PyTorch Day China
""

Submission Reviewer Guidelines

Thank you in advance for your efforts as a PyTorch Day China 2026 program committee member.

Below you will find guides to Sessionize’s evaluation modes in addition to best practices to use when reviewing your set of proposals. Please bookmark this page for easy reference. If you have any questions, please email the PyTorch Events Content Team.

  • Session Schedule Announced: Wednesday, November 18
  • Event Date: Saturday, 2026 December 12

Below are the suggested topics PyTorch Day China 2026. Reviewers should keep these priorities in mind when evaluating submissions to ensure the program reflects the strategic direction of the event.

  • Model Training & Post-Training: Model training, fine-tuning, and alignment using PyTorch and related open-source stacks. Topics include distributed training, supervised fine-tuning (SFT), reinforcement learning, preference optimization, and improvements in training efficiency, stability, and resource utilization.
  • Efficient Inference & Deployment: Deploying models as reliable, continuously running services. Topics include LLM and multimodal inference, quantization, speculative decoding, KV cache optimization, cluster scheduling, and trade-offs among latency, throughput, and cost across cloud, on-premises, and local deployments.
  • PyTorch Core & Hardware Acceleration: Advances in PyTorch compilation, execution, and hardware enablement. Topics include torch.compile, kernel and operator optimization, performance profiling, and adaptation, compatibility, and reproducible benchmarking across GPUs and AI accelerators.
  • Physical AI: Applying PyTorch models to robotics and physical devices. Topics include Vision-Language-Action (VLA) models, perception and control, and simulation-based training.
  • PyTorch Ecosystem & Applications: Tools, domain applications, and community contributions built on PyTorch and related open-source projects. Topics include domain-specific libraries, development and evaluation tools, and open-source collaboration.

Program committee members are diverse in backgrounds and are invited to the Program Committee. To ensure diversity in the committee and mitigate bias, the PC members are carefully selected by taking into account various factors, like their specific interest and domain expertise. Members are also selected from a diverse range of backgrounds, considering gender diversity, geodiversity, and company diversity.

  • The evaluation plan will be activated in Sessionize Wednesday, November 11.
  • Program Committee members will use Star Rating Evaluation Mode to complete reviews by the deadline.
  • Upon completion, your scores will be combined with your fellow program committee members.

Program Committee Benefits

To say thank you for your hard work, we are offering reviewers:

  • A complimentary registration for the event
  • A listing on the Program Committee web page & event welcome slides
  • Each member also receives a Program Committee Member digital badge from Credly.com to display on social channels.

AI Evaluations are not permitted.

Please reach out to creyesjones@lfopensource.com if you have any questions regarding the benefits listed above.

  • Time Commitment: Please plan on committing anywhere from 3-6 hours total to review all submissions, and make final selections depending on the number of submissions received.
  • Process Integrity: It is very important to protect the integrity of the review process, and to avoid undue bias, by keeping the submissions and your comments on them confidential. Please review and adhere to our Code of Conduct.
  • Public & Author Interaction: To ensure an unbiased review process, program committee members should not discuss submissions with authors and/or the overall public (i.e., please no tweeting). Of course, please feel free to tweet about accepted sessions that you are excited to attend once the schedule has been published.
  • Conflict of Interest: Reviewers are asked to wear their “PyTorch” hats rather than the company or other affiliation when scoring submissions so that you rate all submissions fairly. If a submission was written by a colleague you work closely with or someone you are seen to be associated with or in competition with, please skip the question. This will not affect the overall rating.
  • Short Topic + Refereed Presentations: A presentation is delivered by a topic expert with a fresh or unique point of view. Some things to keep in mind when reviewing presentation proposals:
    • Is the submission well written?
    • Is the topic relevant, original and are they considered to be subject matter experts?
    • Are they talking about a specific product from their company? If so, is it engaging in a way that is not advertorial? Keep in mind that sessions that come across as a pitch or infomercial for their company are most often rated very poorly among the audience.
    • Who is their target audience? Does the abstract and description match up with the expertise required?
Screenshot of Star Rating Mode
The progress bar updates in real-time, letting you know how far in the evaluation process you’ve come

The Star rating evaluation mode is highly user-friendly, requiring only an examination of the session information and a rating between one to five stars. Additionally, half-star ratings (0.5, 1.5, 2.5, 3.5, 4.5) are also available for use. Upon completion, simply click the Save and Continue button to confirm your rating and proceed to the next session.

Screenshot of Star Rating Overview

PyTorch Day China utilizes a Stars rating evaluation plan that involves multiple criteria. As an evaluator, you will be required to assess sessions based on the following four criteria as opposed to assigning a singular overall rating:

  • Content: The relevance and coherence of the session’s content, the quality of the proposal, and the likelihood of effective delivery by the speaker.
  • Originality: The degree to which the session presents new and innovative ideas or approaches, as well as the originality of its delivery.
  • Relevance: The extent to which the session’s content provides new and exciting insights or information that is relevant to the conference.
  • Speaker(s): The suitability of the proposed speaker(s) based on their expertise and alignment with the subject matter.

In addition, we ask that if you are able, please provide feedback in the form of a comment for each session. It is important to ensure that feedback is constructive, especially for rejected proposals, as submitting authors may range from a VP at a large company to a university student. Constructive feedback may include highlighting the positive aspects of a proposal, offering helpful suggestions, and providing factual feedback.

