Installing PyTorch on Fedora: the 2 GiB repo route versus a 209 MB pip CPU wheel

There are two ways to install PyTorch on Fedora, and the measured cost differs by an order of magnitude.

Route one: the Fedora repo (not recommended)

dnf list --available python3-torch

python3-torch.x86_64   2.12.0-7.fc45   fedora

That looks like 399 MiB of download and 755 MiB installed, but the real run says:

Total size of inbound packages is 2 GiB. Need to download 2 GiB.
After this operation, 5 GiB extra will be used (install 5 GiB, remove 3 MiB).
[  1/747] cpuinfo-0:25.11.14-1.gitf858c 100% | 14.9 KiB/s | 44.7 KiB | 00m03s

747 packages, 2 GiB of downloads, 5 GiB of disk. On this lab machine the download cache grew at roughly 43 KB/s, which puts the remainder at well over ten hours.

Item	Value
total download	2 GiB
dependency packages	747
disk usage	5 GiB
measured rate here	~43 KB/s

Route two: the pip CPU wheel (recommended)

First check pip — Fedora does not install it by default:

python3 -m pip --version

/usr/bin/python3: No module named pip

Install it, then create a virtual environment:

sudo dnf install python3-pip
python3 -m venv ~/torch-venv

The measured python3-pip version is 26.1.2-5.fc45.

Now pull the CPU-only wheel:

~/torch-venv/bin/pip install --no-deps \
  --index-url https://download.pytorch.org/whl/cpu \
  "torch==2.15.0+cpu"

Downloading torch-2.15.0%2Bcpu-cp315-cp315-manylinux_2_28_x86_64.whl (209.5 MB)
   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 209.5/209.5 MB 5.9 MB/s  0:00:37

209.5 MB at 5.9 MB/s, 37 seconds, versus 2 GiB at 43 KB/s — two orders of magnitude.

Caution

--index-url https://download.pytorch.org/whl/cpu points every dependency at pytorch.org. Small packages there are slow or stall — measured networkx (2.1 MB) at 29.3 kB/s and sympy (6.3 MB) unfinished after ten-plus minutes.

So split the commands: install the main wheel with --no-deps, then fetch dependencies from a domestic mirror (next section).

Filling in the dependencies

import torch actually needs only a few small packages:

~/torch-venv/bin/pip install \
  --index-url https://pypi.tuna.tsinghua.edu.cn/simple \
  typing_extensions filelock jinja2 fsspec

After that import torch works — sympy and networkx are not required.

NumPy is optional, but leaving it out produces a warning:

UserWarning: Failed to initialize NumPy: No module named 'numpy'

If .numpy() should work, add it:

~/torch-venv/bin/pip install numpy

Verifying the install

import torch, time
print("version:", torch.__version__)
print("cuda available:", torch.cuda.is_available())
print("threads:", torch.get_num_threads())
a = torch.randn(512, 512); b = torch.randn(512, 512)
t0 = time.time(); c = a @ b; dt = (time.time() - t0) * 1000
print("matmul 512x512: %.2f ms" % dt)
print("shape:", tuple(c.shape))
print("mean:", float(torch.tensor([1.0, 2.0, 3.0]).mean()))

Measured output:

version: 2.15.0+cpu
cuda available: False
threads: 4
matmul 512x512: 2.11 ms
shape: (512, 512)
mean: 2.0

Four things stand out:

• the version carries the +cpu suffix — a CPU-only build
• cuda available: False is correct on a machine without a GPU, not a mistake
• threads: 4 is CPU parallelism, derived from the core count
• a 512×512 matrix multiply takes 2.11 ms

With numpy present, tensor conversion works too:

~/torch-venv/bin/python -c "import torch;print(torch.tensor([1.0,2.0]).numpy())"

[1. 2.]

Disk cost

du -sm ~/torch-venv

906

The whole venv is 906 MB (the 209.5 MB wheel expands to roughly 900 MB), far below the 5 GiB of the repo route.

Choosing between them

	Fedora python3-torch	pip CPU wheel
download	2 GiB / 747 packages	209.5 MB / 37 s
disk	5 GiB	906 MB (venv)
version	2.12.0	2.15.0+cpu
isolation	installed into system Python	venv, delete the directory to uninstall
good for	system-wide management	almost everything

Tip

A venv is also clean to remove: rm -rf ~/torch-venv leaves nothing behind, unlike 900 MB buried in system Python.

Common traps

Symptom	Cause
No module named pip	Fedora does not ship pip; dnf install python3-pip
dependencies hang on sympy	the primary index points at pytorch.org, where small packages crawl
Failed to initialize NumPy warning	numpy absent; .numpy() unavailable
cuda available: False	CPU wheel on a machine without a GPU — expected
disk dropped by 5 GiB	you took the dnf route
need the GPU build	switch to a CUDA index such as --index-url https://download.pytorch.org/whl/cu126

Related pages

• Install Python — interpreters and venv basics
• Ollama — another local-AI route
• Domestic mirrors — package repos and mirrors
• Official: PyTorch install page