# RFC **URL:** https://discuss.tvm.apache.org/c/development/rfc/8.md [Latest](https://discuss.tvm.apache.org/latest.md) · [Categories](https://discuss.tvm.apache.org/categories.md) --- ## [About the RFC category](https://discuss.tvm.apache.org/t/about-the-rfc-category/78)
**Author:** [@tqchen](https://discuss.tvm.apache.org/u/tqchen)\ **Replies:** 0\ **Last updated:** [April 19, 2018, 3:07am UTC](https://discuss.tvm.apache.org/t/about-the-rfc-category/78 "2018-04-19T03:07:14Z")
Discussion about design proposals. --- ## [\[DISCUSS\] Split TIRX foundation into a separate repo](https://discuss.tvm.apache.org/t/discuss-split-tirx-foundation-into-a-separate-repo/19116)
**Author:** [@tqchen](https://discuss.tvm.apache.org/u/tqchen)\ **Replies:** 4\ **Last updated:** [October 9, 2026, 9:40am UTC](https://discuss.tvm.apache.org/t/discuss-split-tirx-foundation-into-a-separate-repo/19116 "2026-10-09T09:40:56Z")
Last year, we had great success in bringing tvm ffi into its own independent module, as of now, tvm-ffi serves as a standalone ABI that powers many solutions in ML systems and compilers. The trend of agentic era brings … --- ## [\[pre-rfc\] bazel overlay](https://discuss.tvm.apache.org/t/pre-rfc-bazel-overlay/18976)
**Author:** [@Wheest](https://discuss.tvm.apache.org/u/Wheest)\ **Replies:** 0\ **Last updated:** [March 31, 2026, 3:28pm UTC](https://discuss.tvm.apache.org/t/pre-rfc-bazel-overlay/18976 "2026-03-31T15:28:58Z")
In this post, there was a discussion about adding bazel support. This was a few years ago, and since then Bazel is now being used in more projects, and also has an increasingly mature upstream LLVM overlay. I was looki… --- ## [Proposal: A Lighter, Faster RFC Process](https://discuss.tvm.apache.org/t/proposal-a-lighter-faster-rfc-process/18567)
**Author:** [@Hzfengsy](https://discuss.tvm.apache.org/u/Hzfengsy)\ **Replies:** 4\ **Last updated:** [February 26, 2026, 2:31am UTC](https://discuss.tvm.apache.org/t/proposal-a-lighter-faster-rfc-process/18567 "2026-02-26T02:31:11Z")
Hi TVM community, Our current RFC process, which requires a PR to the tvm-rfcs repo, is robust but can be too heavyweight for the rapid pace of AI development. To encourage more ideas and faster iteration, I propose we … --- ## [Evolving and Modernize Tensor-level IR](https://discuss.tvm.apache.org/t/evolving-and-modernize-tensor-level-ir/18852)
**Author:** [@tqchen](https://discuss.tvm.apache.org/u/tqchen)\ **Replies:** 4\ **Last updated:** [December 25, 2025, 1:07pm UTC](https://discuss.tvm.apache.org/t/evolving-and-modernize-tensor-level-ir/18852 "2025-12-25T13:07:26Z")
Over the past year, we have been successfully modernizing the foundational ffi module that is now helpful to the broader community. This thread aims to begin a discussion on modernizing the TIR components. Up until now, … --- ## [\[DISCUSS\] Bring TVM FFI into a Seperate Repo](https://discuss.tvm.apache.org/t/discuss-bring-tvm-ffi-into-a-seperate-repo/18566)
**Author:** [@tqchen](https://discuss.tvm.apache.org/u/tqchen)\ **Replies:** 7\ **Last updated:** [September 18, 2025, 12:13am UTC](https://discuss.tvm.apache.org/t/discuss-bring-tvm-ffi-into-a-seperate-repo/18566 "2025-09-18T00:13:06Z")
We have been working a series of refactors to rebuild the ffi mechanism from ground up As of the above PR, the ffi module is now a standalone module build from ground up, that takes minimal dependencies from the rest … --- ## [\[RFC\] DietCode: An Auto-scheduler for Dynamic Tensor Programs](https://discuss.tvm.apache.org/t/rfc-dietcode-an-auto-scheduler-for-dynamic-tensor-programs/12757)
