to generate it via constrained beam search. Toolsense shows that this regime has two critical drawbacks: it destroys parametric tool knowledge during training, and its beam-search decoding is too slow for real-time deployment. We introduce TRACE (Tool Retrieval via Augmented Chain-of-thought and Enterprise rules), a two-stage curriculum that resolves this dissociation. Stage 1 reuses the multi-format memorization SFT from ToolSense to seed tool knowledge with LoRA. Stage 2 is our core contribution: the model is trained to emit a thinking trace before producing a JSON list of tool tokens, using two data sources -- RRB pairs from ToolSense and queries synthesized to target business rules curated by domain experts -- both augmented with reasoning traces. This training objective preserves Stage 1 MCQ and QA probing accuracy while enabling single-beam greedy decoding at production latency. Evaluated on a combined enterprise catalog of 8,300+ tools across two enterprise product lines, TRACE training for Stage 2 not only preserves but improves tool understanding: MCQ accuracy gains +3.2 pp and QA probing gains +9 pp over Stage 1. On retrieval, TRACE achieves ~86% recall on Domain A and ~60% on Domain B -- compared to embedding baseline performance of ~27% & ~52% -- both with single-beam greedy decoding, making it directly deployable at production latency.

Subjects:	Artificial Intelligence (cs.AI)
Cite as:	arXiv:2607.22639 [cs.AI]
	(or arXiv:2607.22639v2 [cs.AI] for this version)
	https://doi.org/10.48550/arXiv.2607.22639 arXiv-issued DOI via DataCite

Submission history

From: Ashutosh Hathidara [view email]
[v1] Mon, 22 Jun 2026 23:15:35 UTC (458 KB)
[v2] Tue, 10 Nov 2026 15:45:44 UTC (467 KB)