This was extracted (@ 2025-07-04 18:10) from a list of minutes
which have been approved by the Board.
Please Note
The Board typically approves the minutes of the previous meeting at the
beginning of every Board meeting; therefore, the list below does not
normally contain details from the minutes of the most recent Board meeting.
ASF Members may have access to a
private draft
WARNING: these pages may omit some original contents of the minutes.
Report was filed, but display is awaiting the approval of the Board minutes.
## Description: The mission of Apache TVM is the creation and maintenance of software related to compilation of machine learning models to run on a wide range of hardware platforms ## Project Status: Current project status: ongoung Issues for the board: none ## Membership Data: Apache TVM was founded 2020-11-17 (4 years ago) There are currently 78 committers and 29 PMC members in this project. The Committer-to-PMC ratio is roughly 5:2. Community changes, past quarter: - Mengshiun Yu was added as committer on 2025-02-13 ## Project Activity: Recent releases - 0.19.0 was released on 2025-01-28 - 0.18.0 was released on 2024-10-27 The project started to bring more focus on the latest python-first compiler strategies and LLM workloads. Some recent improvements include: - Support for new data types, FP8, BF16 and F4 - Better support for Torch importers - KVCache runtime for latest reasoning models - Support for MLA kernels in deepseek models. - OpenCLML offload for relax. The project also started to phase out legacy components to clean up technical debts and focus on coherent developments for python-first ML compiler. ## Community Health: Generally, we see more interest in new models and generative AI workloads. The community is focusing on new LLM-related directions to meet those interest shifts. Overall, the project merges about 70 PRs per month. We are also spend good amount of energy on architecting to ensure we can have a solid foundation as we bring in support for new workloads.
## Description: The mission of Apache TVM is the creation and maintenance of software related to compilation of machine learning models to run on a wide range of hardware platforms ## Project Status: Current project status: ongoing Issues for the board: none ## Membership Data: Apache TVM was founded 2020-11-17 (4 years ago) There are currently 77 committers and 29 PMC members in this project. The Committer-to-PMC ratio is roughly 5:2. Community changes, past quarter: - Star Yuan was added to the PMC on 2024-10-27 - Masahiro Hiramori was added as committer on 2024-09-23 We also continue to find new committers. ## Project Activity: Recent releases: - 0.18.0 was released on 2024-11-27. - 0.17.0 was released on 2024-09-10. - 0.16.0 was released on 2024-06-12. The community continues to develop the project. Most of recent developments has been on maintainace fronts. The community also start to phase out legacy components to reduce overall complexity of the codebase. Notable developments include - KVCache attention kernel improvements for various backends - Make ORCJIT LLVM executor the default one - ONNX frontend coverage of the relax flow ## Community Health: The project get about 225 commits in the past quarter. As the field of ML moves fast, the community needs to keep up with the interest shift of the ML community. We still see interesting needs on areas like LLM inference and generative AI models. The overall developments starts to shift to stablizing and maintainace, with also mindful effort on bringing new momentums on the latest generative AI fronts.
## Description: The mission of Apache TVM is the creation and maintenance of software related to compilation of machine learning models to run on a wide range of hardware platforms ## Project Status: Current project status: Ongoing Issues for the board: none ## Membership Data: Apache TVM was founded 2020-11-17 (4 years ago) There are currently 76 committers and 28 PMC members in this project. The Committer-to-PMC ratio is roughly 5:2. There are active discussions on bringing in committers and PMC members. ## Project Activity: Recent releases: - 0.17.0 was released on 2024-09-08. - 0.16.0 was released on 2024-06-12. - 0.15.0 was released on 2024-03-10. The project continues to bring latest set of changes. Major contributions centers around documentations to transition to the new generative AI focused topic. We see specific improvements around - Relax development in general - Dlight support for high performance GPU kernels across backends - Distributed and multiGPU support in large language models ## Community Health: We are entering a period of steady state development after the TVM unity transition. The community starts to move the major documentation towards the new flows, which can help in terms of productivity in supporting new models. Overall, the community gets received from 39 contributors in the past quarter, with 70 commits monthly. Most activities are around stabilizing the new features and existing ones. As the field of machine learning and related user community shifts to generative AI workloads, the community also starts to adapt to the latest set of interests and keep up with the changes in the overall landscape of AI/ML. We need to continue to find approaches to keep up with the latest innovations in the latest AI/ML space.
## Description: The mission of Apache TVM is the creation and maintenance of software related to compilation of machine learning models to run on a wide range of hardware platforms ## Project Status: Current project status: Ongoing Issues for the board: none ## Membership Data: Apache TVM was founded 2020-11-17 (4 years ago) There are currently 76 committers and 28 PMC members in this project. The Committer-to-PMC ratio is roughly 5:2. Community changes, past quarter: - Balint Cristian was added as committer on 2024-05-20 - Park Sunghyun was added as committer on 2024-05-30 - Tong Meng was added as committer on 2024-04-09 ## Project Activity: Recent releases: 0.16.0 was released on 2024-06-12. 0.15.0 was released on 2024-03-10. 0.14.0 was released on 2023-11-03. The project continues to bring in new features; highlights include: - Better multiGPU support - Improvements for LLM applications and KV cache management - Better backend support for WebGPU and other emerging platforms. - Supporting latest H100 GPU and FP8 We start to see a lot of developments now centering around new LLM workloads and applications start to arise for deploying these latest models. ## Community Health: The community get contributions from 64 contributors in past quarter, with 200 commit smonthly. We are entering a period of steady state development after the TVM unity transition. Most activities are around stabilizing the new features and existing ones. We also started to see interest in areas like WebGPU and H100 FP8 support. As the field of machine learning and related user community shifts to generative AI workloads, the community also starts to adapt to the latest set of interests and keep up with the changes in the overall landscape of AI/ML.
