Pith. sign in

REVIEW 3 major objections 6 minor 32 references

FASTNav: Fine-tuned Adaptive Small-language-models Trained for Multi-point Robot Navigation

T0 review · 3 major / 6 minor · reviewed 2026-08-12 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read FASTNav claims that fine-tuned sub-3B language models, boosted by teacher-student iteration, reach near-GPT-4 accuracy on multi-point robot navigation while staying lightweight enough to run on edge devices.

desk verdict Plausible recipe for edge LLM navigation, but the paper must prove its iteration loop isn't tuning prompts on the test set before the headline numbers mean anything. read the letter →

arxiv 2411.13262 v1 pith:ATUURZSS submitted 2024-11-20 cs.RO cs.AIcs.HC

Yuxuan Chen , Yixin Han , Xiao Li This is my paper · ORCID
classification cs.ROcs.AIcs.HC
keywords smalllanguagemodelsrobotnavigationLoRAfine-tuningteacher-studentiterationedgedeploymentmulti-pointmodelcompressionJSONoutputformat
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper claims that small language models (SLMs) with under three billion parameters can be boosted through LoRA fine-tuning plus a teacher-student iteration loop to perform multi-point robot navigation almost as accurately as much larger models such as GPT-4, while remaining fast enough to run locally on an edge computer. The motivation is that relying on cloud API calls for robot navigation raises latency, privacy, and network-availability problems, so a locally deployable model that still solves complex navigation instructions would be practically valuable. The paper reports that in a hospital simulation, a 3B model with FASTNav reached a 70% success rate, outperforming quantized 7-8B models, and a 0.5B model reached 63.33%, with inference times around 3 seconds. It also demonstrates the same trend on a real wheel-footed robot in two indoor environments.

What carries the argument

The machinery is a three-module pipeline: (1) LoRA fine-tuning of a sub-3B model on a human-in-the-loop generated dataset, with outputs constrained to JSON containing an explanation and an ordered list of x–y goal coordinates; (2) a teacher-student iteration in which a large model (GPT-4) acts as prompt engineer and feedback receiver, generating prompts for the student and feeding the student's mistakes back into the next round of prompts; and (3) a Navigation2 controller that executes the coordinate list. The load-bearing identity is the mapping f(W, M) = {p1, ..., pn} from a natural-language command and a static text map to an ordered waypoint sequence.

What would settle it

Shift 20% of landmark coordinates in the hospital simulation between training and testing, and measure the success rate; if the model relied on memorized static positions rather than reasoning from the map, success should collapse. Alternatively, issue a command whose target is described by a visual attribute absent from the text map (e.g., 'go to the red chair') and check whether the model can produce a coordinate at all.

Watch

Extended reading notes

Core claim

The central discovery is that a small language model fine-tuned on a domain-specific dataset of navigation tasks and then iteratively guided by a larger teacher model can close most of the performance gap with the teacher on the specific task of translating natural-language instructions into ordered coordinates. In the paper's experiments, fine-tuning constrains the output into a JSON format and encodes map knowledge, raising accuracy sharply, while the teacher-student iteration adds a further 30–40% success-rate improvement by having the teacher rewrite prompts and provide feedback on previous failures. As a result, openllama-3B reaches 70% success and tau-0.5B reaches 63.33% in simulation, compared with 36.67% for Llama3-8B-GPTQ, and the small models keep memory usage at 2–8 GB and inference below three seconds.

Load-bearing premise

The method assumes that multi-point navigation can be fully captured by outputting an ordered list of coordinates read from a static text map of landmarks, so if landmarks move, the map is incomplete, or the robot needs to see a target to identify it, the approach fails regardless of the language model's quality.

