hash mismatch when installing torch dependencies from cu126 index

[rbavery]

Summary

Apologies if this isn't the correct place to raise this issue, not sure where to raise it after reading some past issues here related to hash mismatches.

I have a simple toml that fails to solve due to hash mismatch

[project]
name = "test-toml"
version = "0.1.0"
description = "model"
readme = "README.md"
requires-python = "==3.11.*"
classifiers = [ "Programming Language :: Python :: 3 :: Only", "Programming Language :: Python :: 3.11" ]
dependencies = [
  "pytorch-triton",
  "torch",
  "torchvision",
]

[[tool.uv.index]]
name = "pytorch-cu126"
url = "https://download.pytorch.org/whl/cu126"

uv sync fails with

→ uv sync
Resolved 31 packages in 16ms
  × Failed to download `nvidia-nvtx-cu12==12.6.77`
  ╰─▶ Hash mismatch for `nvidia-nvtx-cu12==12.6.77`

      Expected:
        sha256:6574241a3ec5fdc9334353ab8c479fe75841dbe8f4532a8fc97ce63503330ba1

      Computed:
        sha256:b90bed3df379fa79afbd21be8e04a0314336b8ae16768b58f2d34cb1d04cd7d2
  help: `nvidia-nvtx-cu12` (v12.6.77) was included because `test-toml` (v0.1.0) depends on `torch`
        (v2.9.1+cu126) which depends on `nvidia-nvtx-cu12`

Originally reported here, https://github.com/NVIDIA/NVTX/issues/129

I've tried to globally uv cache clean, recreate venv, but same result.

Platform

Ubuntu 22.04

Version

0.9.10

Python version

Python 3.11