gpumesh

Borrow your friends' GPUs. A distributed compute mesh that lets you share GPU power across machines on your network — with one decorator, one CLI command, or a Python API.

  ╔═══════════════════════════════════════════════════════════╗
  ║              gpumesh - GPU Mesh Network                   ║
  ║        "like Bluetooth, but for your GPUs"                ║
  ╚═══════════════════════════════════════════════════════════╝

         ┌─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─┐
         │           NETWORK TRAFFIC FLOW            │
         │                                           │
         │   ┌──────────┐     ┌──────────┐           │
         │   │ RTX 4090 │◄───►│ RTX 3080 │           │
         │   │  Server  │     │  Laptop  │           │
         │   │120.5 G/s │     │ 85.2 G/s │           │
         │   └────┬─────┘     └────┬─────┘           │
         │        │                │                 │
         │   ┌────▼────────────────▼────┐            │
         │   │        T4 (12.0)         │            │
         │   │      running tasks       │            │
         │   └──────────────────────────┘            │
         │                                           │
         │   >>> results collected automatically     │
         └─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─┘

[!CAUTION] gpumesh runs code you send it, on machines that trust you. There is no sandbox.

A worker executes arbitrary Python as the OS user that started it — same files, same GPUs, same network, same credentials. A valid token is not a password guarding data; it is a licence to execute code on every machine in the mesh.

It runs both ways: a worker's results are deserialized by whoever submitted the task, so a hostile worker executes code on the submitter. Trust runs in both directions.

gpumesh provides no sandbox and does not try to. Use it on machines and with people you actually trust. Read SECURITY.md and THREAT_MODEL.md before exposing a coordinator to a network.

What is gpumesh?

gpumesh turns multiple machines into a single, unified compute pool. Start a coordinator on one machine, join workers from other machines (laptops, desktops, servers — anything with Python), and run code across all of them as if they were one device.

Use cases:

• Hyperparameter search across multiple GPUs
• Data preprocessing sharded across machines
• Model training on a pool of consumer GPUs
• Any embarrassingly parallel workload

Key features

Quick demo

from gpumesh import GPUMesh, accelerate

mesh = GPUMesh("http://coordinator:8000", token=TOKEN)

@accelerate(mesh)
def train(lr, epochs):
    return {"accuracy": 0.95}

result = train(lr=0.01, epochs=100)         # one mesh worker (or local if none)

results = train.map([                        # spread across all mesh devices
    {"lr": 0.01, "epochs": 100},
    {"lr": 0.05, "epochs": 200},
])

Installation

pip install gpumesh
pip install gpumesh[gpu]       # GPU detection + CUDA benchmarks
pip install gpumesh[tunnel]    # ngrok for public URLs
pip install gpumesh[sysinfo]   # System info (psutil)
pip install gpumesh[notebook]  # DataFrame support (pandas)
pip install gpumesh[ui]        # Setup wizard (rich + questionary)
pip install gpumesh[all]       # Everything above

Or with Docker:

docker pull samurai007ak/gpumesh:latest

Requires: Python 3.9+, cloudpickle (auto-installed). PyTorch is optional (needed for GPU detection).

Verify the install:

$ gpumesh --version
gpumesh 1.3.0 (3.11.9, Windows)

Quickstart

One machine is enough. gpumesh serve joins your own CPU/GPU to the pool by default, so a mesh of one runs exactly the same code path as a mesh of ten — same scheduler, same wire format, same results. Get that working first; adding a second machine is one flag and one command.

All terminal output below is real, copied from actual runs, with addresses and hostnames replaced by placeholders.

1. Generate a token

python -c "import secrets; print(secrets.token_urlsafe(32))"

Do this once and reuse the value. It is not decoration: the token is the only thing standing between anyone who can reach the port and code execution as you. Short, memorable tokens are guessable, and the coordinator's hash is a single SHA-256 round with no KDF — the entropy of the token is the whole defence.