Editorial extensions

If this is right

  • If FASTNav is correct, sub-3B models deployed on edge devices can handle multi-point navigation tasks that currently require API calls to GPT-4-scale models, cutting both response time and privacy exposure.
  • The same fine-tuning-plus-iteration recipe may transfer to other structured output tasks where a small model must produce a constrained format (JSON, PDDL, STL) from natural language, making local deployment feasible beyond navigation.
  • Quantization and distillation baselines (GPTQ on Llama3-8B and Mistral-7B, distilled MiniMA-3B) are outperformed by a smaller model fine-tuned for the task, suggesting that task-specific adaptation matters more than raw parameter count or generic compression.
  • Because the teacher is only needed during training, the deployed system has no dependency on the cloud at inference time, so network failure does not disable navigation.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The coordinate-list abstraction implicitly assumes that navigation goals are fully describable by static map coordinates; a testable extension would be a multimodal version that takes camera images as additional input, which the paper's limitations section names as future work.
  • The method's edge-deployment advantage applies to inference, not training, since both the human-in-the-loop dataset generation and the GPT-4 teacher require access to a strong external model during the build phase.
  • One could probe the transferability of FASTNav by training in the hospital simulation and fine-tuning in the lab hallway with only 100 tasks; if the performance gain persists, the method generalizes beyond a single memorized map, and if not, it is tightly coupled to the training environment.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 6 minor

Summary. FASTNav proposes a pipeline for boosting small language models (SLMs) for multi-point robot navigation, consisting of (1) LoRA fine-tuning on a human-in-the-loop generated dataset, (2) a teacher-student iteration where GPT-4 acts as a prompt engineer that receives success feedback, and (3) a Navigation2-based controller that executes the model's output as an ordered list of coordinates. The paper reports simulation experiments on a hospital environment, comparisons against quantization/distillation baselines and LLM-based planning methods, ablations of the fine-tuning and iteration modules, and real-robot deployments on a DIABLO platform with a Jetson Orin NX. The central claim is that fine-tuned and iterated SLMs (e.g., openllama-3B, tau-0.5B) achieve success rates close to or above those of much larger models—a 4-fold accuracy increase over the base SLMs—while remaining lightweight and locally deployable.

Significance. If the reported results hold, FASTNav would be a practically useful contribution: it demonstrates that a sub-3B model can carry out multi-point navigation with high success after task-specific fine-tuning and iterative prompt refinement, and it validates the approach on real hardware. The paper ships a concrete pipeline with code-level descriptions (via PEFT and Navigation2), real-robot evaluations, and ablations that decompose the contributions of fine-tuning and iteration. These are strengths: the work addresses an actual deployment problem (privacy, latency, edge compute) and provides falsifiable comparisons. However, the significance is currently conditional on resolving evaluation-validity concerns, because the teacher-student loop as described may optimize on the test set and the headline comparisons against compression baselines are not apples-to-apples.