## Description: The mission of Apache TVM is the creation and maintenance of software related to compilation of machine learning models to run on a wide range of hardware platforms ## Project Status: Current project status: ongoing Issues for the board: none ## Membership Data: Apache TVM was founded 2020-11-17 (3 years ago) There are currently 73 committers and 28 PMC members in this project. The Committer-to-PMC ratio is roughly 5:2. Community changes, past quarter: - No new PMC members. Last addition was Ruihang Lai on 2023-09-27. - Star Yuan was added as committer on 2024-01-08 ## Project Activity: Recent releases: - 0.14.0 was released on 2023-11-03. - 0.13.0 was released on 2023-08-08. - 0.12.0 was released on 2023-05-13. The project continue to make strides on enabling emerging workloads such as dynamic shape and LLM models. Some of the highlights include: - Better support for remote debugging in hexagon NPU - GPU sampling for LLM usecases - H100 and fp8 initial support ## Community Health: The project continues to receive contributions from the community. As a project in machine learning we also need to adapt and build solutions for the latest waves of genAI models. The community voted and passed on the transition to enable the unity development effort in the main branch [1]. As of now the transition has been completed, including new features in unity while ensuring all previous testcases pass. This milestone will and helps us to collectively evolve the project to empower genAI. [1] https://lists.apache.org/thread/s9cv79wjf3pp3m453f5kpggksp0zb2qo
## Description: The mission of Apache TVM is the creation and maintenance of software related to compilation of machine learning models to run on a wide range of hardware platforms ## Project Status: Current project status: Ongoing Issues for the board: none ## Membership Data: Apache TVM was founded 2020-11-17 (3 years ago) There are currently 72 committers and 28 PMC members in this project. The Committer-to-PMC ratio is roughly 9:4. Community changes, past quarter: - Bohan Hou was added to the PMC on 2023-09-24 - Ruihang Lai was added to the PMC on 2023-09-27 - Qiang Zhang was added as committer on 2023-09-27 ## Project Activity: Recent releases: - 0.14.0 was released on 2023-11-03. - 0.13.0 was released on 2023-08-08. The project continues to push the frontiers of machine learning compilation. Some feature highlights include: - Better dynamic shape support - Enablement of genAI and LLM models. We have observed major participation in the unity branch development due to the overall shift of the ML community toward emerging foundational models. The developers also continue to bring maintaiance to main and keep the branches in sync. ## Community Health: The field of machine learning is evolving very fast this year with the arrivial of genAI and LLMs. They shape sthe overall interest of open source machine learning ecosystem. Being able to support genAI models timely is likely critical for the community given the focus of the broader ML community we sypport. We continue to see contributions to the projects with ~120 commits monthly from 40 authors. We also observe that in the case of contributions, most new contributions now go to the unity branch due to the heavy interest in genAI directions. The community held conversations about core development strategy in the age of genAI and had a broad consensus about bringing unity branch as main while ensuring all the main modules remain supported and synced. Over the past quarter, the community has continued to maintain and sync the unity branch to make such a transition seamless while still giving some time for all community members. We will likely see such a transition in the incoming quarter, given the broad support from the community.
## Description: The mission of Apache TVM is the creation and maintenance of software related to compilation of machine learning models to run on a wide range of hardware platforms ## Project Status: Current project status: Ongoing Issues for the board: there is no specific issue that need board action. We would like to inform board about some of the latest community resolutions (please see more on community healthy section). ## Membership Data: Apache TVM was founded 2020-11-17 (3 years ago) There are currently 71 committers and 26 PMC members in this project. The Committer-to-PMC ratio is roughly 9:4. Community changes, past quarter: - There is on going threads bringing in new PMC members and committers - Yong Wu was added as committer on 2023-06-13 ## Project Activity: Recent releases: 0.13.0 was released on 2023-08-08. 0.12.0 was released on 2023-05-13. The project continues to push the frontiers of machine learning compilation. Some feature highlights include: - Better dynamic shape support in low-level codegen - Various importer improvements - Initial foundations to support H100 GPUs The community also continues to participate in the unity branch development. We have observed growing interest and participation in tvm unity development, especially around enabling emerging foundational models. The majority of the community, including members developing the core modules, also supports the need to put more investment into foundational models. Such sentiment is also shared among volunteer developers who would like enable the latest models. ## Community Health: We get about 140 monthly commits from ~50 authors. We start to see a growing demand from the community to empowering emerging directions like foundational models and bring strategic innovations for the community. In August, the community had a series of discussions on the desire to bring in innovations for emerging directions (foundational models) as well as the overall process for enabling strategic decisions in the community. These discussions are follow-ups to help us address stagnation and continue to empower everyone, see also the previous actions that were brought in the board report [2]. As the community pushes for different goals that help each other, naturally, there are strategy decision points about overall directions and new module adoptions. These decisions are not fine-grained code-level changes but are important for a community to be viable in the long term. The evolution pace of the ML/AI landscape is also very fast; the cost of stagnation is high, and it is critical for the community to be able to make collective strategic decisions together and empower the community. Taking these and the past year's context into consideration, we follow the ASF process that enables the community to have a collective conversation that gains clarity of everyone's view and then comes to a resolution for the process of strategic decisions. The community passed a clear and concise process for strategy decisions[1]. This item does not need board action, as the TVM PMC handled the case through the standard ASF process and empowers the community. We believe it is important to keep the board informed, so we are bringing it up in the report. -- [1] https://lists.apache.org/thread/ywwsdfxqv0wvz3wz19w6nqgqx23kswql [2] https://github.com/apache/tvm-rfcs/pull/89#issuecomment-1404170556
## Description: The mission of Apache TVM is the creation and maintenance of software related to compilation of machine learning models to run on a wide range of hardware platforms ## Project Status: Current project status: Ongoing with active contributions. Issues for the board: nothing atm ## Membership Data: Apache TVM was founded 2020-11-17 (3 years ago) There are currently 70 committers and 26 PMC members in this project. The Committer-to-PMC ratio is roughly 9:4. Community changes, past quarter: - No new PMC members. Last addition was xinqi on 2023-01-31. - Zihao Ye was added as committer on 2023-04-15 - Jiang Jiajun was added as committer on 2023-06-03 - Nicola Lancellotti was added as committer on 2023-05-02 - Anirudh Sundar Subramaniam was added as committer on 2023-05-12 There are also on going discussions on new members. ## Project Activity: Recent releases: - 0.12.0 was released on 2023-05-13. - 0.11.0 was released on 2023-03-09. The project continues to make strides on various fronts. There are several improvements in frontends and TensorIR. The unity branch development also starts gaining community traction, especially in enabling new emerging areas, such as large-language and stable diffusion models that show value to the community. The community also continues to publish tutorials to help community members to get updates on the latest unity development. Specifically, tutorials on BYOC (bring your own codegen) as well as vertical-focused applications such as how to run language models. ## Community Health: Overall we get ~150 commits from ~50 authors monthly. We also start to see a a growing interest in TVM unity branch to enable stable diffusion and language models. Recap on TVMCon: the talks at TVMCon are now online at https://www.tvmcon.org/ This year we had speakers from diverse backgrounds in industry and academia talking about their use in TVM and ML compilation in general. One highlight was the ability to leverage TVM and bring models like stable diffusion onto server, web browsers and eventually mobile devices. Bringing in new members: the PMC is mindful in welcoming new contributors and members and bring a healthy community. One recent priority of some of us is to bring volunteer contributors from different backgrounds. These contributors usually have more spread-out contributions but are super valuable to the community. We identified several such members in the past quarters and would strive to continue to do so. The unity branch development also serves as a way to encourage new members of the community to participate in the emerging areas they are interested in, hopefully bringing a broader set of members to the community. Supporting the community in the age of generative AI: with the arrival of generative AI, such as stable diffusion and language models, the community's interest also start to arise in these new domains. Supporting these new models require different technical considerations from the traditional models we support. Unity branch development brings these new capabilities to the community. We have seen growing community interest in related areas and unity development, with active recent community meetups and contributions to support LLMs.