2. Start a coordinator

pip install gpumesh
gpumesh serve --port 8000 --token $TOKEN

[gpumesh] config saved to ~/.gpumesh/config.json
[OK] Coordinator listening on 127.0.0.1:8000
   Token: <your token>

   Join from THIS machine:   gpumesh join http://127.0.0.1:8000 --token <your token>
   Other machines CANNOT reach this coordinator (bound to 127.0.0.1 only).
   That is the default now: a worker executes code as the user who started it.
   To let other machines join, re-run with: gpumesh serve --host 0.0.0.0 --port 8000 --token <your token>
   (or set GPUMESH_HOST=0.0.0.0). Read the warning it prints before you do.
   Ctrl+C to stop
[OK] Self-check: server is up (tested on 127.0.0.1 only)
[OK] Self-worker started — this machine's CPU/GPU is part of the pool
   (disable with --no-self-worker)
[worker] device=CPU Intel64 Family 6 Model 154 Stepping 3, GenuineIntel score=0.373 GFLOP/s
[mesh] worker joined: laptop-a (Intel64 Family 6 Model 154 Stepping 3, GenuineIntel, score=0.373)
[worker] joined mesh as 0afecb2dfd44

You now have a working mesh. Confirm it in another terminal:

$ gpumesh workers

  WORKERS
  ------------------------------------------------------
  CPU  0afecb2d    laptop-a          score=0.373     [alive]

Prefer a guided wizard? Run gpumesh setup.

3. Run a distributed task

Save as demo.py:

from gpumesh import mesh          # auto-connects using the saved config

@mesh
def train(lr, epochs):
    return {"lr": lr, "accuracy": round(0.90 + lr, 3)}

print(train(lr=0.01, epochs=100))                              # one worker
print(train.map([{"lr": 0.01, "epochs": 100},                  # every worker
                 {"lr": 0.05, "epochs": 200}]))

$ python demo.py
[gpumesh] connected to coordinator at http://127.0.0.1:8000
{'lr': 0.01, 'accuracy': 0.91}
[{'lr': 0.01, 'accuracy': 0.91}, {'lr': 0.05, 'accuracy': 0.95}]

That is the whole workflow. Kill the coordinator and run it again — you get the identical output, computed locally. Your script does not break because the mesh went away.

More runnable examples, all verified: examples/.

4. Add a second machine

The coordinator binds to loopback, so nothing outside your machine can reach it. That is the default, and opening it up is a decision rather than an oversight — a worker runs whatever code the coordinator hands it, as the OS user who started that worker. Restart with --host 0.0.0.0 when you mean it:

gpumesh serve --host 0.0.0.0 --port 8000 --token $TOKEN

   Join from THIS machine:   gpumesh join http://127.0.0.1:8000 --token <your token>
   Join from ANOTHER machine: gpumesh join http://192.0.2.10:8000 --token <your token>
   (127.0.0.1 always works locally; the LAN IP is for other machines)

  !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
  !! NETWORK-EXPOSED COORDINATOR
  !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
   Bound to:      0.0.0.0:8000 (not loopback — other machines can connect)
   Tasks run as:  alex
   On device:     CPU Intel64 Family 6 Model 154 Stepping 3, GenuineIntel

   Anyone who can reach this port AND has the token can run
   arbitrary code on this machine as alex.
   Only do this on a network you trust, and keep the token secret.
   Loopback-only (the default) is: gpumesh serve --port 8000
  !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!

   Reachability of http://192.0.2.10:8000 from other machines is not tested here — a worker joining confirms it.

Copy the Join from ANOTHER machine line — that is the exact command the second machine needs.

# on machine B
pip install gpumesh
gpumesh join http://192.0.2.10:8000 --token $TOKEN

[worker] device=CPU Intel64 Family 6 Model 154 Stepping 3 score=0.310 GFLOP/s
[worker] joined mesh as 7c1e04ab93f2

$ gpumesh workers

  WORKERS
  ------------------------------------------------------
  CPU  0afecb2d    laptop-a          score=0.373     [alive]
  CPU  7c1e04ab    laptop-b          score=0.310     [alive]

Workers never die. They survive laptop sleep, WiFi drops and coordinator restarts, and reconnect on their own.