**Author:** [@ArmageddonKnight](https://discuss.tvm.apache.org/u/ArmageddonKnight)\ **Replies:** 12\ **Last updated:** [August 29, 2025, 4:10am UTC](https://discuss.tvm.apache.org/t/rfc-dietcode-an-auto-scheduler-for-dynamic-tensor-programs/12757 "2025-08-29T04:10:17Z")
Feature Name: DietCode: An Auto-Scheduler for Dynamic Tensor Programs Start Date: (2022-05-10) RFC PR: apache/tvm-rfcs#xx GitHub Issue: apache/tvm#yy Summary We propose to integrate DietCode, an auto-scheduler for dyna… --- ## [Relax Python Module Design](https://discuss.tvm.apache.org/t/relax-python-module-design/18272)
**Author:** [@MasterJH5574](https://discuss.tvm.apache.org/u/MasterJH5574)\ **Replies:** 4\ **Last updated:** [August 28, 2025, 1:21am UTC](https://discuss.tvm.apache.org/t/relax-python-module-design/18272 "2025-08-28T01:21:55Z")
Motivation As machine learning models—especially large language models—continue to grow in scale, there is an increasing demand for ML compiler runtimes to integrate more deeply with the Python ecosystem. Python-based fr… --- ## [\[RFC-109\] Add a new backend NNAPI for BYOC](https://discuss.tvm.apache.org/t/rfc-109-add-a-new-backend-nnapi-for-byoc/17717)
**Author:** [@mengshyu](https://discuss.tvm.apache.org/u/mengshyu)\ **Replies:** 2\ **Last updated:** [September 20, 2024, 9:12pm UTC](https://discuss.tvm.apache.org/t/rfc-109-add-a-new-backend-nnapi-for-byoc/17717 "2024-09-20T21:12:25Z")
Co-authored-by: Ming-Long Huang mlhuang@pllab.cs.nthu.edu.tw Co-authored-by: HMZ mzhuang@pllab.cs.nthu.edu.tw Hi, we are from the Programming Language Lab, National Tsing Hua University, Taiwan. We have been working on… --- ## [\[RFC\]\[OpenCLML\] OpenCL ML integration into TVM as BYOC](https://discuss.tvm.apache.org/t/rfc-openclml-opencl-ml-integration-into-tvm-as-byoc/11911)
**Author:** [@srkreddy1238](https://discuss.tvm.apache.org/u/srkreddy1238)\ **Replies:** 10\ **Last updated:** [April 5, 2023, 4:46pm UTC](https://discuss.tvm.apache.org/t/rfc-openclml-opencl-ml-integration-into-tvm-as-byoc/11911 "2023-04-05T16:46:09Z")
Summary: OpenCL ML is an extension (cl\_qcom\_ml\_ops) over OpenCL spec developed by Qualcomm to accelerate the machine learning at operation level. OpenCL SDK is publicly available at OpenCL Machine Learning Acceleration … --- ## [\[pre-RFC\] TVM Explorer Infrastructure](https://discuss.tvm.apache.org/t/pre-rfc-tvm-explorer-infrastructure/13457)
**Author:** [@chunit](https://discuss.tvm.apache.org/u/chunit)\ **Replies:** 35\ **Last updated:** [September 11, 2024, 7:42am UTC](https://discuss.tvm.apache.org/t/pre-rfc-tvm-explorer-infrastructure/13457 "2024-09-11T07:42:52Z")
Hi community: We would like to propose a Pre-RFC to enhance the debugging ability of TVM. This Pre-RFC will describe the details of how do we enhance the current functionality. On the other hand we will mention about wh… --- ## [\[pre-RFC\] Language Specification for TIR](https://discuss.tvm.apache.org/t/pre-rfc-language-specification-for-tir/14844)
**Author:** [@slyubomirsky](https://discuss.tvm.apache.org/u/slyubomirsky)\ **Replies:** 11\ **Last updated:** [July 2, 2024, 9:25am UTC](https://discuss.tvm.apache.org/t/pre-rfc-language-specification-for-tir/14844 "2024-07-02T09:25:00Z")
Pre-RFC: Language Specification for TIR Thanks to the assistance of several core TIR contributors at OctoML, I have written a draft of a language specification for TIR. I present it here to the community for further revi… --- ## [\[Relay\]\[Onnx\] Support saveing relay ir to onnx for visualization](https://discuss.tvm.apache.org/t/relay-onnx-support-saveing-relay-ir-to-onnx-for-visualization/17020)
