diff --git a/DESCRIPTION b/DESCRIPTION
index 4f12e0e9..8fc0d3fc 100644
--- a/DESCRIPTION
+++ b/DESCRIPTION
@@ -62,6 +62,7 @@ Suggests:
     rsample,
     RSpectra,
     survival (>= 3.2-10),
+    tabnet,
     testthat (>= 3.0.0),
     TH.data,
     usethis (>= 1.5.0),
diff --git a/NAMESPACE b/NAMESPACE
index 993a1abc..aa746464 100644
--- a/NAMESPACE
+++ b/NAMESPACE
@@ -86,6 +86,7 @@ S3method(axe_env,terms)
 S3method(axe_env,train)
 S3method(axe_env,train.recipe)
 S3method(axe_env,xgb.Booster)
+S3method(axe_fitted,"_tabnet_fit")
 S3method(axe_fitted,C5.0)
 S3method(axe_fitted,default)
 S3method(axe_fitted,earth)
diff --git a/NEWS.md b/NEWS.md
index d15bfb6b..1bc61eff 100644
--- a/NEWS.md
+++ b/NEWS.md
@@ -1,5 +1,7 @@
 # butcher (development version)
 
+* Added butcher methods for `tabnet()` (@cregouby #226).
+
 # butcher 0.3.0
 
 * Julia Silge is now the maintainer (#230).
diff --git a/R/tabnet_fit.R b/R/tabnet_fit.R
new file mode 100644
index 00000000..f5f06c47
--- /dev/null
+++ b/R/tabnet_fit.R
@@ -0,0 +1,45 @@
+#' Axing a tabnet_fit.
+#'
+#' @inheritParams butcher
+#'
+#' @return Axed tabnet_fit object.
+#'
+#' @examplesIf rlang::is_installed("tabnet")
+#'
+#' # Load libraries
+#' suppressWarnings(suppressMessages(library(parsnip)))
+#' suppressWarnings(suppressMessages(library(rsample)))
+#'
+#' # Load data
+#' split <- initial_split(mtcars, props = 9/10)
+#' car_train <- training(split)
+#'
+#' # Create model and fit
+#' mtcar_fit <- tabnet() %>%
+#'   set_mode("regression") %>%
+#'   set_engine("torch")
+#'   fit(mpg ~ ., data = car_train)
+#'
+#' out <- butcher(mtcar_fit, verbose = TRUE)
+#'
+#' @name axe-tabnet_fit
+NULL
+
+#' Remove fitted values.
+#'
+#' @rdname axe-tabnet_fit
+#' @export
+axe_fitted._tabnet_fit <- function(x, verbose = FALSE, ...) {
+  old <- x
+  x$fit$fit <- exchange(x$fit$fit, "checkpoints", list(NULL))
+  x$fit$fit$importances <- exchange(x$fit$fit$importances, "variables", list(NULL))
+  x$fit$fit$importances <- exchange(x$fit$fit$importances, "importance", list(NULL))
+
+  add_butcher_attributes(
+    x,
+    old,
+    disabled = NULL,
+    add_class = FALSE,
+    verbose = verbose
+  )
+}
diff --git a/man/axe-tabnet_fit.Rd b/man/axe-tabnet_fit.Rd
new file mode 100644
index 00000000..d2a8b261
--- /dev/null
+++ b/man/axe-tabnet_fit.Rd
@@ -0,0 +1,46 @@
+% Generated by roxygen2: do not edit by hand
+% Please edit documentation in R/tabnet_fit.R
+\name{axe-tabnet_fit}
+\alias{axe-tabnet_fit}
+\alias{axe_fitted._tabnet_fit}
+\title{Axing a tabnet_fit.}
+\usage{
+\method{axe_fitted}{`_tabnet_fit`}(x, verbose = FALSE, ...)
+}
+\arguments{
+\item{x}{A model object.}
+
+\item{verbose}{Print information each time an axe method is executed.
+Notes how much memory is released and what functions are
+disabled. Default is \code{FALSE}.}
+
+\item{...}{Any additional arguments related to axing.}
+}
+\value{
+Axed tabnet_fit object.
+}
+\description{
+Axing a tabnet_fit.
+
+Remove fitted values.
+}
+\examples{
+\dontshow{if (rlang::is_installed("tabnet")) (if (getRversion() >= "3.4") withAutoprint else force)(\{ # examplesIf}
+
+# Load libraries
+suppressWarnings(suppressMessages(library(parsnip)))
+suppressWarnings(suppressMessages(library(rsample)))
+
+# Load data
+split <- initial_split(mtcars, props = 9/10)
+car_train <- training(split)
+
+# Create model and fit
+mtcar_fit <- tabnet() \%>\%
+  set_mode("regression") \%>\%
+  set_engine("torch")
+  fit(mpg ~ ., data = car_train)
+
+out <- butcher(mtcar_fit, verbose = TRUE)
+\dontshow{\}) # examplesIf}
+}
diff --git a/tests/testthat/test-tabnet_fit.R b/tests/testthat/test-tabnet_fit.R
new file mode 100644
index 00000000..c4de0a68
--- /dev/null
+++ b/tests/testthat/test-tabnet_fit.R
@@ -0,0 +1,41 @@
+test_that("tabnet_fit + axe_fitted() works", {
+  skip_on_cran()
+  skip_if_not_installed("tabnet")
+  suppressPackageStartupMessages(library(parsnip))
+  # Create model and fit
+  tabnet_fit <- tabnet::tabnet(epochs = 10) %>%
+    set_mode("regression") %>%
+    set_engine("torch") %>%
+    fit(mpg ~ ., data = mtcars)
+
+  expect_error(axed_out <- axe_fitted(tabnet_fit, verbose = TRUE), NA)
+  expect_lt(lobstr::obj_size(axed_out),lobstr::obj_size(tabnet_fit))
+})
+
+test_that("tabnet_fit + butcher() works", {
+  skip_on_cran()
+  skip_if_not_installed("tabnet")
+  suppressPackageStartupMessages(library(parsnip))
+  # Create model and fit
+  tabnet_fit <- tabnet::tabnet(epochs = 10) %>%
+    set_mode("regression") %>%
+    set_engine("torch") %>%
+    fit(mpg ~ ., data = mtcars)
+
+  expect_error(tabnet_out <- butcher(tabnet_fit, verbose = TRUE), NA)
+})
+
+test_that("tabnet_fit + predict() works", {
+  skip_on_cran()
+  skip_if_not_installed("tabnet")
+  suppressPackageStartupMessages(library(parsnip))
+  # Create model and fit
+  tabnet_fit <- tabnet::tabnet(epochs = 10) %>%
+    set_mode("regression") %>%
+    set_engine("torch") %>%
+    fit(mpg ~ ., data = mtcars)
+
+  tabnet_out <- butcher(tabnet_fit, verbose = TRUE)
+  new_data <- as.matrix(mtcars[1:3, 2:11])
+  expect_equal(predict(tabnet_out,new_data), predict(tabnet_fit, new_data))
+})