## Description: The mission of Apache TVM is the creation and maintenance of software related to compilation of machine learning models to run on a wide range of hardware platforms ## Issues: There are no specific issues that would need actions from the board. We would like to disclose the recent resolutions by PMC this quarter to unblock some of the development in the community. See more in project activity section. ## Membership Data: Apache TVM was founded 2020-11-17 (2 years ago) There are currently 66 committers and 26 PMC members in this project. The Committer-to-PMC ratio is roughly 9:4. Community changes, past quarter: - xinqi was added to the PMC on 2023-01-31 - Hongyi Jin was added as committer on 2023-01-17 - Yaxing Cai was added as committer on 2023-01-27 ## Project Activity: Recent releases: - 0.10.0 was released on 2022-10-17. - 0.9.0 was released on 2022-07-19. The project continues to bring new contributions. The latest set of improvement include highlighted areas: - Better TVMScript printing/parsing support. - Micro tvm and embedded devices support. - Improved support for nvidia TensorCore. - Better arithmetic support. On January, the TVM PMC passed a procedural resolution to enable branch development following lessons from other ASF projects and helping unblock some of the stagnation. Last Aug, a new module was proposed to add to the project, and the proposal was welcomed by many community members from more than eight organizations. The proposal brings a new module that potentially overlaps functionalities with existing modules. Majority of members participating the discussions were not concerned about the overlap and welcomed the proposal. However, there were disagreements from some community members on procedural operations the proposal. The discussions mainly boiled down to procedural questions about whether a migration plan and commitment to migration are necessary for the new module proposal at the point of the proposal or if the new module can co-exist with existing ones and empower different community members to develop them. The discussions has stagnated around winter break. After the winter break, the PMC took steps to resolve the stagnation and empower the community given a lot of asks from community in the thread about the empowerment. The first step compromise was to enable such development first in a branch. This would gives opportunity to empower members who would like to participate in the development and provide examples to those who wants to see more feasibility analysis of migrating some modules. We do not think this is something that would need board action as the PMC operates normally to bring a resolution and provide oversight support as being empowered by the ASF mechanism. We would like to disclose this to the board in the report for visibility. Please see the following threads for full context: the original public thread and resolution[1](also mirrored in dev). [1] https://github.com/apache/tvm-rfcs/pull/89#issuecomment-1404170556 The new branch development was established, with many community members support and contributions to the unity branch developments. The members are also actively publishing discussion, tutorials, and examples so there will be more awareness and hopefully also bring more lights on questions other community members might have. ## Community Health: The community continues to receive new contributions. Every month we received contributions from ~70 authors with ~ 130 commits to main branch. With the stagnation resolved, there are also a good amount of influx and interest of new contributions since the unity branch get established both in terms of code, demos, and examples. The community will hold the online community conference in the incoming week. The conference will talks from both the TVM community members and speakers from the broader machine learning compilation community. The community members also self-organize local meetups. Local meetup in shanghai was organized last week with over 100 people attended and engaged in lively discussions and communications (along with public thread summary), covering topics around TVM for NPU, SYCL, and possible opportunities in TVM unity.
## Description: The mission of Apache TVM is the creation and maintenance of software related to compilation of machine learning models to run on a wide range of hardware platforms ## Issues: There are no issues requiring board attention. ## Membership Data: Apache TVM was founded 2020-11-17 (2 years ago) There are currently 64 committers and 25 PMC members in this project. The Committer-to-PMC ratio is roughly 2:1. Community changes, past quarter: - No new PMC members. Last addition was Josh Fromm on 2022-09-12. - Ashutosh Parkhi was added as committer on 2022-11-02 - Egor Churaev was added as committer on 2022-11-16 - Gavin Uberti was added as committer on 2022-12-02 ## Project Activity: Recent releases: - 0.10.0 was released on 2022-10-17. - 0.9.0 was released on 2022-07-19. Recent improvement highlights include: - Better support for meta-schedule. - Better support for hexagon devices. - New TVMScript parser - MicroTVM improvements ## Community Health: The projects merges 175 monthly. The community members continue to discuss modules proposals on various front. We are also in the process of organizing another community meeting in the incoming quarter. The initial call for proposal received 33 proposals and we plan to do another round of solicitations in the incoming new year.
## Description: The mission of Apache TVM is the creation and maintenance of software related to compilation of machine learning models to run on a wide range of hardware platforms ## Issues: There are no issues requiring board attention ## Membership Data: Apache TVM was founded 2020-11-17 (2 years ago) There are currently 61 committers and 24 PMC members in this project. The Committer-to-PMC ratio is roughly 8:3. Community changes, past quarter: - Wuwei Lin was added to the PMC on 2022-08-30 - David Riazati was added as committer on 2022-06-14 - Elen Kalda was added as committer on 2022-09-06 ## Project Activity: 0.9.0 was released on 2022-07-19. The project continues to make strides towards bringing new features to the community. Some of the highlights include: - New TVMScript parser support to enable meta-programming. - Improved support to qualcomm DSPs - Better PyTorch and code-generation interpolation - MicroNPU support The community also continues to explore new module support to enable dynamic shape and other needs. ## Community Health: The project maintains a monthly contribution of ~230 commits from 70 contributors. We continue to recognizes new members and welcome them to the community. We are also working together towards' TVMCon that brings together developers, users and researchers in the field to talk about overall machine learning compilation.
