Trajectory Supervision for Continual Tool-Use Learning in LLMs

• URL: http://arxiv.org/abs/2605.09734v1
• Date: Sun, 10 May 2026 20:09:41 GMT
• Title: Trajectory Supervision for Continual Tool-Use Learning in LLMs
• Abstract summary: We fine-tune Llama 3.1 8B Instruct with QLoRA on API-Bank using four sequential domain blocks.<n> Condition A strips previous API request/response lines from the prompt and trains the model to predict the next API call.<n> Condition B reaches 56.9% final exact full-call accuracy compared with 39.2% for Condition A.
• Score: 0.3823356975862005
• License: http://creativecommons.org/licenses/by-sa/4.0/
• Abstract: Most language-model training data shows final artifacts, not the process that produced them. We study a tractable version of this question in tool use: when a model learns a stream of new API domains, does keeping tool-use trajectories help compared with stripping the intermediate API trace? We fine-tune Llama 3.1 8B Instruct with QLoRA on API-Bank using four sequential domain blocks. Condition A strips previous API request/response lines from the prompt and trains the model to predict the next API call. Condition B keeps the trajectory context. In a single-seed pilot, full held-out generation evaluation shows that Condition B reaches 56.9\% final exact full-call accuracy compared with 39.2\% for Condition A. B also improves final API-name accuracy by 7.7 points. However, B uses 25.1\% more training tokens, the run uses one seed, and the task is next-call prediction rather than full dialogue success.