major comments (3)
  1. [§V-B, Algorithm 1, Figure 4] The teacher-student iteration procedure as written may leak test-set information into prompt optimization. Algorithm 1 takes expected goals G as input and, inside the loop, sets success = 1 if result[positions] == Gi, then calls UpdateFeedback(result, success). If this feedback is computed for test tasks, the teacher (GPT-4) receives per-task ground-truth success signals for the evaluation set and can adjust prompts accordingly. Figure 4's caption explicitly states that the plotted curves show accuracy on the test set during fine-tuning and iteration. The paper never states that the iteration phase is restricted to a training split and that test tasks are never shown to the teacher. If the reported 30–40% SR improvement from iteration (Figure 6b) is obtained by per-instance prompt optimization on the test set, the central claim of student generalization is not supported. Please clarify whether the iteration curves and Table II/Figure 6 results reflect held-out tasks, and if so, report the exact data-splitting protocol. If iteration was performed on the test set, the experiments must be redone on a separate validation split.
  2. [§VI-A, Table II] The comparison in Table II is not a controlled comparison of methods: FASTNav models receive task-specific LoRA fine-tuning and teacher-student iteration, whereas the compression baselines (Llama3-8B-GPTQ, Mistral-7B-GPTQ, Qwen-7B-Chat-Int4, distilled-MiniMA-3B) are used off-the-shelf with no task adaptation. The reported ≥30% success-rate gap therefore conflates the proposed method with the fact that the baselines were not adapted to the navigation task. A fair comparison would fine-tune the compressed models on the same dataset, or evaluate FASTNav without the fine-tuning/iteration modules, and then compare. As written, the claim that FASTNav 'improves on other model compression methods' is overstated because it compares an adapted system against unadapted checkpoints.
  3. [§V-A, §V-B, Figure 4] The evaluation of 'close to GPT-4' is partly self-referential: GPT-4 is used to generate the fine-tuning dataset (Section V-A), to act as the teacher that produces prompts and receives feedback (Section V-B), and as the reference model in Figure 4. Because the test set is human-designed and success is judged by navigation outcomes, this is not a fatal circularity, but the 'close to GPT-4' claim measures how well the student mimics a teacher on tasks that the same model family helped shape. The paper should state what independent evidence supports the absolute accuracy claim, and should report whether the test tasks were never seen by the teacher during iteration.
minor comments (6)
  1. [§VI-A, Table II] The success-rate values in Table II (e.g., 36.67, 23.33, 6.67) appear to be based on only 30 test tasks; no confidence intervals or statistical tests are reported. Please add error bars or exact counts (e.g., 11/30) for the main comparisons.
  2. [Table I] The column 'Performance (%)' is undefined; specify which benchmark or metric it refers to, and add the evaluation condition (e.g., zero-shot, few-shot).
  3. [§V-B, Algorithm 1] The pseudo-code does not define the functions ReadFeedback, GeneratePrompt, Completion, and UpdateFeedback in terms of inputs and outputs, and it is unclear how the teacher selects prompts across iterations. Providing a formal specification would help reproducibility.
  4. [§VI-A] The dataset sizes are given as 'approximately 1400' and 'about 100' tasks; please report exact numbers, the train/test split, and the number of tasks per complexity bin (single-, two-, three-, multi-goal).
  5. [Abstract] There is a typographical spacing issue in 'F ASTNav' in the abstract; please fix to 'FASTNav'.
  6. [§VI-B, Equation (4)] The Navigation Error metric as written averages distances over goals, but the normalization (e.g., per task, per goal) is not specified; clarify how d(ĝ_i, g_i) is aggregated when tasks have different numbers of goals.

Circularity Check

1 steps flagged · score 6.0 of 10

Teacher–student iteration feeds test-set ground-truth goals back into prompt optimization, so the reported accuracy gains are partially fitted rather than predicted.

  1. fitted input called prediction [Section V-B, Algorithm 1; Figure 4 caption, Section VI-B]
    "Algorithm 1: 'Inputs: tasks T ; environment information E; names of models N; expected goals G' ... 'if result[positions] == Gi then success = 1 ... UpdateFeedback(result, success)' ... Figure 4 caption: 'This graph shows the accuracy of five models and GPT4 on the test set during fine-tuning and iteration processes.'"

    The success criterion inside the iteration loop is the same quantity reported as test accuracy: Algorithm 1 takes expected goals G as input and calls UpdateFeedback(result, success) whenever result[positions] == Gi, and Figure 4 reports accuracy on the test set during the iteration process. Since Algorithm 1 contains no parameter or weight update for the student, the reported test-set accuracy can increase during iteration only by having the teacher rewrite prompts for those same test tasks using their Gi as feedback.