## Description: The mission of Apache TVM is the creation and maintenance of software related to compilation of machine learning models to run on a wide range of hardware platforms ## Issues: There are no issues requiring board attention. ## Membership Data: Apache TVM was founded 2020-11-17 (2 years ago) There are currently 58 committers and 23 PMC members in this project. The Committer-to-PMC ratio is roughly 8:3. Community changes, past quarter: - Krzysztof Parzyszek was added to the PMC on 2022-03-18 - Siyuan Feng was added to the PMC on 2022-03-20 - Gustavo Romero was added as committer on 2022-04-08 - Luke Hutton was added as committer on 2022-04-19 - Mehrdad Hessar was added as committer on 2022-04-05 - Ruihang Lai was added as committer on 2022-04-14 - xinqi was added as committer on 2022-04-14 - Xiyou Zhou was added as committer on 2022-04-23 ## Project Activity: Recent releases: 0.8.0 was released on 2021-11-25. The project community actively works towards improving the project in various areas, highlights include: - Improved support for qualcomm/ARM devices - Foundational improvements in arithmetic modules - Better usability and productivity via TVMScript. ## Community Health: The community continues to bring in new contributors. We have brought 8 new PMC members and committers in the past quarter. ~250 PRs are merged every month. There are also more organized planning through the RFC mechanism. We see a good amount of RFCs to support new features, frontends and hardware backends.
## Description: The mission of Apache TVM is the creation and maintenance of software related to compilation of machine learning models to run on a wide range of hardware platforms ## Issues: There are no issues requiring board attention. ## Membership Data: Apache TVM was founded 2020-11-17 (a year ago) There are currently 52 committers and 21 PMC members in this project. The Committer-to-PMC ratio is roughly 7:3. Community changes, past quarter: - Leandro Nunes was added to the PMC on 2022-03-01 - Andrew Luo was added as committer on 2022-01-11 - Bohan Hou was added as committer on 2022-01-06 - Chris Sullivan was added as committer on 2022-02-22 - Lily Orth-Smith was added as committer on 2022-02-28 - Eric Lunderberg was added as committer on 2021-12-17 ## Project Activity: Recent releases: 0.8.0 was released on 2021-11-25. The community continues to push TVM Unity -- the technical vision that we collectively set during TVMCon. Specifically with an emphasize on enabling collaboration and cross layer optimizations. Some of the recent highlights include: - Metaschedule, the latest automation mechanism is now fully upstreamed. - AOT compilation support for embedded devices - Hexagon DSP support - Overall performance improvements. ## Community Health: The community successfully held a virtual developer conference TVMCon in the past Dec. TVMCon attracted more than 1000 registered developers from all over the world. On average 200 PRs are merge every month from around 80 authors. We also start to see new features being proposed through the formal RFC process. We continue to welcome new committers and PMC members to the community.
## Description: The mission of Apache TVM is the creation and maintenance of software related to compilation of machine learning models to run on a wide range of hardware platforms ## Issues: There are no issues requiring board attention. ## Membership Data: Apache TVM was founded 2020-11-17 (a year ago) There are currently 47 committers and 20 PMC members in this project. The Committer-to-PMC ratio is roughly 2:1. Community changes, past quarter: - Andrew Reusch was added to the PMC on 2021-11-21 - Chris Sidebottom was added as committer on 2021-11-23 There are also ongoing nominations. ## Project Activity: Last release: 0.8.0 was released on 2021-11-25. The community works very hard to complete the 0.8.0 release. There are a lot of active development activities over the past quarter. Including, but not limited to: - Better TE, TIR, TVMScript support - More frontends (Paddle paddle) - Stablizing AutoScheduler - Stablizing Target system - TensorIR and meta-schedule support - TVMC and microTVM improvements - Improved Vulkan backend ## Community Health: The community has been receiving healthy contributions. Monthly about 70 authors contributing 150 commits. The community RFC process also starts to pick momentum, receiving RFCs from various contributors. We will hold the annual community conference TVMCon next week. This year we have one day tutorial and two days of talks with broad participation from both industry and academia.
## Description: The mission of Apache TVM is the creation and maintenance of software related to compilation of machine learning models to run on a wide range of hardware platforms ## Issues: There are no issues requiring board attention. ## Membership Data: Apache TVM was founded 2020-11-17 (10 months ago) There are currently 46 committers and 19 PMC members in this project. The Committer-to-PMC ratio is roughly 2:1. Community changes, past quarter: - Cody Yu was added to the PMC on 2021-07-13 - Junru Shao was added to the PMC on 2021-07-08 - Giuseppe Rossini was added as committer on 2021-09-04 - Manupa Karunaratne was added as committer on 2021-08-26 - Siyuan Feng was added as committer on 2021-09-04 ## Project Activity: Last release: 2020-10-09 The community starts the 0.8 release planning process. This release is going to contain a few major reworks of the TVM's internal architecture and bring next set of improvements. The project have been quite healthy and maintains the rate of 210 commits per month. Recent highlights include: - TensorIR scheduling system close to land in a month. - New frontend support: Oneflow, Paddle - Project API for micro tvm. - Automatic mixed precision support to enable customized floating points such as fp16. ## Community Health: We continue to bring new people from different organizations as committers and PMC members. The new RFC process(apache/tvm-rfcs) is now in-place. Overall the community reacted positively to the new RFC process as it allows more people to participate in the design proposal phase of the architecture. We also start to see a lot of collaborations around RFC discussions. The community also start to take a stab in improving code review process by update the review guidelines to contain more recommendations around code convention, documentation and consensus building. The proposed guideline improvements is voted and set to be incorporated into the contributor guide.
