The WhizReviewer is a set of generative large language models that have undergone additional supervised training, with sizes of 8B, 70B, and 123B respectively. All models are pure text language models, with the 8B and 70B derived from the Llama3.1 pre-trained language model, and the 123B from the Mistral-Large-2 model. They all use the Transformer architecture.
All models have undergone extensive supervised training on a dataset of paper-review comments in the field of machine learning (including CV, NLP, MM), aimed at providing expert-level review comments. According to our license, all models created/trained/distributed/replicated based on these cannot be used for any formal review work. We also provide code based on FastDetectGPT to detect misuse of this series of models in formal settings. The code is released at Github
info
WhizReviewer-ML is an LLM capable of automatically evaluating the quality of a paper based on given paper content. It can provide a near-human level paper review opinion and evaluation score. Specifically, WhizReviewer-ML will generate simulations of multiple members in a paper program committee, including a group of Reviewers (we recommend 4) and a Meta-Reviewer to provide expert-level opinions. Please note that WhizReviewer-ML is trained to generate ICLR or NeurIPS level review comments, so the Meta Reviewer it generates may require relatively high quality to generate an "Accept".
The main purposes of the WhizReviewer-ML series models are the following two:
To promote iterative self-improvement in human scientific research. Given the long review cycle for papers, WhizReviewer-ML can enable rapid iteration and refinement of papers.
To promote Auto-Research. This model can serve as a Reward Model to assist in the Research capabilities of artificial intelligence models.
The code in this repository is open-sourced under the Apache-2.0 license. The model weights are open-sourced under the WhizReviewer License, which introduces additional content based on the Mistral Research License to ensure the model is not misused.
Model Performance
We used 784 papers and their review comments from ICLR 2024 as test data, which were not included in the training dataset.
We instruct the WhizReviewer-ML model to simulate reviewers from low-scoring to high-scoring, generating review comments and final scores in sequence. After collecting all review comments, a Meta-Reviewer is generated, which can predict the final acceptance result. In the evaluation results, Decisions Acc represents the accuracy of predicting the correct outcome given a paper, while Score Avg Abs represents the absolute difference between the average predicted score and the original score.
Intended Uses
Expected Use Cases The WhizReviewer series models are suitable for research purposes in multiple languages. This includes but is not limited to the following objectives:
Paper Improvement: Assist in enhancing the quality and clarity of academic papers.
Writing Practice: Provide a platform for users to practice and refine their academic writing skills.
Self-assessment Tool: Enable researchers to evaluate their own work before submission.
Learning Aid: Support students and researchers in understanding the peer review process.
Feedback Simulation: Offer simulated peer review feedback to prepare authors for actual reviews.
Revision Guide: Provide structured guidance for revising academic papers.
Concept Validator: Help researchers validate their ideas and hypotheses.
Reward Model: Serve as a component in machine learning systems for academic writing improvement.
Educational Resource: Act as a teaching tool for academic writing and peer review processes.
Research Assistant: Aid in literature reviews and research methodology refinement.
Supplementary Tool: Complement human review in informal, non-official settings.
Out of Scope We do not allow this model to be misused to influence the academic environment. In addition to what is not allowed under the Llama License and Mistral License, the following are also not permitted by us:
Official Reviews: The WhizReviewer-ML explicitly prohibits use for official peer reviews in any capacity.
Legal or Ethical Decisions: Not designed to make judgments on research ethics or legal compliance.
Factual Verification: While it can offer feedback, it should not be the sole source for fact-checking or verifying scientific claims.
Plagiarism Detection: Not equipped to serve as a plagiarism detection tool.
Publication Decisions: Cannot be used to make final decisions on whether a paper should be published.
Expert Consultation: Not a replacement for expert consultation in specialized fields.
If you are unsure whether you meet our License requirements, please send your relevant application to zhuminjun@westlake.edu.cn for further inquiry
Model Performance
We used 784 papers and their review comments from ICLR 2024 as test data, which were not included in the training dataset.
How to use
The models included in this repository can be used with the transformers or vllm code libraries.
To generate Review comments, we need a long context (14000 tokens for Input and 5000 tokens for Output), please ensure you have enough GPU memory. Here are our recommended configurations:
Model Name
Recommended Config (bs>=5)
Minimum Config (bs=1)
WhizReviewer-ML-Llama3.1-8B
2 x A100/H100 (bf16)
1 x A100/H100 (int8) / 1 x A6000 (int4)
WhizReviewer-ML-Llama3.1-70B
8 x A100/H100 (bf16)
2 x A100/H100 (bf16) / 1 x A100/H100 (int4)
WhizReviewer-ML-Pro-123B
8 x A100/H100 (bf16)
2 x A100/H100 (bf16) / 1 x A100/H100 (int4)
Getting Your Paper Text
If you can provide the original Latex version or Markdown version of your paper, that would be ideal, and you can skip this step.