full rationale

The central success-rate claim is partially circular by construction. Algorithm 1 takes the expected goals G as inputs, computes success by comparing the student's output positions with Gi, and feeds that success back to the teacher, which rewrites the prompt for the same task. Because the loop updates no student weights, the accuracy curve in Figure 4, reported on the test set during fine-tuning and iteration, is best read as per-task prompt optimization against the test tasks' ground-truth goals. The post-iteration numbers supporting the '4-fold increase' and 'close to GPT-4' headlines therefore are fitted on the test set rather than held-out predictions. The paper never states that iteration is confined to a training split, and the Algorithm 1 pseudocode makes no such restriction. Other parts of the paper are not circular: fine-tuning uses a separately human-scored dataset, the test set is human-designed, and the comparison baselines and edge-deployment measurements are independent. There is no load-bearing self-citation: the only author-overlapping reference, AutoTAMP, is used as a comparison baseline, not as justification for the method. The use of GPT-4 as both teacher and reference model is not itself circular because final success is judged against ground-truth coordinates, not against GPT-4 outputs. Despite those independent elements, the leaked iteration loop compromises the reported accuracy improvements, so the empirical headline reduces in part to the test-set feedback signal.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The central claim rests on domain assumptions about map stability, text-only world representation, the effectiveness of GPT-4's prompts, and the reliability of the Navigation2 controller. No new physical entities are introduced. The fine-tuning and iteration hyperparameters that the results depend on are neither reported nor released, which weakens reproducibility rather than adding circularity.

free parameters (3)
  • LoRA rank r = Not reported.
    The LoRA rank is central to the fine-tuning configuration in Section V-A, but the paper only states r << min(d,k) and never gives the value used in experiments.
  • Fine-tuning learning rate and number of epochs = Not reported.
    Section VI-A says parameters such as learning rate and epochs were adjusted for comparison, but no values are reported, so the training setup is not reproducible.
  • Teacher-student iteration count max(iter) = Not reported.
    Algorithm 1 loops until max(iter), but max(iter) is not specified; Figures 4 and 8 show iteration curves without exact endpoint counts.
assumptions (4)
  • domain assumption Most objects' positions on the map, including landmarks, remain stable; only small dynamic obstacles are allowed.
    Stated in Section IV; the SLM is trained to output coordinates from a static map, so changing landmark positions would invalidate the outputs.
  • domain assumption The map M can be fully described by named landmarks with coordinates and attributes, and this text representation is sufficient for the navigation task.
    Used throughout Section V; the model receives map landmarks as text and outputs coordinates, ignoring visual or sensor data.
  • domain assumption GPT-4's generated prompts and feedback can effectively coach the fine-tuned SLM to improve task accuracy.
    The teacher-student iteration in Section V-B assumes the teacher's feedback is useful and transferable; no analysis of prompt quality is provided.
  • domain assumption The low-level Navigation2 controller reliably executes the coordinate waypoints produced by the language model.
    Section V-C delegates motion to Navigation2; if waypoint following fails, overall success rate drops independent of the SLM.

how reviews work

0 comments
Cite this review

Pith. "Pith review of FASTNav: Fine-tuned Adaptive Small-language-models Trained for Multi-point Robot Navigation." pith.science (2026). https://pith.science/paper/ATUURZSS

@misc{pith2026241113262,
  author       = {Pith},
  title        = {Pith review of: FASTNav: Fine-tuned Adaptive Small-language-models Trained for Multi-point Robot Navigation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ATUURZSS}},
  note         = {Machine review of arXiv:2411.13262}
}
read the original abstract

With the rapid development of large language models (LLM), robots are starting to enjoy the benefits of new interaction methods that large language models bring. Because edge computing fulfills the needs for rapid response, privacy, and network autonomy, we believe it facilitates the extensive deployment of large models for robot navigation across various industries. To enable local deployment of language models on edge devices, we adopt some model boosting methods. In this paper, we propose FASTNav - a method for boosting lightweight LLMs, also known as small language models (SLMs), for robot navigation. The proposed method contains three modules: fine-tuning, teacher-student iteration, and language-based multi-point robot navigation. We train and evaluate models with FASTNav in both simulation and real robots, proving that we can deploy them with low cost, high accuracy and low response time. Compared to other model compression methods, FASTNav shows potential in the local deployment of language models and tends to be a promising solution for language-guided robot navigation on edge devices.