## Description: The mission of Apache TVM is the creation and maintenance of software related to compilation of machine learning models to run on a wide range of hardware platforms ## Issues: The project does not have issues that need board attention. ## Membership Data: Apache TVM was founded 2020-11-17 (7 months ago) There are currently 43 committers and 17 PMC members in this project. The Committer-to-PMC ratio is roughly 2:1. Community changes, past quarter: - Yao Wang was added to the PMC on 2021-04-05 - Andrew Reusch was added as committer on 2021-03-19 - Leandro Nunes was added as committer on 2021-05-07 - Steven Lyubomirsky was added as committer on 2021-05-04 - Trevor Morris was added as committer on 2021-05-21 ## Project Activity: Last release: 2020-10-09 The community is actively working on the new iteration towards the next release. The project have been quite healthy and maintains the rate of 150 commits per month. The improvement highlights include: - A major rework of the new TensorIR scheduling system. - uTVM to enable embedded device support - Better vulkan support to enable global coverage. - Better ONNX frontend support ## Community Health: The community is healthy. We continue to bring new people from different organizations as committers and PMC members. In our particular case, the PMC members follows only nominates people who are outside their own organization, encourage identification of candidates who are actively participating per apache way. The community members also continues to participate in design and general discussions Additionally, the community has voted to use a new RFC process that sends formal RFCs to a separate repo apache/tvm-rfcs after broad discussion among the community members.
## Description: The mission of Apache TVM is the creation and maintenance of software related to compilation of machine learning models to run on a wide range of hardware platforms ## Issues: No issue that needs board attention ## Membership Data: Apache TVM was founded 2020-11-17 (4 months ago) There are currently 39 committers and 16 PMC members in this project. The Committer-to-PMC ratio is roughly 5:2. Community changes, past quarter: - Leyuan Wang was added to the PMC on 2021-01-16 - Chenfan was added as committer on 2020-12-18 - Josh Fromm was added as committer on 2021-01-19 ## Project Activity: The community merged 130 pull requests from 69 authors in the past month. The improvements highlights includes: - Better vulkan and SPIRV support - Support of irregular control flow such as while loop - Initial support of TensorIR for tensor-core libraries - Various improvements to the Torch and Tensorflow frontends. ## Community Health: The developer community is healthy, with broad participations from various organizations. The community discussed about start experimenting a new RFC mechanism that opens pull request to a separate github repo (apache/tvm-rfcs, whose pull requests will be mirrored to dev@). This would allow reviews and tracking of the RFC more easily. Additionally, there are quite a few engaged discussion about the Device API, micro controller support, and Torchscript integeration. The community members also give talks in major workshops. The PMC is actively discussing new candidates for PMC and committership.
## Description: The mission of Apache TVM is the creation and maintenance of software related to compilation of machine learning models to run on a wide range of hardware platforms ## Issues: No issue that need board attention ## Membership Data: Apache TVM was founded 2020-11-17 (3 months ago) There are currently 39 committers and 16 PMC members in this project. The Committer-to-PMC ratio is roughly 5:2. Community changes, past quarter: - Leyuan Wang was added to the PMC on 2021-01-16 - Chenfan was added as committer on 2020-12-18 - Josh Fromm was added as committer on 2021-01-19 ## Project Activity: The project has been quite healthy in the past month. On main, 373 files have changed and there have been 14,511 additions and 4,001 deletions. The improvements come in areas include(but not limited to): - Better micro-controller support. - More documentation improvements. - Automatic scheduler enhancement to support more workloads. - Better test coverage and ## Community Health: 177 pull requests were bought up, and 135 of them were merged. Those contributions come from 55 authors. The community members are also actively engaged in RFC discussions, formalizing proposals for features like micro-controller support, compiler updates and accelerator support under the Apache way.
## Description: Apache TVM’s extensible full-stack framework enables deep learning applications to efficiently deploy across an array of hardware modules, platforms, and systems, including mobile phones, wearables, specialized chips, and embedded devices. ## Issues: There are no issues requiring board attention. ## Membership Data: Apache TVM was founded 2020-11-17 (2 months ago) There are currently 38 committers and 15 PMC members in this project. The Committer-to-PMC ratio is roughly 5:2. Community changes, past month: - The community is discussing and voting new PMC members. - Chenfan was added as committer on 2020-12-18 ## Project Activity: Software development activity: - We land the first complete version of the automatic scheduling code generation. - The community has been working on improving the robustness of the internals (better logging and error messaging). - Improvements to uTVM (micro-controller compilation support). - Last project release was v0.7.0 (2020-10-02), the community is working hard to push for the v0.8 release cycle. Meetups and Conferences: - Annual developer conference(see the community health) ## Community Health: Overall community health is good. The amount of PRs drop to 100 (was 150) monthly. But this is expected due to the holiday season. Followed by a successful announcement of the TLP. The community successfully held a three day virtual developer conference (https://tvmconf.org/) on Dec 2-4. Nearly 1000 people registered. The conference attracted talks from major industry users, such as AMD, ARM, AWS, Qualcomm, OctoML, Xilinx, Alibaba, Huawei, and academic contributors from UW, UC Berkeley, Cornell, UCLA, Beihang University. During the conference, the community celebrated the apache way and talked about their applications and contributions to the Apache TVM.
