**Summary:**
The paper introduces InvMSAFold, a novel approach to inverse protein folding that aims to generate diverse amino acid sequences for a given protein structure. The method uses an encoder-decoder architecture, where a structure is encoded, and the complete set of parameters for a lightweight model is generated, which is trained end-to-end to capture the broader probability distribution of sequences corresponding to a specific fold. The lightweight model, based on the pairwise family, reduces the number of parameters, boosting training efficiency. The model is validated through various out-of-sample tests, showing its ability to generate diverse sequences, capturing evolutionary patterns, and enabling the sampling of a wider range of protein properties.

**Strengths:**
- **Innovative Approach:** The use of a lightweight model and a single forward pass for sequence generation makes the method computationally efficient and scalable.
- **Broad Probability Distribution:** The model captures a broader probability distribution of sequences, allowing for the generation of diverse sequences.
- **Validation Through Experiments:** The paper presents a comprehensive validation through various experiments, including covariance reconstruction, sequence patterns, protein property sampling, and structure prediction, demonstrating the model's effectiveness.

**Weaknesses:**
- **Limited Evaluation on Real-World Data:** While the paper validates the method through synthetic and simulated data, there is a lack of comprehensive evaluation on real-world protein structures and sequences.
- **Training Data Dependence:** The effectiveness of the method relies heavily on the training data, which might limit its generalizability to structures not seen during training.

**Questions:**
- How does the model perform on real-world protein structures compared to synthetic data?
- What is the robustness of the method when applied to a diverse set of structures not seen during training?

**Soundness:**
3 good

**Presentation:**
4 excellent

The paper is well-structured, clearly written, and includes all necessary details for reproducing the research. The methodology is presented in a logical flow, and the results are presented with clear figures and tables for easy understanding.

**Contribution:**
4 excellent

The paper contributes a novel and efficient method for inverse protein folding that addresses the limitations of existing approaches by generating diverse sequences for a given structure. The method has the potential to expand the sequence design space and enable the discovery of novel protein structures with desirable properties.

**Rating:**
6 marginally above the acceptance threshold

While the paper presents a significant contribution to the field of inverse protein folding, it could benefit from more comprehensive evaluation on real-world data and a discussion on its limitations and potential areas for improvement. The paper is well-presented and technically sound, making it a strong candidate for acceptance.

**Paper Decision:**
- Decision: Accept
- Reasons: The paper presents a novel and efficient method for inverse protein folding with a strong theoretical foundation and comprehensive experimental validation. The contributions are significant to the field, despite the need for further evaluation on real-world data.