Figures

Figures reproduced from arXiv: 2411.13262 by the authors.

Figure 1
Figure 1. FASTNav is a method for boosting SLMs for robot navigation. We improve the performance of SLMs for multi-point robot navigation by fine-tuning and teacher-student iteration. demonstration videos. By reducing the size and complexity of large models, model compression methods can significantly decrease computational load and inference time, offering a promising solution to address those inefficiencies. Addition￾ally, … view at source ↗
Figure 2
Figure 2. Architecture of FASTNav. The architecture contains three components: (1) fine-tuning of SLMs based on PEFT: we use PEFT to give SLMs access to the necessary environment information and to constrain their output format; (2) teacher-student iteration: the fine￾tuned SLMs cooperate with teacher models to solve navigation tasks, where the teacher serves as a prompt engineer and feedback receiver, transferring knowledge … view at source ↗
Figure 3
Figure 3. Human-in-the-loop generation of the dataset. [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: The accuracy curve during fine-tuning and iteration. [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 5
Figure 5. Figure 5: The sizes of the environments used in our experiments. [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]
Figure 6
Figure 6. Figure 6: Ablation experiments. (a) FASTNav method and fine￾tuning-only method. Through FASTNav, models apparently have better SR than those only fine-tuned. (b) FASTNav method and iteration-only method. SLMs without fine-tuning perform terribly compared to SLMs with FASTNav. SL…
Figure 7
Figure 7. Figure 7: Performance of the language model on one task under different conditions. [PITH_FULL_IMAGE:figures/full_fig_p007_7.png]
Figure 8
Figure 8. Figure 8: Experiments on a real robot. After being processed by FASTNav, TinyLlama-1.1B shows higher SR and better per￾formance on a real robot than before adjustments. Increasing the number of iterations also enhances the model’s capabilities. We conducted experiments in two re…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

32 extracted references · 13 canonical work pages

  1. [1]

    Lm-nav: Robotic navigation with large pre-trained models of language, vision, and action,

    D. Shah, B. Osinski, B. Ichter, and S. Levine, “Lm-nav: Robotic navigation with large pre-trained models of language, vision, and action,” Conference on Robot Learning, 2022

    work page 2022

  2. [2]

    Do as i can, not as i say: Grounding language in robotic affordances,

    M. Ahn, A. Brohan, and N. B. et al., “Do as i can, not as i say: Grounding language in robotic affordances,” Conference on Robot Learning, pp. 287–318, 2022

    work page 2022

  3. [3]

    Visual language maps for robot navigation,

    C. Huang, O. Mees, A. Zeng, and W. Burgard, “Visual language maps for robot navigation,” 2023 IEEE International Conference on Robotics and Automation (ICRA), pp. 10 608–10 615, 2022

    work page 2023

  4. [4]

    Lora: Low-rank adaptation of large language models,

    J. E. Hu, Y . Shen, P. Wallis, Z. Allen-Zhu, Y . Li, S. Wang, and W. Chen, “Lora: Low-rank adaptation of large language models,” ArXiv, vol. abs/2106.09685, 2021

    arXiv 2021

  5. [5]

    Vision-only robot navigation in a neural radiance world,

    M. Adamkiewicz, T. Chen, A. Caccavale, R. Gardner, P. Culbertson, J. Bohg, and M. Schwager, “Vision-only robot navigation in a neural radiance world,” IEEE Robotics and Automation Letters, vol. 7, no. 2, pp. 4606–4613, 2022

    work page 2022

  6. [6]

    Combination and competition between path integration and landmark navigation in the estimation of heading direction,

    S. K. Harootonian, A. D. Ekstrom, and R. C. Wilson, “Combination and competition between path integration and landmark navigation in the estimation of heading direction,” PLOS Computational Biology, vol. 18, pp. 1–26, 02 2022

    work page 2022

  7. [7]