**Author:** [@ShawnZhuang](https://discuss.tvm.apache.org/u/ShawnZhuang)\ **Replies:** 0\ **Last updated:** [April 4, 2024, 1:02pm UTC](https://discuss.tvm.apache.org/t/relay-onnx-support-saveing-relay-ir-to-onnx-for-visualization/17020 "2024-04-04T13:02:13Z")
relay ir may not be easy to be viewed in netron. Awareness of distinct Attr types is required in some solutions. this commit(\[relay\]\[feature\] save relay IR as onnx for visualize by ShawnZhuang · Pull Request #16847 · ap… --- ## [CSourceMetaData Module : A CSourceModule to Hold Function Registry for uTVM](https://discuss.tvm.apache.org/t/csourcemetadata-module-a-csourcemodule-to-hold-function-registry-for-utvm/8554)
**Author:** [@manupa-arm](https://discuss.tvm.apache.org/u/manupa-arm)\ **Replies:** 18\ **Last updated:** [March 18, 2024, 9:05pm UTC](https://discuss.tvm.apache.org/t/csourcemetadata-module-a-csourcemodule-to-hold-function-registry-for-utvm/8554 "2024-03-18T21:05:13Z")
There had been a need for a common c-source module to hold the metadata for uTVM where it mostly deals with “c” or “llvm” modules. One of the main need is to have a model-wide function registry for c-runtime in the ab… --- ## [\[RFC\]\[BYOC\] Arm Compute Library integration](https://discuss.tvm.apache.org/t/rfc-byoc-arm-compute-library-integration/7082)
**Author:** [@lhutton1](https://discuss.tvm.apache.org/u/lhutton1)\ **Replies:** 8\ **Last updated:** [March 18, 2024, 11:49am UTC](https://discuss.tvm.apache.org/t/rfc-byoc-arm-compute-library-integration/7082 "2024-03-18T11:49:58Z")
Motivation Arm Compute Library (ACL) is an open source project that provides hand-crafted assembler routines for Arm CPU’s and GPU’s. This integration will look at how we can accelerate CPU performance for Arm devices in… --- ## [\[DISCUSS\]Introduce RISC-V Vector/Matrix extension](https://discuss.tvm.apache.org/t/discuss-introduce-risc-v-vector-matrix-extension/16490)
**Author:** [@zhupijuan\_lkl](https://discuss.tvm.apache.org/u/zhupijuan_lkl)\ **Replies:** 6\ **Last updated:** [March 16, 2024, 8:53am UTC](https://discuss.tvm.apache.org/t/discuss-introduce-risc-v-vector-matrix-extension/16490 "2024-03-16T08:53:19Z")
Hi All, Background Recently, RISC-V has been developing rapidly. Among them, hardware extensions for accelerating AI applications (such as Vector/Matrix extensions) are gradually being commercialized, such as products l… --- ## [\[RFC\]\[Tensor Core\] Optimization of CNNs on Tensor Core](https://discuss.tvm.apache.org/t/rfc-tensor-core-optimization-of-cnns-on-tensor-core/6004)
**Author:** [@Shawn\_Inspur](https://discuss.tvm.apache.org/u/Shawn_Inspur)\ **Replies:** 37\ **Last updated:** [February 22, 2024, 8:47am UTC](https://discuss.tvm.apache.org/t/rfc-tensor-core-optimization-of-cnns-on-tensor-core/6004 "2024-02-22T08:47:49Z")
Introduction The optimizations were built on existing work of @Hzfengsy, see #4052 for details. We added tensor core enabled conv2d, dense, and Tensor Core instructions in Topi, and modified codes in Relay to enable aut… --- ## [\[RFC\] Sphinx and the Documentation Refactor](https://discuss.tvm.apache.org/t/rfc-sphinx-and-the-documentation-refactor/8705)
**Author:** [@areusch](https://discuss.tvm.apache.org/u/areusch)\ **Replies:** 1\ **Last updated:** [March 8, 2021, 11:19pm UTC](https://discuss.tvm.apache.org/t/rfc-sphinx-and-the-documentation-refactor/8705 "2021-03-08T23:19:06Z")
Overview Documentation is one of the pieces of TVM that we would like to improve in the near future. Currently, much of our documentation is tutorial-focused, but we would like to add several new types of documentation i… --- ## [\[RFC\]\[Quantization\] Quantization in TVM](https://discuss.tvm.apache.org/t/rfc-quantization-quantization-in-tvm/9161)