Windows: run gpumesh serve as Administrator so the firewall rule is added for you. Without it, workers on other machines may be blocked.

If the worker reports a timeout instead, it could not reach that address — see Troubleshooting.

examples/second_machine.py reads your live mesh and prints the exact command, filled in with your address, port and token.

Prefer the CLI?

Submit a script instead of decorating a function. (The examples/ files ship with the repository, not the pip package — git clone first, or point the command at any script of your own that reads a JSON payload on stdin.)

$ gpumesh submit examples/grid_search.py --payloads examples/payloads.json --wait
[OK] Submitted job c3f81cc498f0
Job: examples/grid_search.py (c3f81cc498f0)

  #################### 100%  6/6 done  (10s)

  Status: finished
  Counts: {'done': 6}

  ✓ task 463a58e92490 [done]  cost=1.0  worker=0afecb2dfd44
    result: {"lr": 0.01, "epochs": 100, "l2": 0.0, "val_accuracy": 0.948, "weights": [0.2271, -0.2757, 0.0019]}
  ✓ task 7fa79a95725f [done]  cost=10.0  worker=0afecb2dfd44
    result: {"lr": 0.3, "epochs": 1000, "l2": 0.01, "val_accuracy": 0.952, "weights": [2.1907, -2.9448, 0.0191]}

Code normally

This is the entire point. Once a worker is connected, every machine sees the same pool. Write normal Python and mark the heavy functions:

from gpumesh import mesh   # auto-connects from saved config

@mesh
def train(lr, epochs):
    return {"accuracy": 0.95}

# Single call — runs on a mesh worker (your own machine if nothing else joined)
result = train(lr=0.01, epochs=100)

# .map() — spreads across EVERY connected laptop + your machine
results = train.map([{"lr": 0.01}, {"lr": 0.05}, {"lr": 0.1}])

Works in VS Code, Jupyter, PyCharm, or a plain terminal. No job submission, no CLI commands, no ceremony.

Your function returns whatever it normally returns — a dict, an int, a list, a numpy array, a torch tensor — and you get that same object back:

@mesh
def evaluate(seed):
    import numpy as np
    return {"loss": np.float32(0.12), "preds": np.arange(10)}

out = evaluate(seed=1)      # {'loss': np.float32(0.12), 'preds': array([0, ..., 9])}

Jupyter notebooks

Load the extension once, in its own cell:

%load_ext gpumesh

Then %%mesh as the first line of any cell wraps every function defined in that cell with @mesh:

%%mesh
def preprocess(chunk_id, rows):
    return {"chunk": chunk_id, "rows": rows * rows}

results = preprocess.map([{"chunk_id": i, "rows": 100 + i} for i in range(6)])

Like %%time, the cell's own output displays normally. Loading the extension also injects a bare @mesh decorator into the notebook namespace, so you can decorate individual functions instead of a whole cell.

CLI reference

Two flags are global and go before the subcommand: -v / --verbose for debug-level logging, and --json-logs to emit logs as JSON lines (for Docker).

Server & connection

serve and join also take --color / --no-color. They are the two commands that keep printing for hours into a Docker log or a systemd journal, which is exactly where isatty() says "no terminal" and you may still want colour — or emphatically not want it.

gpumesh doctor is read-only: it adds no firewall rules, saves no connection, installs nothing, and never prints the token. It reports which gpumesh is imported and from where, the Python interpreter, torch/CUDA (and whether nvidia-smi disagrees with it), the cloudpickle version, the saved coordinator and its workers, the addresses this machine would bind and advertise, and on Windows whether a firewall rule for the port exists. It exits 1 only for a fault on this machine — no coordinator configured, or an unreachable one, is a warning and still exits 0, so it works as a pre-flight check in a script. --json emits the same report as a document (warnings are diverted to stderr so the stdout stays parseable).