    Olfaction-based mobile robot navigation,

    “Olfaction-based mobile robot navigation,” Thin Solid Films, vol. 418, no. 1, pp. 51–58, 2002

    work page 2002

  8. [8]

    Navigation with large language models: Semantic guesswork as a heuristic for planning,

    D. Shah, M. R. Equi, B. Osi ´nski, F. Xia, B. Ichter, and S. Levine, “Navigation with large language models: Semantic guesswork as a heuristic for planning,” in Proceedings of The 7th Conference on Robot Learning, ser. Proceedings of Machine Learning Research, J. Tan, M. Toussaint, and K. Darvish, Eds., vol. 229. PMLR, 06–09 Nov 2023, pp. 2683–2699

    work page 2023

Show all 32 references
  1. [9]

    Srlm: Human-in-loop interactive social robot navigation with large language model and deep reinforcement learning,

    W. Wang, L. Mao, R. Wang, and B.-C. Min, “Srlm: Human-in-loop interactive social robot navigation with large language model and deep reinforcement learning,” 2024

    2024

  2. [10]

    Towards learning a generic agent for vision-and-language navigation via pre-training,

    W. Hao, C. Li, X. Li, L. Carin, and J. Gao, “Towards learning a generic agent for vision-and-language navigation via pre-training,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), June 2020

    2020

  3. [11]

    Llm+p: Empowering large language models with optimal planning proficiency,

    B. Liu, Y . Jiang, X. Zhang, Q. Liu, S. Zhang, J. Biswas, and P. Stone, “Llm+p: Empowering large language models with optimal planning proficiency,” ArXiv, vol. abs/2304.11477, 2023

    2023 arXiv

  4. [12]

    Dynamic planning with a llm,

    G. Dagan, F. Keller, and A. Lascarides, “Dynamic planning with a llm,” ArXiv, vol. abs/2308.06391, 2023

    2023 arXiv

  5. [13]

    Autotamp: Autoregressive task and motion planning with llms as translators and checkers,

    Y . Chen, J. Arkin, Y . Zhang, N. A. Roy, and C. Fan, “Autotamp: Autoregressive task and motion planning with llms as translators and checkers,” 2024 IEEE International Conference on Robotics and Automation (ICRA), pp. 6695–6702, 2023

    2024

  6. [14]

    Pruning meets low-rank parameter-efficient fine-tuning,

    M. Zhang, H. Chen, C. Shen, Z. Yang, L. Ou, X. Yu, and B. Zhuang, “Pruning meets low-rank parameter-efficient fine-tuning,” ArXiv, vol. abs/2305.18403, 2023

    2023 arXiv

  7. [15]

    Llm-pruner: On the structural pruning of large language models,

    X. Ma, G. Fang, and X. Wang, “Llm-pruner: On the structural pruning of large language models,” ArXiv, vol. abs/2305.11627, 2023

    2023 arXiv

  8. [16]

    Knowledge distillation of large language models,

    Y . Gu, L. Dong, F. Wei, and M. Huang, “Knowledge distillation of large language models,” ArXiv, vol. abs/2306.08543, 2023

    2023 arXiv

  9. [17]

    In-context learning dis- tillation: Transferring few-shot learning ability of pre-trained language models,

    Y . Huang, Y . Chen, Z. Yu, and K. McKeown, “In-context learning dis- tillation: Transferring few-shot learning ability of pre-trained language models,” ArXiv, vol. abs/2212.10670, 2022

    2022 arXiv

  10. [18]

    Gptq: Accu- rate post-training quantization for generative pre-trained transformers,

    E. Frantar, S. Ashkboos, T. Hoefler, and D. Alistarh, “Gptq: Accu- rate post-training quantization for generative pre-trained transformers,” ArXiv, vol. abs/2210.17323, 2022

    2022 arXiv

  11. [19]