**Author:** [@electriclilies](https://discuss.tvm.apache.org/u/electriclilies)\ **Replies:** 14\ **Last updated:** [January 9, 2024, 6:17am UTC](https://discuss.tvm.apache.org/t/rfc-quantization-quantization-in-tvm/9161 "2024-01-09T06:17:27Z")
Quantization in TVM (This RFC corresponds to a PR #7474) The goal of this work is to create a flexible and extensible framework for quantizing and calibrating models. Specifically, I want to Allow arbitrary patterns… --- ## [\[preRFC\]\[BYOC\] Integrating Marvell ML/AI Accelerator to the TVM BYOC framework](https://discuss.tvm.apache.org/t/prerfc-byoc-integrating-marvell-ml-ai-accelerator-to-the-tvm-byoc-framework/16155)
**Author:** [@f2013519](https://discuss.tvm.apache.org/u/f2013519)\ **Replies:** 0\ **Last updated:** [December 20, 2023, 11:45am UTC](https://discuss.tvm.apache.org/t/prerfc-byoc-integrating-marvell-ml-ai-accelerator-to-the-tvm-byoc-framework/16155 "2023-12-20T11:45:53Z")
Summary Integrate Marvell’s Machine Learning Inference Processor (MLIP) with TVM BYOC framework to bring the TVM ecosystem to Marvell customers. This preRFC deprecates this previous preRFC thread \[pre-RFC\]\[BYOC\] Marvell… --- ## [\[pre-RFC\]\[BYOC\] Marvell ML/AI Accelerator Integration](https://discuss.tvm.apache.org/t/pre-rfc-byoc-marvell-ml-ai-accelerator-integration/11691)
**Author:** [@ccjoechou](https://discuss.tvm.apache.org/u/ccjoechou)\ **Replies:** 1\ **Last updated:** [December 6, 2023, 3:25am UTC](https://discuss.tvm.apache.org/t/pre-rfc-byoc-marvell-ml-ai-accelerator-integration/11691 "2023-12-06T03:25:37Z")
Summary Integrate Marvell’s ML/AI accelerator with TVM BYOC framework in order to bring the TVM ecosystem to Marvell customers. Motivation Marvell MLIP is an ML/AI inference accelerator and is embedded on our ARM Neov… --- ## [\[Frontend\] Preserving Pytorch Spans](https://discuss.tvm.apache.org/t/frontend-preserving-pytorch-spans/16044)
**Author:** [@navya-mehta](https://discuss.tvm.apache.org/u/navya-mehta)\ **Replies:** 9\ **Last updated:** [November 30, 2023, 7:38am UTC](https://discuss.tvm.apache.org/t/frontend-preserving-pytorch-spans/16044 "2023-11-30T07:38:49Z")
Hey all, wanted to discuss a new feature proposal with the community. I’ve not contributed to TVM before so please let me know if there is a strict process to follow. When loading a traced pytorch model using tvm.relay.f… --- ## [\[RFC\] TVM Target Specification](https://discuss.tvm.apache.org/t/rfc-tvm-target-specification/6844)
**Author:** [@tqchen](https://discuss.tvm.apache.org/u/tqchen)\ **Replies:** 61\ **Last updated:** [November 8, 2023, 7:49am UTC](https://discuss.tvm.apache.org/t/rfc-tvm-target-specification/6844 "2023-11-08T07:49:15Z")
In collaboration with @areusch A target object in TVM provides all the information needed for code lowering and code generation. Currently, a target is specified by a string in the format of \ \[attributes\], wh… --- ## [Target and Attributes](https://discuss.tvm.apache.org/t/target-and-attributes/6013)
**Author:** [@tqchen](https://discuss.tvm.apache.org/u/tqchen)\ **Replies:** 12\ **Last updated:** [November 8, 2023, 3:18am UTC](https://discuss.tvm.apache.org/t/target-and-attributes/6013 "2023-11-08T03:18:39Z")
Motivation We start to introduce quite a few target-related concepts when we start to add more flexibilities into code generation, autotvm and tensorization. One particular example is the recent introduced “compiler” con… --- ## [\[RFC\]\[BYOC\] Android NNAPI Integration](https://discuss.tvm.apache.org/t/rfc-byoc-android-nnapi-integration/9072)