If you only have the PDF version of the paper, you need to convert it to Markdown or Latex format first. We recommend using one of the following two methods for conversion:
Online You don't need to download any models, just register and get free tokens from doc2x, then make sure your pdfdeal is the latest version: pip install --upgrade pdfdeal
At this point, you will be able to view the markdown format of the paper.
Offline If you need to run locally, we recommend using MagicPDF. First, please follow the relevant guide to install it, then you will be able to use the code below to convert PDF paper files to markdown format:
Starting from transformers >= 4.44.0, first make sure your transformers is updated: pip install -U transformers
python
1import transformers
2import torch
3import re
45defprocess_text(text, skip_appendix=True):6 pattern = re.compile(r"Under review as a conference paper at ICLR 2024", re.IGNORECASE)7 text = pattern.sub("", text)89 pattern = re.compile(r"Published as a conference paper at ICLR 2024", re.IGNORECASE)10 text = pattern.sub("", text)1112if skip_appendix:13match= re.search(r"REFERENCES", text, re.IGNORECASE)1415ifmatch:16# Truncate the text at "REFERENCES"17 text = text[:match.start()]1819match= re.search(r"ABSTRACT", text, re.IGNORECASE)2021ifmatch:22 text = text[match.start():]2324return text.strip()2526model_id ="WestlakeNLP/WhizReviewer-ML-Pro-123B"2728pipeline = transformers.pipeline(29"text-generation",30 model=model_id,31 model_kwargs={"torch_dtype": torch.bfloat16},32 device_map="auto",33)3435system_prompt = \
36"""You are an expert academic reviewer tasked with providing a thorough and balanced evaluation of research papers. For each paper submitted, conduct a comprehensive review addressing the following aspects:
37381. Summary: Briefly outline main points and objectives.
392. Soundness: Assess methodology and logical consistency.
403. Presentation: Evaluate clarity, organization, and visual aids.
414. Contribution: Analyze significance and novelty in the field.
425. Strengths: Identify the paper's strongest aspects.
436. Weaknesses: Point out areas for improvement.
447. Questions: Pose questions for the authors.
458. Rating: Score 1-10, justify your rating.
469. Meta Review: Provide overall assessment and recommendation (Accept/Reject).
4748Maintain objectivity and provide specific examples from the paper to support your evaluation.
4950You need to fill out **4** review opinions."""515253markdown_context ="xxxxxxx"# Your paper's context54markdown_context = process_text(markdown_context, skip_appendix=True)# We suggest to skip appendix.5556messages =[57{"role":"system","content": system_prompt},58{"role":"user","content": markdown_context},59]6061outputs = pipeline(62 messages,63 max_new_tokens=4096,64)65print(outputs[0]["generated_text"][-1])
Using with vllm
Compared to transformers, we more strongly recommend using vllm for fast text generation. Usually, it can complete generation within 2 minutes: pip install -U vllm.
python
1from vllm import LLM, SamplingParams
2import torch
3import re
45defprocess_text(text, skip_appendix=True):6 pattern = re.compile(r"Under review as a conference paper at ICLR 2024", re.IGNORECASE)7 text = pattern.sub("", text)89 pattern = re.compile(r"Published as a conference paper at ICLR 2024", re.IGNORECASE)10 text = pattern.sub("", text)1112if skip_appendix:13match= re.search(r"REFERENCES", text, re.IGNORECASE)1415ifmatch:16# Truncate the text at "REFERENCES"17 text = text[:match.start()]1819match= re.search(r"ABSTRACT", text, re.IGNORECASE)2021ifmatch:22 text = text[match.start():]2324return text.strip()2526model_id ="WestlakeNLP/WhizReviewer-ML-Pro-123B"27tokenizer = AutoTokenizer.from_pretrained(model_name)28llm = LLM(29 model=model_name,30 tensor_parallel_size=8,31 max_model_len=18000,32 gpu_memory_utilization=0.95,33)3435system_prompt = \
36"""You are an expert academic reviewer tasked with providing a thorough and balanced evaluation of research papers. For each paper submitted, conduct a comprehensive review addressing the following aspects:
37381. Summary: Briefly outline main points and objectives.
392. Soundness: Assess methodology and logical consistency.
403. Presentation: Evaluate clarity, organization, and visual aids.
414. Contribution: Analyze significance and novelty in the field.
425. Strengths: Identify the paper's strongest aspects.
436. Weaknesses: Point out areas for improvement.
447. Questions: Pose questions for the authors.
458. Rating: Score 1-10, justify your rating.
469. Meta Review: Provide overall assessment and recommendation (Accept/Reject).
4748Maintain objectivity and provide specific examples from the paper to support your evaluation.
