Compute Backends: CPU, GPU and Cloud

library(tidylearn)

Where a fit runs

Most of the time, nowhere interesting: tl_model() runs on the CPU and you never think about it. This vignette is about the cases where you should — a fit that would take hours, or one that will not fit in memory at all.

tidylearn separates two questions that are easy to conflate:

What this machine can do

tl_check_gpu()

tl_check_gpu() parses nvidia-smi and checks which GPU-capable backends are installed. It is cheap by design: it does not load Python, import TensorFlow, or fit anything. It returns an object with a print() method describing the CUDA driver, the devices found, and which of xgboost, keras, tensorflow and torch are installed.

Only two methods have a GPU path

Of the thirteen supervised methods, only "xgboost" and "deep" have an upstream GPU implementation. The other eleven — linear, polynomial and logistic regression, ridge, LASSO, elastic net, decision trees, random forests, gradient boosting, SVM and neural networks — wrap packages that have no CUDA path at all. Asking for a GPU gets you a warning and a CPU fit:

# Routed to CUDA when a capable backend is present
model <- tl_model(data, y ~ ., method = "xgboost", compute = "gpu")

# Warns and falls back to CPU: randomForest has no GPU implementation
model <- tl_model(data, y ~ ., method = "forest", compute = "gpu")

# Let tidylearn decide per call
model <- tl_model(data, y ~ ., method = "xgboost", compute = "auto")

For those eleven methods the question to ask is whether the job fits in memory, and how many cores you can throw at it.

Estimating before you commit

tl_compute_advisor() answers that. It is arithmetic over the problem dimensions, so it needs neither a GPU nor the backend package installed:

advice <- tl_compute_advisor(
  "xgboost", iris, Species ~ .,
  hyperparams = list(nrounds = 1000)
)

advice$recommendation
#> [1] "cpu"
advice
#> <tidylearn compute advice>
#> Problem:        xgboost on 150 rows x 4 cols (~0.0 MB)
#> 
#> Tier estimates (order-of-magnitude):
#>   Local CPU:    0.0s   (peak RAM ~0 MB, 16 cores)
#>   Local GPU:    --   [not applicable]
#>   Cloud:        45.0s   (~$0.01) [T4 (16 GB VRAM / 16 GB RAM)]   [not configured]
#> 
#> Recommendation: cpu
#> 
#> Reasoning:
#>   - Estimated local CPU runtime ~0.0s. Cloud cold-start (~45s) would dominate; just run it locally.
#> 
#> Notes:
#>   - Method 'xgboost' could use GPU, but no GPU-capable backend was detected. See ?tl_check_gpu.
#>   - Cloud integration is not yet configured in tidylearn. Estimates shown so users can see the tier's shape; actual submission is not yet supported.

The advisor covers all thirteen supervised methods, and treats cloud as a memory headroom tier. It will recommend cloud for a CPU-only method such as random forest or SVM when the job is RAM-infeasible locally, which is the common case for large data. On a small dataset like iris it will tell you to stay local.

Estimates are order-of-magnitude. Runtime constants are calibrated per method, and Modal pricing is approximate as of early 2026. Treat the output as a decision aid.

Unsupervised methods are not modelled. Calling the advisor on one errors.

Cloud compute

Cloud execution is not available yet. compute = "cloud" errors with a message saying so, and the advisor’s cloud tier is reported for planning only. What follows describes what is already in place, so that you can see the safety model before the feature that uses it.

A cloud fit uploads your training data to your own Modal account. That is a third party, and it may be something your organisation’s data-handling rules forbid. tidylearn treats that transfer as the consequential step it is.

Destinations are checked, not trusted

The endpoint is read from an environment variable:

Sys.setenv(
  TIDYLEARN_MODAL_ENDPOINT = "https://you--tidylearn-fit.modal.run"
)

An environment variable rather than an R option, deliberately: a shared .Rprofile can set an option without you noticing, and the whole point of this check is that a destination should not appear silently.

Before anything is sent, the URL must parse, use https, and resolve to a host on the allowlist. Anything else errors. The default allowlist is Modal’s own domains:

tl_cloud_allowed_hosts()
#> [1] "modal.run" "modal.com"

Modal customers serving Web Functions from a custom domain can add it, for the session:

tl_cloud_allow_host("fits.example.com")
#> Cloud uploads may now also go to: fits.example.com. This is in addition to Modal's own hosts and lasts for this R session only.
tl_cloud_allowed_hosts()
#> [1] "modal.run"        "modal.com"        "fits.example.com"

Matching is anchored, so allowing a host allows exactly it and its subdomains. fits.example.com does not admit example.com, evil-fits.example.com, or fits.example.com.evil.test. Single-label names such as "com" are refused outright, since they would open an entire top-level domain.

Bounding the cost

A job submitted to Modal runs there to completion whatever your R session does next. Pressing Ctrl-C, closing the IDE, a crashed session and a closed laptop all leave it running, because the session was only polling for a result. A job you thought you had abandoned keeps billing until it finishes or something kills it.

No amount of care in the R client can guarantee anything against that, because a killed session runs no cleanup handlers. So the controls are layered by what survives:

What holds even if your session dies. Every submission sets an explicit timeout, derived from the estimate with headroom and capped well below Modal’s 24-hour maximum. Modal’s inherited default is never used and neither is no timeout at all. The worker runs with retries disabled, because Modal applies timeouts per attempt and three retries would bill four full timeouts for one hung job. You should also set a spend budget on your Modal workspace, which is the only true hard cap and is not tidylearn’s to set.

What is best-effort. The pre-flight gates, the job registry and cancel-on-interrupt catch the ordinary cases, and none of them survive a killed session.

Before a fit, the figure you are asked to accept is the worst case — the timeout at the tier’s rate — rather than the estimate:

model <- tl_model(data, y ~ ., method = "xgboost", compute = "cloud",
                  confirm_upload = TRUE, max_cost = 5)

That refuses before anything is uploaded if the worst case exceeds max_cost. It also refuses a fit whose estimate is so large that the job would be killed by its own timeout before finishing — submitting that would bill the full timeout and return nothing.

Nothing in flight is invisible:

tl_cloud_jobs()
#> # A tibble: 0 × 5
#> # ℹ 5 variables: call_id <chr>, method <chr>, submitted_at <dttm>,
#> #   timeout_seconds <int>, worst_case_cost <dbl>

A job leaves that list when its result is collected or it is cancelled. The list itself does not outlive the session; the timeout does.

The contract is written down

What cloud compute will and will not do is documented as a threat model that ships with the package, including an audit checklist a reviewer can run against the source:

file.show(
  system.file("security/threat-model.md", package = "tidylearn")
)

It covers token handling, egress consent, ephemeral compute, the absence of telemetry, and destination validation.

What is coming

The remaining work is the submission path itself — building the authenticated request, uploading, polling for the result, and restoring the fitted model in your session. Models cross that boundary as bytes; twelve of the thirteen methods serialise natively, and "deep" carries its keras weights separately because a keras model is a reference to a Python object rather than an R one.

Until that lands, use tl_compute_advisor() to size the problem and run locally.

See also