--host and --host-ip are different things, and mixing them up costs an hour of firewall debugging:

Setting --host-ip alone never opens the port up. Use --host 0.0.0.0 for that, and read the banner it prints. --host-ip is for when auto-detection picks a VPN or hypervisor adapter and remote workers time out against an address only your machine can route to.

Jobs

Each payload object is handed to your script as JSON on stdin. Four keys are read by the scheduler instead of being left to your code:

These are filters, not preferences. A task nobody can satisfy stays pending until the coordinator gives up on it (60s) and fails it with a message naming the requirement, rather than queueing forever.

All four also apply to payloads passed to mesh.submit(...) from the Python API, and to the payload dicts you hand to .map(). Full shapes in docs/protocol.md; what is versioned and what a bump promises is in docs/stability.md.

Monitoring

Job and monitoring commands (submit, status, cancel, retry, workers, devices, kill) accept --url URL --token TOKEN. If you omit them, gpumesh falls back to the GPUMESH_URL / GPUMESH_TOKEN environment variables, then to the connection saved by join/serve.

Environment variables

GPUMESH_CLAIM_HOST is deliberately not GPUMESH_HOST. The two answer different questions and are routinely set on the same machine: GPUMESH_HOST=127.0.0.1 is a perfectly sensible coordinator setting, and if the claim port read it too, every worker on that box would silently become unclaimable. A non-loopback claim bind prints a banner naming the OS user that claims will run as.

GPUMESH_COLOR is the environment form. gpumesh serve and gpumesh join also take --color (force it on) and --no-color (force it off), which are mutually exclusive. The flags work by setting GPUMESH_COLOR and re-running the detection, so the choice reaches this process and the function-task subprocess a worker spawns, which inherits the environment. (Script tasks run with a deliberately minimal allowlisted environment and never see it.)

One wrinkle: the auto-detection probes stdout, while -v log lines are written to stderr. For the uncommon shape where only stderr is redirected (gpumesh serve 2> run.log from a terminal), the detection says "terminal" and the log file gets escape sequences. --no-color or GPUMESH_COLOR=0 is the escape hatch.

Python API

from gpumesh import GPUMesh

mesh = GPUMesh("http://coordinator:8000", token=TOKEN)

Distribute a function

results = mesh.distribute(
    function=train_model,
    params=[{"lr": 0.01, "epochs": 100}, {"lr": 0.05, "epochs": 200}],
    timeout=600,
)

Inspect the pool

workers = mesh.workers()        # [{'id', 'device', 'device_name', 'hostname', 'score', 'alive'}]
devices = mesh.devices()        # unified pool view
count   = mesh.device_count()   # alive machines contributing compute (GPU or CPU)
gpus    = mesh.gpu_count()      # alive GPUs only
total   = mesh.total_score()    # combined compute score
best    = mesh.auto_device()    # most powerful alive device

Job management

job_id = mesh.submit(name="preprocess", script="process.py",
                     payloads=[{"file": "data.csv"}])
status = mesh.status(job_id)
df     = mesh.results_to_dataframe(results)   # requires pandas

From Python, non-blocking

GPUMesh.start_coordinator(port=8000, token=TOKEN)   # binds 127.0.0.1; pass host="0.0.0.0" to open it up
GPUMesh.add_worker("http://coordinator:8000", token=TOKEN)

@accelerate patterns

from gpumesh import GPUMesh, accelerate

mesh = GPUMesh("http://coordinator:8000", token=TOKEN)

# Basic
@accelerate(mesh)
def preprocess(chunk_id, data_path):
    import pandas as pd
    df = pd.read_parquet(data_path)
    return {"chunk": chunk_id, "rows": len(df)}

# Hardware selection — only run on an A100
@accelerate(mesh, gpu="A100")
def train(model):
    return model.cuda().forward(x)

# Resource specs
@accelerate(mesh, cores=8, memory="16GB", timeout=300)
def heavy_computation(data):
    return processed

# Batch: spread across every device
results = train.map([{"lr": 0.01}, {"lr": 0.05}])

# Bind to a specific device
gpu_predict = predict.to("cuda")
result = gpu_predict(x)

# Global install — @accelerate with no arguments
accelerate.install(mesh)

@accelerate
def train(lr, epochs):
    return {"accuracy": 0.95}

Smart routing

Either path returns the identical value, so switching between them never changes your results. Note that gpumesh serve joins your own machine to the pool by default, so a single call is dispatched through the mesh even when you are the only participant — pass --no-self-worker if you want it to stay purely local.