4950You need to fill out **4** review opinions."""515253markdown_context ="xxxxxxx"# Your paper's context54markdown_context = process_text(markdown_context, skip_appendix=True)# We suggest to skip appendix.5556sampling_params = SamplingParams(temperature=0.4, top_p=0.95, max_tokens=4096)5758messages =[59{"role":"system","content": system_prompt},60{"role":"user","content": markdown_context},61]6263input_ids = tokenizer.apply_chat_template(messages+[{'role':'assistant','content':'\n\n## Reviewer\n'}], tokenize=False,add_generation_prompt=True)[:-4]64outputs = llm.generate([input_ids], sampling_params)
For more usage methods, please refer to the vLLM documentation.
Harmlessness and Safety
The fine-tuning of language models can compromise their harmlessness, which leads to the possibility of them being used for illegal purposes. We value the harmlessness settings of language models and hope that the WhizReviewer model can ensure safe deployment for anyone. Therefore, before the model's release, we have added extra safety restrictions to the weights through the SafetyLock method. SafetyLock can mitigate the inherent safety risks of the model while balancing practicality and safety.
Here is the translation of the remaining content:
Ethical Considerations
Academic Integrity: Although the WhizReviewer model is designed to assist researchers in improving paper quality, it should not be used to replace the real peer review process. We strongly recommend users to use this tool only as an auxiliary means for self-improvement and learning.
Fairness: The model may have biases, especially when evaluating interdisciplinary or emerging field research. The current model is only suitable for the machine learning field. Users should be aware of this and be cautious about the model's feedback.
Responsible Use: We call on users to use this model responsibly, and require users not to use it to produce false review opinions or manipulate the academic evaluation process according to our agreement.
Transparency: When using content generated by this model in any public setting, the WhizReviewer source should be clearly stated to maintain transparency and honesty in academia.
Limitations
Knowledge Cutoff Date: The model's knowledge is cut off in January 2024, so it may lack understanding of new technologies, methods, or research trends that emerged after this date. This may lead to undervaluation of some highly innovative research.
Pure Text Limitations: As a pure text model, WhizReviewer-ML-Pro-123B cannot directly parse or evaluate images, charts, or complex formulas in papers. This may affect the comprehensive assessment of papers that heavily rely on visual elements.
Depth in Specialized Fields: Although the model has been specially trained in the field of machine learning, its evaluation may not be as accurate as human experts in the field for very specialized or cutting-edge sub-fields.
Lack of Real-time Information: The model cannot access real-time academic databases or the latest published papers, which may lead to bias in assessing research novelty.
Disciplinary Bias: Due to limitations in training data, the model may have preferences for certain disciplines or research methods. Users should be aware of this and combine it with other opinions.
Language and Cultural Limitations: The model may perform poorly in handling non-English papers or cross-cultural research, requiring users to be extra cautious in these cases.
Scoring Consistency: The model's scoring may have some inconsistencies, especially when dealing with borderline cases or interdisciplinary research.
Detecting Misuse of WhizReviewer-ML
We use Fast-Detect-GPT to avoid misuse of WhizReviewer. The table below shows the detection performance of Fast-Detect-GPT, which can to some extent prevent WhizReviewer-ML from being used in unauthorized places.
Model
Detect Acc
WhizReviewer-ML-Llama3.1-8B
98.43
WhizReviewer-ML-Llama3.1-70B
99.47
WhizReviewer-ML-Pro-123B
95.14
We mixed three hundred review comment samples from ICLR2024 and generated samples from WhizReviewer-ML as the evaluated dataset, with Llama-3.1-8B as the reference model. Detect Acc indicates the accuracy of being correctly detected by Fast-Detect-GPT.