## Description: Apache TVM’s extensible full-stack framework enables deep learning applications to efficiently deploy across an array of hardware modules, platforms, and systems, including mobile phones, wearables, specialized chips, and embedded devices. ## Project Status: - Apache TVM graduated as a TLP last month - Community contribution are healthy, ~150 PRs merged last month - The community is working on new features on automatic scheduling, stabilizing the API, rust integration, micro-controllers, tensorflow and pytorch support ## Community - The community welcomed one new committer last month. There are also ongoing discussions about new committer. - The community is hosting an developer conference on Dec 2-4 https://tvmconf.org/ - Press release about TLP out, thanks to Sally! ## Releases: - Apache TVM 0.7.0 was released Oct 9 2020
WHEREAS, the Board of Directors deems it to be in the best interests of the Foundation and consistent with the Foundation's purpose to establish a Project Management Committee charged with the creation and maintenance of open-source software, for distribution at no charge to the public, related to compilation of machine learning models to run on a wide range of hardware platforms. NOW, THEREFORE, BE IT RESOLVED, that a Project Management Committee (PMC), to be known as the "Apache TVM Project", be and hereby is established pursuant to Bylaws of the Foundation; and be it further RESOLVED, that the Apache TVM Project be and hereby is responsible for the creation and maintenance of software related to compilation of machine learning models to run on a wide range of hardware platforms; and be it further RESOLVED, that the office of "Vice President, Apache TVM" be and hereby is created, the person holding such office to serve at the direction of the Board of Directors as the chair of the Apache TVM Project, and to have primary responsibility for management of the projects within the scope of responsibility of the Apache TVM Project; and be it further RESOLVED, that the persons listed immediately below be and hereby are appointed to serve as the initial members of the Apache TVM Project: * Tianqi Chen <tqchen@apache.org> * Timothy Chen <tnachen@apache.org> * Zhi Chen <zhic@apache.org> * Byung-Gon Chun <bgchun@apache.org> * Ziheng Jiang <ziheng@apache.org> * Furkan Kamaci <kamaci@apache.org> * YiZhi Liu <liuyizhi@apache.org> * Masahiro Masuda <masahi@apache.org> * Thierry Moreau <moreau@apache.org> * Jared Roesch <jroesch@apache.org> * Henry Saputra <hsaputra@apache.org> * Haichen Shen <haichen@apache.org> * Markus Weimer <weimer@apache.org> * Eddie Yan <eqy@apache.org> * Lianmin Zheng <lmzheng@apache.org> NOW, THEREFORE, BE IT FURTHER RESOLVED, that Tianqi Chen be appointed to the office of Vice President, Apache TVM, to serve in accordance with and subject to the direction of the Board of Directors and the Bylaws of the Foundation until death, resignation, retirement, removal or disqualification, or until a successor is appointed; and be it further RESOLVED, that the Apache TVM Project be and hereby is tasked with the migration and rationalization of the Apache Incubator TVM podling; and be it further RESOLVED, that all responsibilities pertaining to the Apache Incubator TVM podling encumbered upon the Apache Incubator PMC are hereafter Discharged. Special Order 7D, Establish the Apache TVM Project, was approved by Unanimous Vote of the directors present.
TVM is a full stack open deep learning compiler stack for CPUs, GPUs, and
specialized accelerators. It aims to close the gap between the
productivity- focused deep learning frameworks, and the performance- or
efficiency- oriented hardware backends.
TVM has been incubating since 2019-03-06.
### Three most important unfinished issues to address before graduating:
- Keep growing the community
### Are there any issues that the IPMC or ASF Board need to be aware of?
no
### How has the community developed since the last report?
TVM community has welcomed three new committers/PPMC members since last
report. There are also on-going new committer nominations. The community
is active and vibrate, with wide collaborations from many contributors.
The total number of contributors has grown to 440.
The community actively works to resolve the items under the guide of the
Apache maturity model.
### How has the project developed since the last report?
A lot of improvements have been made. Including automatic scheduling,
better documentations, command line support
See also our monthly reports for detailed improvements
- July https://discuss.tvm.apache.org/t/tvm-monthly-jul-2020/7570
- Aug https://discuss.tvm.apache.org/t/tvm-monthly-aug-2020/7791
### How would you assess the podling's maturity?
- [ ] Initial setup
- [ ] Working towards first release
- [X] Community building
- [X] Nearing graduation
- [ ] Other:
### Date of last release:
2020-07-10
The community is voting on a new release now
### When were the last committers or PPMC members elected?
2020-09-28
### Have your mentors been helpful and responsive?
Our mentors are super helpful.
### Is the PPMC managing the podling's brand / trademarks?
yes
### Signed-off-by:
- [X] (tvm) Byung-Gon Chun
Comments:
- [ ] (tvm) Sebastian Schelter
Comments:
- [X] (tvm) Henry Saputra
Comments: Peak for graduation.
- [ ] (tvm) Timothy Chen
Comments:
- [X] (tvm) Furkan Kamaci
Comments:
- [X] (tvm) Tianqi Chen
Comments:
- [X] (tvm) Markus Weimer
Comments: The project is ready for graduation.
### IPMC/Shepherd notes:
TVM is a full stack open deep learning compiler stack for CPUs, GPUs, and
specialized accelerators. It aims to close the gap between the productivity-
focused deep learning frameworks, and the performance- or efficiency-
oriented hardware backends.
TVM has been incubating since 2019-03-06.
### Three most important unfinished issues to address before graduating:
1. Keep growing the community
### Are there any issues that the IPMC or ASF Board need to be aware of?
no
### How has the community developed since the last report?
TVM community has welcomed four new committers/PPMC members since last
report. There are also on-going new committer nominations.
The community is active and vibrate, with wide collaborations from many
contributors. The total number of contributors has grown to 387.
The community actively works to resolve the items under the guide of the
Apache maturity model
https://docs.google.com/document/d/18nyAH-fcptVezAxPQe6H3FeTKPRkujOp1tc1YRSP
Lok/edit?usp=sharing
### How has the project developed since the last report?
A lot of improvements have been made. Including wasm/webgpu backend,
performance improvement,
operator/backend coverage, codebase refactor
See also our monthly reports for detailed improvements
- Mar https://discuss.tvm.ai/t/tvm-monthly-march-2020/6199
- Apr https://discuss.tvm.ai/t/tvm-monthly-april-2020/6570
- May https://discuss.tvm.ai/t/tvm-monthly-may-2020/6992
### How would you assess the podling's maturity?
- [ ] Initial setup
- [ ] Working towards first release
- [X] Community building
- [X] Nearing graduation
- [ ] Other:
### Date of last release:
2019-12-1
The community is voting on a new release now
### When were the last committers or PPMC members elected?
No answer.
### Have your mentors been helpful and responsive?
Our mentors are super helpful.
### Is the PPMC managing the podling's brand / trademarks?
yes
### Signed-off-by:
- [X] (tvm) Byung-Gon Chun
Comments:
- [ ] (tvm) Sebastian Schelter
Comments:
- [X] (tvm) Henry Saputra
Comments: Community is growing and healthy
- [ ] (tvm) Timothy Chen
Comments:
- [X] (tvm) Furkan Kamaci
Comments:
- [X] (tvm) Tianqi Chen
Comments:
- [X] (tvm) Markus Weimer
Comments: Looking forward to graduation soon.
### IPMC/Shepherd notes:
TVM is a full stack open deep learning compiler stack for CPUs, GPUs, and
specialized accelerators. It aims to close the gap between the productivity-
focused deep learning frameworks, and the performance- or efficiency-
oriented hardware backends.
TVM has been incubating since 2019-03-06.
### Three most important unfinished issues to address before graduating:
1. Keep growing the community
2. Make a few more Apache releases
3. Improve documentations
### Are there any issues that the IPMC or ASF Board need to be aware of?
no
### How has the community developed since the last report?
TVM community has welcomed two new committers/PPMC members since last
report.