    Llm-qat: Data-free quantization aware training for large language models,

    Z. Liu, B. O ˘guz, C. Zhao, E. Chang, P. Stock, Y . Mehdad, Y . Shi, R. Krishnamoorthi, and V . Chandra, “Llm-qat: Data-free quantization aware training for large language models,” ArXiv, vol. abs/2305.17888, 2023

    2023 arXiv

  12. [20]

    Zeroquant-fp: A leap forward in llms post- training w4a8 quantization using floating-point formats,

    X. Wu, Z. Yao, and Y . He, “Zeroquant-fp: A leap forward in llms post- training w4a8 quantization using floating-point formats,” ArXiv, vol. abs/2307.09782, 2023

    2023 arXiv

  13. [21]

    Orca 2: Teaching small language models how to reason,

    A. Mitra, L. D. Corro, and S. M. et al., “Orca 2: Teaching small language models how to reason,” ArXiv, vol. abs/2311.11045, 2023

    2023 arXiv

  14. [22]

    Small language models fine-tuned to coordinate larger language models improve complex reasoning,

    G. Juneja, S. Dutta, S. Chakrabarti, S. Manchanda, and T. Chakraborty, “Small language models fine-tuned to coordinate larger language models improve complex reasoning,” in Conference on Empirical Methods in Natural Language Processing, 2023

    2023

  15. [23]

    Peft: State-of-the-art parameter-efficient fine-tuning methods,

    S. Mangrulkar, S. Gugger, L. Debut, Y . Belkada, S. Paul, and B. Bossan, “Peft: State-of-the-art parameter-efficient fine-tuning methods,” https:// github.com/huggingface/peft, 2022

    2022

  16. [24]

    Grimoire is all you need for enhancing large language models,

    D. Chen, S. Song, Q. Yu, Z. Li, W. Wang, F. Xiong, and B. Tang, “Grimoire is all you need for enhancing large language models,” ArXiv, vol. abs/2401.03385, 2024

    2024 arXiv

  17. [25]

    Metagpt: Meta programming for multi-agent collaborative framework,

    S. Hong, X. Zheng, and J. P. C. et al., “Metagpt: Meta programming for multi-agent collaborative framework,” ArXiv, vol. abs/2308.00352, 2023

    2023 arXiv

  18. [26]

    Autogen: Enabling next-gen llm applications via multi-agent conversation framework,

    Q. Wu, G. Bansal, J. Zhang, Y . Wu, S. Zhang, E. Zhu, B. Li, L. Jiang, X. Zhang, and C. Wang, “Autogen: Enabling next-gen llm applications via multi-agent conversation framework,” ArXiv, vol. abs/2308.08155, 2023

    2023 arXiv

  19. [27]

    The marathon 2: A navigation system,

    S. Macenski, F. J. P. Mart’in, R. White, and J. G. Clavero, “The marathon 2: A navigation system,” 2020 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pp. 2718–2725, 2020

    2020

  20. [28]

    aws-robomaker-hospital-world,

    A. RoboMaker, “aws-robomaker-hospital-world,” 2000. [Online]. Avail- able: https://github.com/aws-robotics/aws-robomaker-hospital-world

    2000

  21. [29]

    Llama 3 model card,

    AI@Meta, “Llama 3 model card,” 2024

    2024

  22. [30]

    Mistral 7b,

    A. Q. Jiang, A. Sablayrolles, and A. M. et al., “Mistral 7b,” ArXiv, vol. abs/2310.06825, 2023

    2023 arXiv

  23. [31]

    Qwen technical report,

    J. Bai, S. Bai, and Y . C. et al., “Qwen technical report,” ArXiv, vol. abs/2309.16609, 2023

    2023 arXiv

  24. [32]

    Towards the law of capacity gap in distilling language models,

    C. Zhang, D. Song, Z. Ye, and Y . Gao, “Towards the law of capacity gap in distilling language models,” ArXiv, vol. abs/2311.07052, 2023

    2023 arXiv

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.