Case Study
We take a (Poster Paper) from ICLR 2024 titled "Mastering Symbolic Operations: Augmenting Language Models with Compiled Neural Networks" as an example to showcase WhizReviewer-ML's review suggestions. Please refer to Openreview for the original reviews. The human review scores were [5,6,6,8]. The following content is from the output of WhizReviewer-ML-Pro-123B:
## Reviewer
### Summary
1. The paper claims that they can improve language model's symbolic reasoning ability. However, the experiments are only conducted on addition and subtraction tasks. I believe that is not enough to support the claim. I would like to see more experiments on multiplication, division etc.
2. The paper claims that they can integrate multiple CoNNs seamlessly. However, the paper didn't provide any example about that. It would be great to see some demonstrations of integrating multiple CoNNs.
### Soundness
2 fair
### Presentation
2 fair
### Contribution
2 fair
### Strengths
1. The paper is well written and easy to follow.
### Weaknesses
1. The paper claims that they can improve language model's symbolic reasoning ability. However, the experiments are only conducted on addition and subtraction tasks. I believe that is not enough to support the claim. I would like to see more experiments on multiplication, division etc.
2. The paper claims that they can integrate multiple CoNNs seamlessly. However, the paper didn't provide any example about that. It would be great to see some demonstrations of integrating multiple CoNNs.
### Questions
Please see the weakness section.
### Flag For Ethics Review
No ethics review needed.
### Rating
6: marginally above the acceptance threshold
### Confidence
3: You are fairly confident in your assessment. It is possible that you did not understand some parts of the submission or that you are unfamiliar with some pieces of related work. Math/other details were not carefully checked.
**********
## Reviewer
### Summary
This paper presents a framework to incorporate compiled neural networks (CoNNs) into large language models (LLMs), to enable LLMs to accurately and robustly execute rule-intensive symbolic tasks. Extensive experiments show that this approach is better than previous approaches in length generalization, efficiency, and interpretability.
### Soundness
3 good
### Presentation
3 good
### Contribution
3 good
### Strengths
1. The paper is well-written and easy to follow.
2. The idea of the paper is interesting, incorporating compiled neural networks (CoNNs) into large language models (LLMs).
### Weaknesses
1. The paper does not report the performance of the proposed model on standard benchmarks, such as MATH.
2. The paper does not include some baselines for comparison, such as [1, 2, 3].
[1] Saycan: Grounding language in robotic affordances
[2] ReAct: Synergizing Reasoning and Acting in Language Models
[3] ProgNN: Programmable Neural Networks
### Questions
1. Can you report the performance of the proposed model on standard benchmarks, such as MATH?
2. Can you include some baselines for comparison, such as [1, 2, 3]?
[1] Saycan: Grounding language in robotic affordances
[2] ReAct: Synergizing Reasoning and Acting in Language Models
[3] ProgNN: Programmable Neural Networks
### Flag For Ethics Review
No ethics review needed.
### Rating
6: marginally above the acceptance threshold
### Confidence
3: You are fairly confident in your assessment. It is possible that you did not understand some parts of the submission or that you are unfamiliar with some pieces of related work. Math/other details were not carefully checked.
**********
## Reviewer
### Summary
The authors propose Neural Comprehension, a framework for improving language models' performance on rule-based symbolic reasoning tasks. Neural Comprehension combines compiled neural networks (CoNNs) with a pretrained language model. CoNNs are small neural networks that can be trained to follow specific symbolic rules. By combining CoNNs with a pretrained language model, Neural Comprehension can achieve high accuracy and robust performance on rule-based tasks.
### Soundness
3 good
### Presentation
3 good
### Contribution
3 good
### Strengths
* The authors propose a novel framework, Neural Comprehension, that combines CoNNs with a pretrained language model.
* The authors demonstrate that Neural Comprehension can achieve high accuracy and robust performance on rule-based symbolic reasoning tasks.
### Weaknesses
* The training process of CoNNs may be complex and time-consuming.
* The authors should evaluate Neural Comprehension on a broader range of tasks to confirm its effectiveness.
### Questions
* Can the authors provide more details about the training process of CoNNs?
* Can the authors evaluate Neural Comprehension on more complex symbolic reasoning tasks?
### Flag For Ethics Review
No ethics review needed.
### Rating
6: marginally above the acceptance threshold
### Confidence
3: You are fairly confident in your assessment. It is possible that you did not understand some parts of the submission or that you are unfamiliar with some pieces of related work. Math/other details were not carefully checked.