There are also two on-going new committer nomination that will close in a
week.
The community hosted many meetups and an annual developer conference with
more than 200+ attendees.
These contents are made publically available as per apache way.
This has been a great year for us doing the Apache way and grow the
community.
The community has grown 70% in terms of number of contributors and
committers,
while these statistics do not necessarily indicate success, they suggest
we are on the right track.
### How has the project developed since the last report?
We made our first Apache release. Thanks to the help of IPMC members, in
particular Justin, we were able to
hold the Apache standard and release without using the WIP disclaimer
A lot of improvements have been made. Including TensorCore support,
embedded system support and performance improvements
See also our monthly report for detailed improvements
- Sep https://discuss.tvm.ai/t/tvm-monthly-september-2019/4219
- Oct https://discuss.tvm.ai/t/tvm-monthly-oct-2019/4587
- Nov https://discuss.tvm.ai/t/tvm-monthly-nov-2019/5038
### How would you assess the podling's maturity?
Please feel free to add your own commentary.
- [ ] Initial setup
- [ ] Working towards first release
- [x] Community building
- [x] Nearing graduation
- [ ] Other:
### Date of last release:
2019-12-1
### When were the last committers or PPMC members elected?
Dec 1 2019
### Have your mentors been helpful and responsive?
Our mentors are super helpful.
### Is the PPMC managing the podling's brand / trademarks?
We have yet check with the VP.
Given the long history of the project, we want to keep the name,
will reach out to brand formally.
### Signed-off-by:
- [x] (tvm) Byung-Gon Chun
Comments: Great progress! Good job on the first Apache release!
- [ ] (tvm) Sebastian Schelter
Comments:
- [X] (tvm) Henry Saputra
Comments: Community is healthy and congrats on the first release under ASF
- [ ] (tvm) Timothy Chen
Comments:
- [X] (tvm) Furkan Kamaci
Comments:
- [x] (tvm) Tianqi Chen
Comments:
- [x] (tvm) Markus Weimer
Comments: With the first release done, we should look towards
graduation.
### IPMC/Shepherd notes:
Justin Mclean: I think it may be a little too early to consider
graduation. Has the podling filling the the optional maturity model,
doing so may point out when there's still more work to do.
TVM is a full stack open deep learning compiler stack for CPUs, GPUs, and
specialized accelerators. It aims to close the gap between the productivity-
focused deep learning frameworks, and the performance- or efficiency-
oriented hardware backends.
TVM has been incubating since 2019-03-06.
### Three most important unfinished issues to address before graduating:
1. Source code and website migration to ASF infra
2. Make the first Apache release
3. Continue to grow the community
### Are there any issues that the IPMC or ASF Board need to be aware of?
No
### How has the community developed since the last report?
TVM community has welcomed four committers since last report.
We also made our first presentation at ApacheCon NA.
### How has the project developed since the last report?
The project has been quite health, with ~120 pull requests being merged
every months.
These pull requests are authored by a diverse set of contributors(~50
authors last month).
Some highlights of recent developments:
- New integer analysis infrastructure
- Initial computation support
- Support for running quanitized models
For detailed information about the project development, please refer to
the
monthly TVM community's monthly summary:
- Oct: https://discuss.tvm.ai/t/tvm-monthly-september-2019/4219
- Aug: https://discuss.tvm.ai/t/tvm-monthly-august-2019/3904
- July: https://discuss.tvm.ai/t/tvm-monthly-july-2019/3600
- June: https://discuss.tvm.ai/t/tvm-monthly-june-2019/3202
### How would you assess the podling's maturity?
Please feel free to add your own commentary.
- [x] Initial setup
- [x] Working towards first release
- [x] Community building
- [ ] Nearing graduation
- [ ] Other:
### Date of last release:
no release yet
### When were the last committers or PPMC members elected?
Sep 30 2019
### Have your mentors been helpful and responsive?
Mentors are very helpful in providing helpful guidance.
### Signed-off-by:
- [X] (tvm) Byung-Gon Chun
Comments: great progress
- [ ] (tvm) Sebastian Schelter
Comments:
- [X] (tvm) Henry Saputra
Comments: Will start working on repo migration
- [ ] (tvm) Timothy Chen
Comments:
- [X] (tvm) Furkan Kamaci
Comments:
- [x] (tvm) Tianqi Chen
Comments:
### IPMC/Shepherd notes:
TVM is a full stack open deep learning compiler stack for CPUs, GPUs, and
specialized accelerators. It aims to close the gap between the productivity-
focused deep learning frameworks, and the performance- or efficiency-
oriented hardware backends.
TVM has been incubating since 2019-03-06.
### Three most important unfinished issues to address before graduating:
1. Source code and website migration to ASF infra
2. Make the first Apache release
3. Continue to grow the community
### Are there any issues that the IPMC or ASF Board need to be aware of?
N/A
### How has the community developed since the last report?
TVM community has welcomed one committer in the past month.
### How has the project developed since the last report?
Summary statistics: In the past month authors have pushed 142 commits to
master and 142 commits to all branches. On master, 421 files have changed
and there have been 22,680 additions and 5,388 deletions.
Some highlights of recent developments:
- Robust coverage of windows support.
- Improvements in the integer simplifier interface.
- Major improvements in high level ir(relay) support
- For detailed information about the project development, please refer to
the monthly TVM community's monthly summary:
- May: https://discuss.tvm.ai/t/tvm-monthly-may-2019/2793
- June: https://discuss.tvm.ai/t/tvm-monthly-june-2019/3202
### How would you assess the podling's maturity?
Please feel free to add your own commentary.
- [x] Initial setup
- [x] Working towards first release
- [x] Community building
- [ ] Nearing graduation
- [ ] Other:
### Date of last release:
N/A
### When were the last committers or PPMC members elected?
June 13 2019
### Have your mentors been helpful and responsive?
Mentors are very helpful in providing guidance.
### Signed-off-by:
- [X] (tvm) Byung-Gon Chun
Comments:
- [ ] (tvm) Sebastian Schelter
Comments:
- [X] (tvm) Henry Saputra
Comments: Working on migration or resources to ASF infra and managed
resources. But the community still growing and active.