Docker

A prebuilt image is available on Docker Hub (samurai007ak/gpumesh).

First, a token:

export GPUMESH_TOKEN=$(python -c "import secrets; print(secrets.token_urlsafe(32))")

There are two host decisions in a container, and they are not the same one:

1. What the coordinator binds to inside the container. It must be 0.0.0.0, because anything reaching it arrives from outside the container — a loopback bind inside a container answers nothing, including the port Docker published.
2. What Docker publishes on the machine running Docker. This is the one that decides who can reach you. -p 8732:8732 binds 0.0.0.0 on the host, which puts a coordinator on every network the machine is attached to. Prefix the host side with 127.0.0.1 unless you mean otherwise.

# Coordinator — reachable from this machine only
docker run -d --name gpumesh-coordinator \
  -p 127.0.0.1:8732:8732 -p 127.0.0.1:48900:48900/udp \
  -e GPUMESH_TOKEN \
  samurai007ak/gpumesh:1.3.0 \
  serve --host 0.0.0.0 --port 8732

# Worker
docker run -d --name gpumesh-worker \
  -e GPUMESH_TOKEN \
  samurai007ak/gpumesh:1.3.0 \
  join http://coordinator-ip:8732

Drop the 127.0.0.1: prefix — or better, name a specific interface, e.g. -p 192.0.2.10:8732:8732 — only when you deliberately want other machines to join. That is the same decision gpumesh serve --host 0.0.0.0 asks you to make on the host, and it deserves the same pause.

Both serve and join read GPUMESH_TOKEN from the environment, so -e GPUMESH_TOKEN is enough and the token never appears in docker ps output. join takes the coordinator URL as a positional argument, so there is no GPUMESH_URL to set here — that variable is only read by the job and monitoring commands.

Or use the included docker-compose.yaml for a coordinator + N workers, with a healthcheck the workers wait on before starting. It pulls the published image, so the file works on its own:

GPUMESH_TOKEN=$TOKEN docker compose up -d                      # localhost only
WORKER_REPLICAS=4 GPUMESH_TOKEN=$TOKEN docker compose up -d    # scale workers
GPUMESH_BIND=0.0.0.0 GPUMESH_TOKEN=$TOKEN docker compose up -d # reachable from the LAN

GPUMESH_TOKEN is required — compose stops with a clear message if it is unset, rather than starting a coordinator with a random token that no worker could authenticate against. GPUMESH_BIND defaults to 127.0.0.1 and is the host-side publish address from decision (2) above.

Ports: 8732 (TCP API) and 48900/udp (LAN discovery). Persistent state (the SQLite database and ~/.gpumesh/config.json) lives under /data; the compose file mounts a named volume there.

The container listens on 8732, while gpumesh serve on the host defaults to 8000. That is deliberate — the image pins an explicit port so published docker run and compose recipes stay stable. Both are just defaults: pass --port to use whatever you like, and make sure workers point at the same number the coordinator is listening on.

Network options

• LAN: gpumesh serve --host 0.0.0.0 + gpumesh join http://192.0.2.10:8000 --token $TOKEN. UDP broadcast helps workers find a coordinator, but joining still needs the token and a coordinator that is actually bound beyond loopback.
• Tailscale: gpumesh serve --host 0.0.0.0 --port 8000 --tailscale, then join via the Tailscale IP. This is the option to reach for if the mesh crosses any network you do not control — the tunnel, not the token, becomes the boundary.
• ngrok: gpumesh serve --host 0.0.0.0 --port 8000 --public prints a public https://... URL that workers anywhere can join. "Anywhere" includes people you did not invite; the token is the only thing keeping them out.

