Following installation guide, unable to start pie

[stevel408]

I followed the installation guide to the "verify installation" section (https://pie-project.org/docs/installation/#verify-installation) and encountered the following error.

$ pie run text-completion -- --prompt "Hello, world"

╭─ Pie Run ───────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────╮
│ Inferlet       text-completion                                                                                                                                                          │
│ Model          Qwen/Qwen3-0.6B                                                                                                                                                          │
│ Device         mps                                                                                                                                                                      │
╰─────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────╯

Starting engine...
 Loading weights │████████████████████████████████████████│ 100% • 226/226 • 423.02tensors/s • ETA: 00:00
✓ Engine running. Press Ctrl+C to stop

✗ Error: Failed to launch instance from registry: Registry returned error 404 Not Found for std/text-completion @ latest: {"detail":"Not Found"}

This is done on a MacOS with a M3 processor with the latest OS version.

Thank you guys for the good work.

[shsym]

Hi @stevel408, thanks for reporting this.

@ingim, do we plan to publish the updated package to PyPI, since this issue was fixed in d00710b?

@zyma98, it seems the root issue is that the registry is maintaining compatibility with the upstream main branch rather than with PyPI.

[zyma98]

Yes, the registry is maintaining compatibility with main. The PyPI release is very old. There are lots of breaking changes going on so for now I'd recommend installing from source.

In addition, the registry maintained text-completion is also stale. It doesn't work with the latest engine anymore. For this, I'd recommend submitting from local, not from registry.

Please follow the commands below to build from source and submit locally @stevel408 :

Prepare the environment

git clone https://github.com/pie-project/pie.git # Cloned directory referenced as $PIE below
cd $PIE/pie
uv sync --extra metal

Build the example

cd $PIE/std/text-completion
cargo build --release --target wasm32-wasip2

Run the example

cd $PIE/pie
uv run \
    pie run \
    --path $PIE/std/text-completion/target/wasm32-wasip2/release/text_completion.wasm \
    --manifest $PIE/std/text-completion/Pie.toml \
    -- \
    --prompt "Where is capital of France?"

[stevel408]

Thank you guys for such quick responses. I'm not in a hurry as I'm providing such feedback on suggestion from your prof., and would be happy to try it out again after you guy flush everything and merge into the main branch, and/or hopefully also update the web site itself.

Kudos...

[YangLiuWillow]

Yes, the registry is maintaining compatibility with main. The PyPI release is very old. There are lots of breaking changes going on so for now I'd recommend installing from source.

In addition, the registry maintained text-completion is also stale. It doesn't work with the latest engine anymore. For this, I'd recommend submitting from local, not from registry.

Please follow the commands below to build from source and submit locally @stevel408 :

Prepare the environment

git clone https://github.com/pie-project/pie.git # Cloned directory referenced as $PIE below
cd $PIE/pie
uv sync --extra metal

Build the example

cd $PIE/std/text-completion
cargo build --release --target wasm32-wasip2

Run the example

cd $PIE/pie
uv run \
    pie run \
    --path $PIE/std/text-completion/target/wasm32-wasip2/release/text_completion.wasm \
    --manifest $PIE/std/text-completion/Pie.toml \
    -- \
    --prompt "Where is capital of France?"

Thank you so much! This fixes my issue during installation. However, if I were just to follow the installation guide here (https://pie-project.org/docs/installation/#verify-installation), I face a linker error, flagging it here just so that it gets taken care of when you guys get things flushed and merge to the main branch.

$ pie run text-completion -- --prompt "What is the capital of France?"

╭─ Pie Run ────────────────────────────────────────────────────────────────────────────────────────────────╮
│ Inferlet       text-completion                                                                           │
│ Model          Qwen/Qwen3-0.6B                                                                           │
│ Device         mps                                                                                       │
╰──────────────────────────────────────────────────────────────────────────────────────────────────────────╯

Starting engine...
 Loading weights │██████████▍                             │  26% • 59/226 • 589.40tensors/s • ETA: 00:00W0407 13:59:52.734000 57646 torch/distributed/elastic/multiprocessing/redirects.py:29] NOTE: Redirects are currently not supported in Windows or MacOs.
 Loading weights │████████████████████████████████████████│ 100% • 226/226 • 279.40tensors/s • ETA: 00:00
✓ Engine running. Press Ctrl+C to stop

❌ Instance exception: Runtime error: Instantiation error: component imports instance `pie:core/model`, but a matching implementation was not found in the linker

✓ Complete
/opt/anaconda3/lib/python3.11/multiprocessing/resource_tracker.py:254: UserWarning: resource_tracker: There appear to be 1 leaked semaphore objects to clean up at shutdown
  warnings.warn('resource_tracker: There appear to be %d '

[shsym]

Thanks, @YangLiuWillow, for reporting this. This usually happens due to WASI interface changes since the last PyPI release. Please let us know if this happens, even if everything is built from the main branch.

[stevel408]

It seems that you guys modified the instructions a bit, so today I tried again from scratch, and let me share with you guys what I got.

First I followed your build instructions and it was quite smooth. Then I tried to run with your instructions and hit a little snag:

$ pie config init
zsh: command not found: pie

It seems that the instruction says to use the "uv" package manager, but didn't tell me how to get into the execution environment. In any case, I think I figured it out by prefixing everything with "uv run".

$ uv run pie config init
      Built pie-server @ file:///Users/...xxx.../workspace/...xxx.../pie
Uninstalled 1 package in 0.86ms
Installed 1 package in 2ms
✓ Configuration file created at /Users/...xxx.../.pie/config.toml
Downloading Python 3.14 runtime for Python inferlets ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 23.1/23.1 MB 75.8 MB/s
Extracting runtime…
✓ Python 3.14 runtime installed

Model downloading was uneventful as well:

$ uv run pie model download Qwen/Qwen3-0.6B

Downloading: Qwen/Qwen3-0.6B
 ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━  0/? bytes ?
✓ Downloaded to /Users/...xxx.../.cache/huggingface/hub/models--Qwen--Qwen3-0.6B/snapshots/c1899de289a04d12100db370d81485cdf75e47ca

✓ Pie compatible (arch: qwen3)
Add to config.toml:
  hf_repo = "Qwen/Qwen3-0.6B"

But then running the app still isn't successful:

$ uv run pie run text-completion -- --prompt "What is the capital of France?"

╭─ Pie Run ────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────╮
│ Inferlet       text-completion                                                                                                                                                       │
│ Model          Qwen/Qwen3-0.6B                                                                                                                                                       │
│ Device         mps                                                                                                                                                                   │
╰──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────╯

Starting engine...
 Loading weights │▏                                       │   0% • 1/226 •  5.76tensors/s • ETA: 00:39W0428 21:04:08.675000 58382 torch/distributed/elastic/multiprocessing/redirects.py:29] NOTE: Redirects are currently not supported in Windows or MacOs.
 Loading weights │████████████████████████████████████████│ 100% • 226/226 • 136.41tensors/s • ETA: 00:00
✓ Engine running. Press Ctrl+C to stop

❌ Instance exception: Runtime error: Instantiation error: component imports instance `pie:core/model`, but a matching implementation was not found in the linker

✓ Complete
/opt/homebrew/Cellar/python@3.12/3.12.13/Frameworks/Python.framework/Versions/3.12/lib/python3.12/multiprocessing/resource_tracker.py:279: UserWarning: resource_tracker: There appear to be 1 leaked semaphore objects to clean up at shutdown
  warnings.warn('resource_tracker: There appear to be %d '

[shsym]

Hi @stevel408, thank you for reporting this.

1. Yes, uv is required if you install pie from the source. We indeed have a note on the installation page for this ;)

2. The linker issue is the same problem that occurs when it tries to load Inferlets from the old registry. You can try the command @zyma98 mentioned above.

Build the example

cd $PIE/std/text-completion
cargo build --release --target wasm32-wasip2

Run the example

cd $PIE/pie
uv run \
    pie run \
    --path $PIE/std/text-completion/target/wasm32-wasip2/release/text_completion.wasm \
    --manifest $PIE/std/text-completion/Pie.toml \
    -- \
    --prompt "Where is capital of France?"

We have a new release coming out in a week, which will fundamentally solve these issues.