- [ ] (tvm) Timothy Chen
Comments:
- [X] (tvm) Furkan Kamaci
Comments:
- [X] (tvm) Markus Weimer
Comments:
### IPMC/Shepherd notes:
TVM is a full stack open deep learning compiler stack for CPUs, GPUs, and
specialized accelerators. It aims to close the gap between the productivity-
focused deep learning frameworks, and the performance- or
efficiency-oriented
hardware backends.
TVM has been incubating since 2019-03-06.
### Three most important unfinished issues to address before graduating:
1. Source code and website migration to ASF infra
2. Make the first Apache release
3. Continue to grow the community
### Any issues that the Incubator PMC (IPMC) or ASF Board wish/need to be
aware of?
No answer.
### How has the community developed since the last report?
TVM community has welcomed one PPMC member in the past month. There is
also ongoing votes about adding new committers.
### How has the project developed since the last report?
Summary statistics: in the past month, 44 contributors have pushed 98
commits to master and 98 commits to all branches. On master, 366 files
have changed and there have been 21,156 additions and 2,810 deletions.
The contributions covers areas including documentations, bugfixes, user
interface and backend hardware support.
Some highlights of recent developments:
- More robust frontend support to support various machine learning models.
- Cycle accurate simulation to make it easy to add new architecture
backends.
- Quantized models to enable deployment to embedded devices.
For detailed information about the project development, please refer to
the monthly TVM community's monthly summary:
- April: https://discuss.tvm.ai/t/tvm-monthly-april-2019/2426
- May: https://discuss.tvm.ai/t/tvm-monthly-may-2019/2793
### How would you assess the podling's maturity?
Please feel free to add your own commentary.
- [x] Initial setup
- [x] Working towards first release
- [x] Community building
- [ ] Nearing graduation
- [ ] Other:
### Date of last release:
no release yet
### When were the last committers or PPMC members elected?
May 23rd
### Have your mentors been helpful?
Mentors are very helpful in helping to set up the podling and provide
helpful guidance.
### Signed-off-by:
- [X] (tvm) Markus Weimer
Comments:
- [X] (tvm) Byung-Gon Chun
Comments:
- [ ] (tvm) Sebastian Schelter
Comments:
- [X] (tvm) Henry Saputra
Comments:
- [ ] (tvm) Timothy Chen
Comments:
- [X] (tvm) Furkan Kamaci
Comments:
### IPMC/Shepherd notes:
Dave Fisher: This already busy project and community is just getting
started in the Incubator.
They look to be on the right track. They have some resources which may
require a VM.
I suggested that they reach out to Infra on Slack for some answers.
It's good to see mentor engagement.
TVM is a full stack open deep learning compiler stack for CPUs, GPUs, and specialized accelerators. It aims to close the gap between the productivity- focused deep learning frameworks, and the performance- or efficiency-oriented hardware backends. TVM has been incubating since 2019-03-06. Three most important unfinished issues to address before graduating: 1. Source code and website migration to ASF infra 2. Make the first Apache release 3. Continue to grow the community Any issues that the Incubator PMC (IPMC) or ASF Board wish/need to be aware of? How has the community developed since the last report? TVM community has welcomed one new committer in the past month How has the project developed since the last report? In the past month, 51 contributors have pushed 109 commits to master and 109 commits to all branches. On master, 1,763 files have changed and there have been 44,070 additions and 13,852 deletions. See detailed monthly report - Mar: https://discuss.tvm.ai/t/tvm-monthly-march-2019/2083 - April: https://discuss.tvm.ai/t/tvm-monthly-april-2019/2426 How would you assess the podling's maturity? Please feel free to add your own commentary. [x] Initial setup [x] Working towards first release [x] Community building [ ] Nearing graduation [ ] Other: Date of last release: not yet When were the last committers or PPMC members elected? 2019 April 22 Have your mentors been helpful and responsive or are things falling through the cracks? In the latter case, please list any open issues that need to be addressed. Mentors are very helpful in helping to set up the podling and provide helpful guidance. Signed-off-by: [X] (tvm) Markus Weimer Comments: [ ](tvm) Byung-Gon Chun Comments: [ ](tvm) Sebastian Schelter Comments: [X](tvm) Henry Saputra Comments: [ ](tvm) Timothy Chen Comments: [X](tvm) Furkan Kamaci Comments: IPMC/Shepherd notes:
TVM is a full stack open deep learning compiler stack for CPUs, GPUs, and
specialized accelerators. It aims to close the gap between the productivity-
focused deep learning frameworks, and the performance- or
efficiency-oriented
hardware backends.
TVM has been incubating since 2019-03-06.
Three most important unfinished issues to address before graduating:
1. Finish the IP clearance
2. Make the first Apache release
3. Continue to grow the community
Any issues that the Incubator PMC (IPMC) or ASF Board wish/need to be
aware of?
None
How has the community developed since the last report?
- TVM PMC has been fully transitioned to private@
- The development conversations are mirrored to dev@
- TVM community has welcomed two new committers in the past month
See more in https://github.com/dmlc/tvm/search?q=COMMUNITY&type=Issues
NOTE: due to the transition, the vote of the new committers were done in
the pre-apache PMC private list with our mentors as observers.
We are doing procedural votes to bring these committers formally to
Apache soon.
How has the project developed since the last report?
In the past month, 47 contributors have pushed 138 commits.
On master, 480 files have changed and there have been
18,529 additions and 4,290 deletions.
See details in our monthly report
Mar: https://discuss.tvm.ai/t/tvm-monthly-march-2019/2083/2
Feb: https://discuss.tvm.ai/t/tvm-monthly-feb-2019/1801
How would you assess the podling's maturity?
Please feel free to add your own commentary.
[X] Initial setup
[X] Working towards first release
[X] Community building
[ ] Nearing graduation
[ ] Other:
Date of last release:
None yet
When were the last committers or PPMC members elected?
2019 Mar 15
Have your mentors been helpful and responsive or are things falling
through the cracks? In the latter case, please list any open issues
that need to be addressed.
Mentors are very helpful in helping to set up the podling and provide
helpful guidances.
Signed-off-by:
[X](tvm) Byung-Gon Chun
Comments:
[X](tvm) Sebastian Schelter
Comments:
[X](tvm) Henry Saputra
Comments:
[X](tvm) Timothy Chen
Comments:
[X](tvm) Furkan Kamaci
Comments:
[X](tvm) Markus Weimer
Comments:
IPMC/Shepherd notes: