[VOTE] Release Apache TVM v0.26.0.rc0

[MasterJH5574]

Dear TVM community,

This is a call for vote to release Apache TVM version v0.26.0.

Release highlights:

• TIRx bring-up continues: buffers are now typed variables (BufferType), a first-class Return statement, a dedicated boolean-buffer lowering pass, and scalable Ramp lowering on LLVM/RVV.
• TIRx CUDA tile primitives extended with tcgen05 dispatch paths, FA4 and CLC intrinsics, nvfp4 GEMM, and FlashMLA lowering.
• IR type-system refactor: PrimExpr types unified on PrimType instead of DataType, Var identity unified across IR/Relax/TIRx, and Downcast plus the Relax expression wrappers phased out.
• Relax ONNX frontend: GroupNormalization, modern QDQ opset attributes, Shape start/end, 3D AffineGrid with align_corners, dynamic Gather on shape, and an LpPool conversion fix.
• Relax TFLite and PyTorch frontends: static hashtable find, dynamic RANGE and DYNAMIC_UPDATE_SLICE, reverse_sequence, an rnn_tanh.input converter, and symbolic scalar binding in from_fx.
• Relax legalization fixes: grouped conv with symbolic channels, dilated conv_transpose, nn.dropout as an inference no-op, shape_to_tensor on device, bucketize output dtype, and int64 row indices in GPU multinomial sampling.
• Arithmetic: optional Z3-backed proving in Analyzer, Analyzer::Clone for deep-copying analyzer state, and correctness fixes in const-int-bound tightening for Mod/FloorMod and floormod coefficient reduction.
• tvm-ffi bumped to 0.1.13.post2, bringing the stable Optional ABI layout and thread-safe packed-function initialization.
• Backends, runtime and build: Metal 4 shader compilation, a SPIR-V 1.4+ entry-point fix for Vulkan, the FlashInfer 0.6.3 KVCache backend, RISC-V target device keys, CUDA extras narrowed to cuda-bindings, and the Z3 dependency migrated to mlc-z3-static.
• WebGPU/WASM: batched GPU-to-GPU copies with synchronization fixes, a custom allocator linked into the WASM runtime, and deterministic RNG state restore.

Link to release candidate:
https://dist.apache.org/repos/dist/dev/tvm/tvm-v0.26.0-rc0/

Link to GitHub pre-release page (full release notes):
https://github.com/apache/tvm/releases/tag/v0.26.0.rc0

Everyone is welcomed to vote. Please vote by replying to this thread explicitly. According to release procedure, for the dev@ vote, there must be at least 3 binding +1 votes and more +1 votes than -1 votes.

+1 = approve
+0 = no opinion
-1 = disapprove (provide reason)

NOTE: this thread is being mirrored in dev@

This vote will remain open for at least 72 hours.

[MasterJH5574]

Verified the v0.26.0.rc0 candidate two ways — (A) building from the signed source release, and (B) installing the published wheel from PyPI. Kept separate since they are different installation paths.

A. Source build check — tests/scripts/release/test_release_package.sh

Run with the script as shipped in the release candidate:
https://github.com/apache/tvm/blob/v0.26.0.rc0/tests/scripts/release/test_release_package.sh

$ bash test_release_package.sh

+ version=v0.26.0
+ rc=rc0
[setup] Create an isolated virtualenv ...
+ python3 -m venv .venv
...
IGNORE
...
[1/10] Downloading from apache.org ...
...
IGNORE
...
[2/10] Downloading from github.com ...
...
IGNORE
...
[3/10] Check difference between github.com and apache.org ...
+ diff github/md5sum.txt ./apache/md5sum.txt
[4/10] Checking asc ...
+ gpg --verify ./apache-tvm-src-v0.26.0.rc0.tar.gz.asc ./apache-tvm-src-v0.26.0.rc0.tar.gz
gpg:                using RSA key 3EA0F45C3223CF0D81C8F2CCA9F8369106601185
gpg: Good signature from "Ruihang Lai <ruihangl@apache.org>" [unknown]
[5/10] Checking sha512 ...
+ sha512sum -c ./apache-tvm-src-v0.26.0.rc0.tar.gz.sha512
apache-tvm-src-v0.26.0.rc0.tar.gz: OK
[6/10] Unzip ...
+ tar -zxf apache-tvm-src-v0.26.0.rc0.tar.gz
[7/10] Checking whether binary in source code ...
[8/10] Compile and Python Import on Linux ...
+ cmake ..
+ make -j16
...
IGNORE
...
[100%] Built target tvm_objs
[100%] Linking CXX shared library lib/libtvm_compiler.so
[100%] Built target tvm_compiler
[9/10] Install Python runtime dependencies (declared in the source pyproject.toml) ...
Installing declared runtime deps: apache-tvm-ffi>=0.1.13.post2 ml_dtypes numpy typing_extensions
Successfully installed apache-tvm-ffi-0.1.13.post2 ...
[10/10] Import TVM and print path ...
+ python3 -c 'import tvm; print(tvm.__path__)'
['.../github/apache-tvm-src-v0.26.0/python/tvm']

LGTM — md5 (github vs apache.org) identical, GPG signature good (verified against the dev KEYS), sha512 OK, no binaries in source, full compile with no warnings, and import tvm succeeds.

B. Wheel install check — PyPI

All four platform wheels published for apache-tvm==0.26.0rc0 (macOS arm64, Linux x86_64 + aarch64 manylinux_2_28, Windows AMD64). Installed the Linux x86_64 wheel in a clean venv — dependencies (including apache-tvm-ffi) resolve automatically, no source build:

$ python3 -m venv venv && . venv/bin/activate
$ pip install apache-tvm==0.26.0rc0
...
Successfully installed apache-tvm-0.26.0rc0 apache-tvm-ffi-0.1.13.post2 ...
$ python -c "import tvm; print(tvm.__version__)"
0.26.0rc0

LGTM — wheel installs with the declared dependencies (apache-tvm-ffi>=0.1.13.post2) and imports at the expected version. (It is a pre-release, so a bare pip install apache-tvm still resolves the latest stable; testers opt into the RC by pinning ==0.26.0rc0 or using --pre.)

[tlopex]

+1

[spectrometerHBH]

+1

[tqchen]

+1

[cchung100m]

+1

[akaashrp]

+1

[yongwww]

+1