Skip to content

Use GPUArrays caching allocator in train! #2665

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
CarloLucibello merged 1 commit into from
Mar 26, 2026
Merged

Use GPUArrays caching allocator in train! #2665

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 2 additions & 0 deletions Project.toml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -9,6 +9,7 @@ ChainRulesCore = "d360d2e6-b24c-11e9-a2a3-2a2ae2dbcce4"
Compat = "34da2185-b29b-5c13-b0c7-acf172513d20"
EnzymeCore = "f151be2c-9106-41f4-ab19-57ee4f262869"
Functors = "d9f16b24-f501-4c13-a1f2-28368ffc5196"
GPUArrays = "0c68f7d7-f131-5f86-a1c3-88cf8149b2d7"
LinearAlgebra = "37e2e46d-f89d-539d-b4ee-838fcccc9c8e"
MLCore = "c2834f40-e789-41da-a90e-33b280584a8c"
MLDataDevices = "7e8f7934-dd98-4c1a-8fe8-92b47a384d40"
Expand Down Expand Up @@ -58,6 +59,7 @@ Enzyme = "0.13"
EnzymeCore = "0.7.7, 0.8.4"
FiniteDifferences = "0.12"
Functors = "0.5"
GPUArrays = "11.2"
MLCore = "1.0.0"
MLDataDevices = "1.4.2"
MLUtils = "0.4"
Expand Down
15 changes: 10 additions & 5 deletions src/train.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4,6 +4,7 @@ using LinearAlgebra
using Optimisers: Optimisers
using Functors: fmap, fmapstructure
using ..Flux: Flux
using GPUArrays: GPUArrays

using ProgressLogging: @progress, @withprogress, @logprogress
using EnzymeCore: Duplicated
Expand Down Expand Up @@ -110,19 +111,23 @@ It adds only a few features to the loop above:
function train!(loss, adtype::AbstractADType, model, data, opt; cb = nothing)
isnothing(cb) || error("""train! does not support callback functions.
For more control use a loop with `gradient` and `update!`.""")
cache = GPUArrays.AllocCache()
@withprogress for (i,d) in enumerate(data)
d_splat = d isa Tuple ? d : (d,)

l, gs = Flux.withgradient(m -> loss(m, d_splat...), adtype, model)
GPUArrays.@cached cache begin
l, gs = Flux.withgradient(m -> loss(m, d_splat...), adtype, model)

if !isfinite(l)
throw(DomainError(lazy"Loss is $l on data item $i, stopping training"))
end
if !isfinite(l)
throw(DomainError(lazy"Loss is $l on data item $i, stopping training"))
end

opt, model = _update!(opt, model, gs[1])
opt, model = _update!(opt, model, gs[1])
end

@logprogress Base.haslength(data) ? i/length(data) : nothing
end
GPUArrays.unsafe_free!(cache)
end

_update!(opt_state, model, grads) = Optimisers.update!(opt_state, model, grads)
Expand Down
Loading