The default — no flag at all — is loopback only. Every row above is an opt-in to a wider audience.

Architecture

                         COORDINATOR
        ┌─────────────────────────────────────────────────┐
        │                                                 │
        │  ┌──────────┐  ┌──────────┐  ┌──────────────┐  │
        │  │ Job Queue │  │ Task DB  │  │ Worker       │  │
        │  │ (memory)  │  │ (SQLite) │  │ Registry     │  │
        │  └────┬─────┘  └──────────┘  └──────┬───────┘  │
        │       │                              │          │
        │       └──────────┬───────────────────┘          │
        │                  │                              │
        │         HTTP API :8000                          │
        └──────────────────┼──────────────────────────────┘
                           │
              ┌────────────┼────────────┐
              │            │            │
        ┌─────▼────┐ ┌────▼────┐ ┌────▼────┐
        │ Worker 1 │ │Worker 2 │ │Worker 3 │
        │ RTX 4090 │ │RTX 3080 │ │   T4    │
        │Score: 120│ │Score: 85│ │Score: 12│
        └──────────┘ └─────────┘ └─────────┘
              │            │            │
              └────────────┼────────────┘
                           │
                    ┌──────▼──────┐
                    │   Results   │
                    │  Collected  │
                    └─────────────┘

  JOB FLOW:  Submit ─► Queue ─► Claim ─► Execute ─► Report ─► Collect

How it works: jobs are stored in SQLite, workers pull tasks over HTTP with a lease (so a crashed worker's task is automatically re-queued), run each task in an isolated subprocess, and post results back. Workers are scored by a benchmark and the scheduler assigns heavier tasks to stronger workers.

Security

Read that table as "what raises the cost of an attack", not as "what makes this safe". None of it is a sandbox.

The token is a licence to execute code, not a password guarding data. Anyone holding your URL and token can run arbitrary Python on every machine in your mesh, as the user who started each worker. And it runs both ways: a worker's result is deserialized on the submitting machine, so a hostile worker executes code on you. gpumesh is built for trusted networks — home labs, lab benches, your own machines, a team you know.

Use a real token: python -c "import secrets; print(secrets.token_urlsafe(32))". The coordinator hashes it with a single SHA-256 round and a deterministic salt, which is fine for 250 bits of entropy and useless for a token you chose by hand. Entropy is the whole defence.

Run the coordinator with --safe-mode to refuse function distribution and accept submitted scripts only. That closes the pickle path; it does not stop a submitted script from doing anything a script can do.

Rate limiting exempts loopback deliberately. Anyone who can open a socket from 127.0.0.1 can read the token out of the process and the config file rather than guess it, so a lockout there costs an attacker nothing and costs you your own mesh.

Traffic is not encrypted on a plain LAN. Use --tailscale or --public (ngrok) when the mesh crosses a network you do not control.

Full detail: SECURITY.md (what a token grants, how to report a vulnerability) and THREAT_MODEL.md (the same flow traced through the code, file and line).

Benchmark scoring

Each worker runs a benchmark on join and gets a score of gflops * 0.7 + bandwidth_gbps * 0.3. It is a relative, unbounded number — there is no normalisation and no ceiling, so it only means anything next to the other workers in your pool. The scheduler uses it to rank workers, nothing else.

Rough magnitudes, to read the numbers gpumesh workers prints:

Those are illustrative, not a scale — a laptop CPU commonly lands around 0.24, and a faster GPU than anything listed here would simply score higher.

