---
title: "Deep and foundation models"
output: rmarkdown::html_vignette
vignette: >
  %\VignetteIndexEntry{Deep and foundation models}
  %\VignetteEngine{knitr::rmarkdown}
  %\VignetteEncoding{UTF-8}
resource_files:
  - figures/logo.svg
  - figures/logo-dark.svg
---

<!-- cttir-vignette-logo: start -->
<style>
.cttir-vignette-logo { float: right; width: 120px; max-width: 30%; margin: 0 0 1rem 1.5rem; }
.cttir-vignette-logo img { width: 100%; height: auto; border: 0; background: transparent; box-shadow: none; }
.cttir-vignette-logo .cttir-vignette-logo-light { display: block; }
.cttir-vignette-logo .cttir-vignette-logo-dark { display: none; }
@media (prefers-color-scheme: dark) {
  .cttir-vignette-logo .cttir-vignette-logo-light { display: none; }
  .cttir-vignette-logo .cttir-vignette-logo-dark { display: block; }
}
[data-bs-theme="light"] .cttir-vignette-logo .cttir-vignette-logo-light { display: block; }
[data-bs-theme="light"] .cttir-vignette-logo .cttir-vignette-logo-dark { display: none; }
[data-bs-theme="dark"] .cttir-vignette-logo .cttir-vignette-logo-light { display: none; }
[data-bs-theme="dark"] .cttir-vignette-logo .cttir-vignette-logo-dark { display: block; }
</style>
<div class="cttir-vignette-logo">
  <img class="cttir-vignette-logo-light" src="figures/logo.svg" alt="hypermodR logo" width="120" height="139" />
  <img class="cttir-vignette-logo-dark" src="figures/logo-dark.svg" alt="hypermodR logo" width="120" height="139" />
</div>
<!-- cttir-vignette-logo: end -->

```{r setup, include=FALSE}
library(hypermodR)
knitr::opts_chunk$set(collapse = TRUE, comment = "#>", error = FALSE)
```
Model declarations do not download weights or import upstream Python. This
example makes named classification specifications only; no pretrained inference
or scientific result is produced.

```{r named-contracts}
task <- hm_task("class", "classification", "image", "subject_id")
constructors <- list(
  hm_htc, hm_hyperfree, hm_hypersigma, hm_hypervision,
  hm_dual_stream_mhsi, hm_spectralearth, hm_spectralgpt
)
specifications <- lapply(constructors, function(f) f(task))
vapply(specifications, function(x) x$model_id, character(1))
```

HTC, HyperFree and HyperSIGMA have explicit extraction implementations requiring
exact reviewed source inventories, weights, preprocessing and rights. Their
independent installed-package parity remains pending. HyperSIGMA's fixed fused
features are distinct from its upstream trained downstream model. HyperVision,
Dual-Stream-MHSI, SpectralEarth and SpectralGPT retain explicit execution blocks
until their exact source/weights/input contracts are admitted. SpectralEarth is
an adapted spectral track; SpectralGPT is a lossy multispectral auxiliary track.
None is a qualified clinical model.

Package-owned scratch reference IDs use `reference:` followed by `cnn1d`,
`unet2d`, `cnn3d`, `spectral_vit`, `masked_autoencoder`, `contrastive_encoder`,
`vae`, `conditional_gan`, `ddpm`, `multitask`, `domain_adversarial` or
`distillation`. These have separate training objectives and synthetic gradient,
restore and prediction tests. Distillation uses a scratch frozen teacher.
Their worker protocol is distinct from the ordinary R engine workflow; registry
visibility alone does not admit an R `hm_fit()` neural route.

Reproduction is opt-in: resolve and review an exact runtime, use explicit training
roles/group IDs, bound epochs, retain numeric artifacts and run adapter-specific
parity. Do not run a model from a documentation build. GPU/Metal/ROCm/WSL,
PEFT, quantization and scientific transfer are not qualified by CPU reference
tests. See the runtime article before enabling any external execution.