**Author:** [@melsonlai](https://discuss.tvm.apache.org/u/melsonlai)\ **Replies:** 10\ **Last updated:** [November 1, 2023, 4:24pm UTC](https://discuss.tvm.apache.org/t/rfc-byoc-android-nnapi-integration/9072 "2023-11-01T16:24:27Z")
Motivation Android NNAPI is a graph-level neural network inference API provided by the Android runtime. It’s intended to provide the best execution setup to machine learning inferences based on available resources, inclu… --- ## [\[TIR\] Negative or zero extent in for loop](https://discuss.tvm.apache.org/t/tir-negative-or-zero-extent-in-for-loop/15777)
**Author:** [@kparzysz](https://discuss.tvm.apache.org/u/kparzysz)\ **Replies:** 0\ **Last updated:** [October 5, 2023, 4:17pm UTC](https://discuss.tvm.apache.org/t/tir-negative-or-zero-extent-in-for-loop/15777 "2023-10-05T16:17:48Z")
I’m thinking of implementing loop peeling in schedule. Some loops that we’re working with contain conditional code which is true/false for all iterations except the last. Separating (peeling) the last iteration from th… --- ## [\[RFC\]\[VTA\]A HLS C VTA bug](https://discuss.tvm.apache.org/t/rfc-vta-a-hls-c-vta-bug/6743)
**Author:** [@ffffc](https://discuss.tvm.apache.org/u/ffffc)\ **Replies:** 5\ **Last updated:** [January 21, 2022, 8:48am UTC](https://discuss.tvm.apache.org/t/rfc-vta-a-hls-c-vta-bug/6743 "2022-01-21T08:48:52Z")
VTA can’t get correct in Zynq Ultra Scale based device. We once thought this was a coherency problem, but recently I found there may be a bug in HLS C VTA hardware with my partner @hht. Only FINISH instruction can set V… --- ## [\[RFC\] Introduce TPAT (TVM Plugin Autogen Tool)](https://discuss.tvm.apache.org/t/rfc-introduce-tpat-tvm-plugin-autogen-tool/15520)
**Author:** [@Civitasv](https://discuss.tvm.apache.org/u/Civitasv)\ **Replies:** 0\ **Last updated:** [August 14, 2023, 9:18am UTC](https://discuss.tvm.apache.org/t/rfc-introduce-tpat-tvm-plugin-autogen-tool/15520 "2023-08-14T09:18:47Z")
Summary This RFC introduce TPAT, which stands for TVM Plugin Autogen Tool. It utilizes TVM to generate optimized plugins/extensions for vendor-specific acceleration library, in this way, we can effectively to combine … --- ## [\[Process RFC\] Clarify Community Strategy Decision Process](https://discuss.tvm.apache.org/t/process-rfc-clarify-community-strategy-decision-process/15463)
**Author:** [@tqchen](https://discuss.tvm.apache.org/u/tqchen)\ **Replies:** 11\ **Last updated:** [August 10, 2023, 3:17pm UTC](https://discuss.tvm.apache.org/t/process-rfc-clarify-community-strategy-decision-process/15463 "2023-08-10T15:17:11Z")
Dear community, there has been quite a bit of discussions on how we can collectively make strategic decisions forward, including posts by @Hzfengsy and others. After reviewing our conversations in the past years and exi… --- ## [Adding support for Silicon Labs HW accelerator](https://discuss.tvm.apache.org/t/adding-support-for-silicon-labs-hw-accelerator/15279)
**Author:** [@silabs-kjetil](https://discuss.tvm.apache.org/u/silabs-kjetil)\ **Replies:** 3\ **Last updated:** [July 12, 2023, 8:15am UTC](https://discuss.tvm.apache.org/t/adding-support-for-silicon-labs-hw-accelerator/15279 "2023-07-12T08:15:43Z")
Introduction Silicon Labs produces low-power wireless Cortex-M based MCUs. Some of the MCUs, such as the EFR32xG24, contain an AI/ML hardware accelerator called MVP (Matrix Vector Processor). The goal of this post is to … [Next page](https://discuss.tvm.apache.org/c/development/rfc/8.md?page=1)