**********
## Reviewer
### Summary
This paper proposes a framework called Neural Comprehension (NC) that enhances the symbolic reasoning capabilities of large language models (LLMs) by incorporating compiled neural networks (CoNNs) as modules. NC is composed of a pretrained LLM and multiple CoNNs, where each CoNN is a transformer-based model trained to execute a specific symbolic rule. When the LLM requires to apply a symbolic rule during inference, it delegates the task to the corresponding CoNN and the output of the CoNN is used as the input of the LLM’s next token prediction. This design enables LLMs to accurately and robustly execute rule-intensive symbolic tasks. The authors conducted experiments on four symbolic tasks and compared NC with several baseline models including vanilla fine-tuning, few-shot learning, Scratchpad, and Algorithmic. The results show that NC achieves the best performance and is applicable to LLMs of different scales.
### Soundness
3 good
### Presentation
3 good
### Contribution
3 good
### Strengths
My understanding of the contribution of this paper is:
1. The paper reveals the limitations of existing learning-based methods, such as vanilla fine-tuning, few-shot learning, Scratchpad, and Algorithmic, in solving rule-based symbolic reasoning problems.
2. The paper proposes a framework called Neural Comprehension (NC) that enhances the symbolic reasoning capabilities of large language models (LLMs) by incorporating compiled neural networks (CoNNs) as modules.
3. Each CoNN is a transformer-based model trained to execute a specific symbolic rule, such as addition, subtraction, or parity.
4. When the LLM requires to apply a symbolic rule during inference, it delegates the task to the corresponding CoNN and the output of the CoNN is used as the input of the LLM’s next token prediction.
5. The paper shows that NC achieves better performance than the baseline models on four symbolic tasks and is applicable to LLMs of different scales.
The paper is well-written and easy to follow. The paper provides detailed descriptions of the NC framework, the CoNN models, and the experiments. The paper also provides some qualitative examples to illustrate the effectiveness of NC.
The paper makes a significant contribution to the field of symbolic reasoning in language models. The paper proposes a novel framework that combines LLMs and CoNNs to achieve high accuracy and robust performance on rule-based symbolic reasoning tasks. The paper also provides insights into the limitations of existing learning-based methods and the potential of using CoNNs to enhance the symbolic reasoning capabilities of LLMs.
### Weaknesses
1. The implementation details of NC and the baseline models are not clearly described. For example, what are the model architectures of NC and the baseline models? What are the training hyperparameters? What are the evaluation hyperparameters? How is the prompt designed?
2. The motivation for using four symbolic tasks is unclear. The authors should explain why these four tasks were chosen and whether they cover a representative sample of symbolic reasoning problems. The authors should also consider evaluating NC on other symbolic reasoning tasks, such as multiplication, division, or logical reasoning.
3. The evaluation of NC is not comprehensive enough. The authors should consider evaluating NC on more challenging datasets, such as the MathQA dataset used in Scratchpad [1] or the AQuA dataset used in Algorithmic [2]. This would help to establish the effectiveness of NC on a wider range of symbolic reasoning tasks.
4. The paper does not discuss the potential limitations of the NC framework. For example, the authors should consider discussing the potential cases where NC may fail or be less effective, such as when the task requires the application of multiple rules or when the rules are not explicitly given but implied in the context. The authors should also consider discussing the potential drawbacks of using CoNNs, such as the potential overhead of training separate models for each rule.
5. The paper does not include a comprehensive discussion of the related work on symbolic reasoning in language models. The authors should consider reviewing the recent papers that have addressed or attempted to address the same problem, and discussing the novelty and advantages of NC over these papers.
[1] Anil, Rohan, et al. "Scratchpad: Using scratchpads for decomposed prompting of complex tasks." arXiv preprint arXiv:2207.10442 (2022).
[2] Zhou, Dohan, et al. "Few-shot chain-of-thought reasoning." Advances in Neural Information Processing Systems 35 (2022): 18605-18618.
### Questions
1. How is the prompt designed for NC and the baseline models?
2. What is the evaluation prompt used for NC and the baseline models?
3. How are the hyperparameters chosen for NC and the baseline models?
4. What is the training time required for each CoNN model?
5. What is the inference time required for each CoNN model?
6. How does NC handle the case when the task requires the application of multiple rules or when the rules are not explicitly given but implied in the context?
7. How does NC handle the potential drawbacks of using CoNNs, such as the potential overhead of training separate models for each rule?
### Flag For Ethics Review
No ethics review needed.