It is crucial to avoid direct attacks and instead focus on objective feedback that can help improve the proposal. Moreover, we strongly advise against using vague comments like “Scoring was tough, I had to cut this” or “LGTM.” Instead, it is essential to provide thoughtful and insightful comments that will assist the co-chairs in making their final selections.

Once you have finished rating a session based on the criteria and provided constructive feedback, click on Save and continue to proceed to the next session. Please note that if your comments are deemed unconstructive, you may not be invited to serve as a program committee member in the future.

Screenshot of how to save and continue later

If you have a hard time coming up with a decision about a certain session, you have the option to skip it and come back to it later. Simply click on the arrow flanking the Save and continue button to expand the appropriate menu and select the Skip and ask later option. This can be particularly useful if you have only just started with the evaluation process and would like to get a better sense of the overall quality of the submitted sessions.

In case a certain session covers a subject you’re completely unfamiliar with or poses a conflict of interest, click on the Ignore this session button. The evaluation system won’t ask you about that session anymore.

Track your progress

Screen shot of the evaluation bar

During the evaluation process, a progress bar will be displayed at the top of the page, providing an indication of your progress. If, at any point, you need to pause the evaluation process, click on the Stop and continue later button located above the progress bar. Upon returning, you can resume the evaluation from where you left off.

On the right is a box with several useful tabs. The default tab is Recent. You can use it to keep track of your past session evaluations, but it has an additional purpose: you can click on any of the sessions to reopen them and potentially change your evaluation.

Here’s a complete overview of the tabs found in the aforementioned box:

  • Recent – a list of sessions you recently evaluated
  • Speaker – see other sessions submitted by the same speaker (assuming they exist)
  • Similar – browse similar sessions
  • Search – look through all nominated sessions
Screen grab of the four tab options.
Keep track of your past evaluations and reevaluate the ones you aren’t happy with

Complete the evaluation and view your stats

Screenshot of star rating dashboard stats
Change your evaluation even after you’ve completed the initial process

Once you’re done with the evaluation, you’ll automatically be redirected to the Evaluation page. By opening the evaluation plan you’ve just completed, you can view your statistics, as well as potentially change your mind on any of the sessions by clicking on the corresponding edit button.

Your scores will be combined with other members in your review track.

OVERVIEW

Volunteers who help with the planning, organization, or production of a Linux Foundation event are often seen as representatives of the LF community and their actions can meaningfully impact participants’ experience and perception of the event.  Therefore, and in the interest of fostering an open, positive, and welcoming environment for our community, it’s important that all event volunteers hold themselves to a high standard of professional conduct as described below.

These guidelines apply to a volunteer’s conduct and statements that relate to or could impact any Linux Foundation event that the volunteer helps plan, organize, select speakers for, or otherwise serve as a volunteer for.  These guidelines apply to relevant conduct occurring before, during, and after the event, both within community spaces and outside such spaces (including statements from personal social media accounts), and to both virtual and physical events. In addition to these guidelines, event volunteers must comply with LF Open Source Event Code of Conduct and the PyTorch Code of Conduct.

Be professional and courteous

Event volunteers will:

  • Conduct themselves in a professional manner suitable for a workplace environment;
  • Treat other event participants (including speakers, sponsors, exhibitors, attendees, volunteers, and staff) with courtesy and kindness; and
  • In their event-related communications, express their opinions in a courteous and respectful manner, even when disagreeing with others, and refrain from using obscenities, insults, rude or derisive language, excessive profanity, or other unprofessional language, images, or content.

Express feedback constructively, not destructively

The manner in which event volunteers communicate can have a large impact personally and professionally on others in the community. Event volunteers should strive to provide feedback or criticism relating to the event or any person or organization’s participation in the event in a constructive manner that supports others in learning, growing, and improving (e.g., offering suggestions for improvement).  Event volunteers should avoid providing feedback in a destructive or demeaning manner (e.g., insulting or publicly shaming someone for their mistakes).

Be considerate when choosing communication channels

Event volunteers should be considerate in choosing channels for communicating feedback.  Positive or neutral feedback may be communicated in any channel or medium.  In contrast, criticism about any individual event participant, staff member, or volunteer should be communicated in one or more private channels (rather than publicly) to avoid causing unnecessary embarrassment.  Criticism about an event that is not about specific individuals may be expressed privately or publicly, so long as it is expressed in a respectful, considerate, and professional manner.

Treat sensitive data confidentially and with respect

Event volunteers may have access to details about proposed or accepted speakers and the contents of their talks. They are required to adhere to The Linux Foundation’s guidelines regarding use of this information and may only use it for the purpose of choosing talks for an event. They are prohibited from using this data for any other purpose, including but not limited to the following:

  1. Using the information for unrelated business purposes
  2. Contacting speakers for any purpose other than evaluating their submission for this event
  3. Asking a submitter to speak at another event or recruiting them for another role
  4. Sharing the information with anyone outside of the program committee or sharing the acceptance of the talk prior to the schedule and abstract being announced

Changes to These Guidelines and Consequences for Noncompliance 

The event organizers may update these guidelines from time to time, and will notify volunteers by email.  However, any changes to these guidelines will not apply retroactively.  If the Linux Foundation Events team determines that a volunteer has violated these guidelines or LF Open Source Event Code of Conduct, it may result in the volunteer’s immediate suspension or removal from any event-related volunteer positions they hold, including participation in event-related committees. If these guidelines are updated and a volunteer does not wish to agree, their participation in the event-related volunteer position will cease until such time as they do agree.

If you require any assistance reviewing proposals or have questions about the review process or any of the best practices we have suggested, please contact us for assistance.