Troubleshooting

Start with gpumesh doctor. It is read-only, it never prints the token, and its output is meant to be pasted into an issue. For "workers can't connect" it shows in one screen whether the coordinator binds loopback, which address this machine would advertise (and whether that address belongs to a virtual adapter no other machine can route to), whether a Windows firewall rule for the port exists, and whether the saved coordinator answers at all. For "NameError on a task" it prints this machine's Python, gpumesh and cloudpickle versions and flags any connected worker reported as differing — which is the cause nearly every time. If the coordinator does not report worker versions it says so and tells you to run gpumesh doctor on each machine and compare the Python lines by hand. gpumesh doctor --json gives the same report as a parseable document.

Verbose logging: gpumesh -v serve (the flag is global, so it works on any command) — decorator routing messages: GPUMESH_VERBOSE=1 python my_script.py — force local-only: GPUMESH_LOCAL=1 python my_script.py

Development

git clone https://github.com/Samurai007AK/gpumesh.git
cd gpumesh
pip install -e ".[dev]"
pytest                 # on Windows a few skip and one xpasses
python -m build        # build wheel + sdist

CI runs the suite on Linux (Python 3.9, 3.11, 3.12), Windows and macOS for every push and pull request — the badge at the top of this page is the live count, so it cannot drift the way a hand-written number does.

Contributions are welcome — issues and pull requests both. See CONTRIBUTING.md for setup, how the pieces fit together, and what makes a change easy to review.

Documentation

Limitations

• Python only — tasks must be Python functions or scripts
• Arguments and return values must be picklable. Anything tied to a live process — open files, sockets, locks, database handles, CUDA handles — cannot cross machines. Return plain data (numbers, arrays, tensors, DataFrames) instead
• Every worker needs the imports your function uses already installed; gpumesh ships your code, not your environment
• Workers should run the same Python minor version as the submitter — cloudpickle falls back to source when they differ, which does not cover every function
• No GPU memory sharing — each task gets its own process
• No model sharding — each task runs on one machine at a time
• No task graphs — every task is independent; if task B needs task A's output, that plumbing is yours
• No distributed data structures — no shared arrays, dataframes or object store. If your data does not fit on one machine, gpumesh will not help
• Built and tested for meshes of roughly 2–20 machines. Nobody has run it at hundreds of workers
• Single coordinator — single point of failure (use Tailscale for reliability)
• No built-in encryption — use Tailscale for encrypted tunnels
• Trusted networks only — workers run whatever code the coordinator sends, and submitters deserialize whatever workers return

If several of those are dealbreakers, read Why not Ray or Dask? — it says plainly which tool to use instead.

Prior art and credits

gpumesh borrows shamelessly, at the level of API shape and scheduling strategy. No code was copied from any of these projects; what was taken is the idea of what a good interface looks like, and each one is credited in the source where its influence lands.

exo (Apache-2.0) is the closest relative in spirit — running AI workloads on the consumer hardware you already own. Its runner-supervisor pattern is why every gpumesh task runs in its own subprocess, and its crash diagnostics and topology-change events are reflected in worker.py and db.py. hivemind (MIT) supplied the TTL-based worker expiry that prunes a machine that has stopped heartbeating. Petals (MIT) supplied straggler deprioritisation — a worker slower than twice its peers' median gets lighter tasks rather than holding up the batch. cudf.pandas (Apache-2.0) is where accelerate.install(mesh) comes from: an import hook that makes existing code use the accelerator without editing it. Hugging Face Accelerate (Apache-2.0) is the source of the .to(device) placement idea. clustrix (MIT) is where @accelerate(cores=8, memory="16GB") gets its shape — declaring resource requirements on the decorator rather than in a separate config. burla contributed the remote_parallel_map batch pattern behind .map() and the func_gpu="A100" hardware-selection idea; note that burla is licensed FSL-1.1-Apache-2.0, which is source-available, not OSI open source, and reads back to Apache-2.0 after two years. Two smaller influences round it out: distry, for the plain single-function decorator, and ezpz, for setup_torch()-style automatic backend detection.

If you maintain one of these and think the credit is wrong, or the influence is closer to copying than we believe, open an issue and it will be corrected.

License

MIT License. See LICENSE for details.

GitHub · Issues · PyPI