Related papers

• Joint Optimization of Tool Creation and Use for Large Language Model Agents [69.07066297088187]
  SMITH (-grounded Multi-task Iterative Tool Honing) trains tool creation and tool use inside a single policy.<n>A 4B Qwen3 trained with SMITH on 13 procedural reasoning tasks with exact verifiers reaches 79.8 macro-average accuracy on held-out tasks.<n>It also reaches 40.4 on TabMWP-Hard and 42.6 on out-of-domain GQA (+7.6 over the best same-backbone inference-time baseline)
  arXiv  Detail & Related papers  (2026-08-25T13:59:34Z)
• VAKRA: Evaluating Multi-Hop Reasoning Across APIs and Retrieval Under Tool-Use Policies [5.733713042530412]
  VAKRA is a benchmark of over $8,000$ executable APIs across $62$ domains.<n>It tests diverse API interaction styles, multi-hop reasoning over structured APIs, and multi-source reasoning with natural-language tool-use policy constraints.<n>Using a fixed ReAct harness, we evaluate frontier and open-weight models and find that even the best model achieves only 70.4% on single-hop endpoint-style tasks.
  arXiv  Detail & Related papers  (2026-08-12T17:27:27Z)
• REVES: REvision and VErification--Augmented Training for Test-Time Scaling [53.197756110943395]
  We propose a two-stage iterative framework that alternates between online data/prompt augmentation and policy optimization.<n>We observe gains of +6.5 points over the RL baseline and +4.0 points over standard multi-turn training.
  arXiv  Detail & Related papers  (2026-06-17T10:37:23Z)
• AISE-Bench: A Full-Cycle Curated Benchmark for Information Seeking on Academic Knowledge Graphs [55.8213573429699]
  AISE-Bench is a real-world, full-cycle annotated benchmark for information seeking on academic knowledge graphs.<n>AISE-Bench release contains 1,133 QA pairs, including query, full API execution, validated parameters, and source-grounded answers with reference links.<n>We develop a comprehensive evaluation protocol measuring answer quality, reference grounding, API-planning correctness, and execution success.
  arXiv  Detail & Related papers  (2026-06-16T07:59:46Z)
• APITestGenie: Generating Web API Tests from Requirements and API Specifications with LLMs [0.6445087473595952]
  APITestGenie is a novel tool that automatically generates API integration tests directly from business requirements and OpenAPI specifications.<n>The tool was able to generate syntactically and semantically valid test scripts for 89% of the business requirements under test after at most three attempts.<n>Some generated tests revealed previously unknown defects in the APIs, including integration issues between endpoints.
  arXiv  Detail & Related papers  (2026-04-02T13:43:56Z)
• Semantic Soft Bootstrapping: Long Context Reasoning in LLMs without Reinforcement Learning [46.765013720309064]
  Long context reasoning in large language models (LLMs) has demonstrated enhancement of their cognitive capabilities via chain-of-thought (CoT) inference.<n>Training such models is usually done via reinforcement learning with verifiable rewards (RLVR) in reasoning based problems, like math and programming.<n>We propose textbfSemantic Soft Bootstrapping ( SSB), a self-distillation technique, in which the same base language model plays the role of both teacher and student, but receives different semantic contexts about the correctness of its outcome at training time.
  arXiv  Detail & Related papers  (2025-12-04T18:59:18Z)
• A Systematic Evaluation of Large Code Models in API Suggestion: When, Which, and How [53.65636914757381]
  API suggestion is a critical task in modern software development. Recent advancements in large code models (LCMs) have shown promise in the API suggestion task.
  arXiv  Detail & Related papers  (2024-09-20T03:12:35Z)
• ToolACE: Winning the Points of LLM Function Calling [139.07157814653638]
  ToolACE is an automatic agentic pipeline designed to generate accurate, complex, and diverse tool-learning data.<n>We demonstrate that models trained on our synthesized data, even with only 8B parameters, achieve state-of-the-art performance on the Berkeley Function-Calling Leaderboard.
  arXiv  Detail & Related papers  (2024-09-02T03:19:56Z)
• Zephyr: Direct Distillation of LM Alignment [59.03530095974505]
  We aim to produce a smaller language model that is aligned to user intent. Previous research has shown that applying supervised fine-tuning (dSFT) on larger models significantly improves task accuracy. We apply distilled direct preference optimization (dDPO) to learn a chat model with significantly improved intent alignment.
  arXiv  Detail & Related papers  (2023-10-25T19:25:16Z)
• APICom: Automatic API Completion via Prompt Learning and Adversarial Training-based Data Augmentation [6.029137544885093]
  API recommendation is the process of assisting developers in finding the required API among numerous candidate APIs. Previous studies mainly modeled API recommendation as the recommendation task, and developers may not yet be able to find what they need. Motivated by the neural machine translation research domain, we can model this problem as the generation task. We propose a novel approach APICom based on prompt learning, which can generate API related to the query according to the prompts.
  arXiv  Detail & Related papers  (2023-09-13T15:31:50Z)
• Evaluating Embedding APIs for Information Retrieval [51.24236853841468]
  We evaluate the capabilities of existing semantic embedding APIs on domain generalization and multilingual retrieval. We find that re-ranking BM25 results using the APIs is a budget-friendly approach and is most effective in English. For non-English retrieval, re-ranking still improves the results, but a hybrid model with BM25 works best, albeit at a higher cost.
  arXiv  Detail & Related papers  (2023-05-10T16:40:52Z)
• On the Effectiveness of Pretrained Models for API Learning [8.788509467038743]
  Developers frequently use APIs to implement certain functionalities, such as parsing Excel Files, reading and writing text files line by line, etc. Developers can greatly benefit from automatic API usage sequence generation based on natural language queries for building applications in a faster and cleaner manner. Existing approaches utilize information retrieval models to search for matching API sequences given a query or use RNN-based encoder-decoder to generate API sequences.
  arXiv  Detail & Related papers  (2022-04-05T20:33:24Z)
• TicketTalk: Toward human-level performance with end-to-end, transaction-based dialog systems [10.659519248703273]
  We present a data-driven, end-to-end approach to transaction-based dialog systems. We show that the system performs at near-human levels in terms of verbal response quality and factual grounding accuracy. We introduce TicketTalk, a movie ticketing dialog dataset with 23,789 annotated conversations.
  arXiv  Detail & Related papers  (2020-12-23T02:43:37Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.