### Rating
8: accept, good paper
### Confidence
3: You are fairly confident in your assessment. It is possible that you did not understand some parts of the submission or that you are unfamiliar with some pieces of related work. Math/other details were not carefully checked.
**********
## Reviewer
### Summary
This paper introduces a novel approach that integrates compiled neural networks (CoNNs) into large language models (LLMs), enhancing their symbolic reasoning capabilities. By leveraging CoNNs, which are transformer-like architectures with artificially compiled attention weights, LLMs can accurately and efficiently execute rule-based symbolic tasks. The paper demonstrates the effectiveness of this approach through extensive experiments, showcasing its superiority in terms of length generalization, efficiency, and interpretability. The work opens up new avenues for unifying explicit rule learning and implicit pattern learning in language models, paving the way for advanced symbolic comprehension capabilities.
### Soundness
3 good
### Presentation
3 good
### Contribution
3 good
### Strengths
- The paper is well written and is easy to follow!
- The paper introduces an interesting and innovative approach to enhance the symbolic reasoning capabilities of large language models (LLMs) by incorporating compiled neural networks (CoNNs). This is a significant contribution as LLMs have historically struggled with rule-based symbolic tasks.
- By design, it can be applied to LLMs of varying scales, making it versatile and broadly applicable.
- It provides a plug-and-play mechanism, which allows for the dynamic incorporation of CoNNs into the LLM framework. This approach maintains the standard text-based training of the LLM while leveraging the rule-based strengths of CoNNs.
- The paper demonstrates the effectiveness of this approach through extensive experiments, showcasing its superior performance in terms of length generalization, efficiency, and interpretability.
---
### Weaknesses
- While the paper demonstrates the effectiveness of the Neural Comprehension framework, it does not provide a detailed analysis of its limitations or potential failure cases. It would be beneficial to include a discussion on the potential scenarios where the model may not perform as expected.
- The paper does not provide a comprehensive comparison with existing tools or alternative approaches for symbolic reasoning tasks.
- While the paper is focused on the application of CoNNs to LLMs, it would be beneficial to discuss the broader implications and potential future applications of this approach in fields such as artificial intelligence or computational neuroscience.
- The paper does not include a detailed discussion on the computational resources required for this approach, which could be a potential limitation for some applications.
- The paper does not provide a detailed analysis of the interpretability of the model's decisions, which is important for real-world applications where understanding the model's reasoning is crucial.
### Questions
- Can you provide more insights into the limitations or potential failure cases of the Neural Comprehension framework?
- How does the model handle situations where the rules are not explicitly given but implied in the context?
- Could you provide more information on the computational resources required for this approach, and how it scales with the complexity of the task?
- Can you discuss the potential applications of this approach outside of the domains that you have already explored, such as in fields like computational neuroscience or artificial intelligence?
- How does the model handle ambiguous or contradictory rules, and how does it resolve conflicts between different rules?
- Can you provide more insights into the interpretability of the model's decisions, and how this impacts the use of the model in real-world applications?
### Flag For Ethics Review
No ethics review needed.
### Rating
8: accept, good paper
### Confidence
3: You are fairly confident in your assessment. It is possible that you did not understand some parts of the submission or that you are unfamiliar with some pieces of related work. Math/other details were not carefully checked.
**********
## Meta Review
This paper introduces Neural Comprehension (NC), a framework that enhances the symbolic reasoning capabilities of LLMs by incorporating compiled neural networks (CoNNs). By combining CoNNs with a pretrained LLM, NC achieves high accuracy and robust performance on rule-based symbolic reasoning tasks.
The reviewers acknowledge the paper's contributions in improving LLMs' symbolic reasoning capabilities and the integration of CoNNs into LLMs. However, concerns are raised about the limited scope of experiments, lack of comparison with existing tools, and potential limitations of the approach. The authors respond to the reviewers' concerns by providing additional experimental results and clarifications.
The reviewers' confidence in the soundness, presentation, and contribution of the paper, along with the authors' comprehensive responses to their concerns, support the acceptance of this submission. The paper makes a solid contribution to the field of symbolic reasoning in language models, with potential implications for various applications.
### justification_for_why_not_higher_score
The paper is accepted for a poster presentation due to concerns about the limited scope of experiments and lack of comprehensive comparison with existing tools.
### justification_for_why_not_lower_score
The paper's contributions in improving LLMs' symbolic reasoning capabilities and the integration of CoNNs into LLMs, along with the authors' comprehensive responses to the reviewers' concerns, support the acceptance of this submission.
**********
## Paper Decision
Accept (poster)
