FluxML/Flux.jl

Pull Requests Commits

Add Lux + ReactantNitro versions of the GPT-2 and ResNet examples (#2803)

CarloLucibello committed 1 hour ago

Verified 3b933245

Refresh ResNet and GPT-2 examples and benchmarks (#2794)

CarloLucibello committed 4 hours ago

Verified 422d73f7

Bump version from 0.16.14 to 0.16.15

CarloLucibello committed 23 hours ago

Verified 095766de

One GC pacer per model for train! and trainstep!, with a full GC after a step that compiles (#2798)

CarloLucibello committed 23 hours ago

Verified a22f35ef

Make `g` of `WeightNorm` with `dims = :` an array, so that it is trained (#2797)

parth-patel010 committed 23 hours ago

Verified eee35111

Mooncake: move the rule for mse here from Mooncake's Flux extension (#2772)

AstitvaAggarwal committed 1 day ago

Verified f45b4779

Count a change of exactly `min_dist` as no improvement in `early_stopping` and `plateau` (#2793)

parth-patel010 committed 1 day ago

Verified 20a0a8f5

Add a precompile workload (#2792)

CarloLucibello committed 2 days ago

Verified 3b9a07ea

Fix Enzyme zero gradient for a loss returning a NamedTuple with aux arrays (#2790)

CarloLucibello committed 2 days ago

Verified 5d4c57fa

GPT-2 example: add a CUDA.jl version (#2765)

CarloLucibello committed 2 days ago

Verified 07ac41c7

Fix the gradient of `logitcrossentropy` for a 0-dimensional label (#2788)

parth-patel010 committed 2 days ago

Verified fa4f7c93

Bump version from 0.16.13 to 0.16.14

CarloLucibello committed 4 days ago

Verified 53a969e8

Drop the Reactant override of `logsumexp` in the label loss (#2785)

CarloLucibello committed 4 days ago

Verified 102614dd

Fused rule for `logitcrossentropy` with class labels (#2770)

CarloLucibello committed 4 days ago

Verified 6ffe197d

Let layers customize `f16`/`bf16`/`f32`/`f64` with a 2-argument method (#2784)

CarloLucibello committed 4 days ago

Verified ea4312c0

Build error messages lazily to cut first-call latency (#2783)

CarloLucibello committed 4 days ago

Verified e7710810

Remove the ignored Greek keywords of `focal_loss` and `tversky_loss` (#2779)

parth-patel010 committed 4 days ago

Verified 7def663f

Reactant cache test: count only the test model's entries (#2775)

CarloLucibello committed 5 days ago

Verified 848039f6

Use `NNlib.layernorm` in the `LayerNorm` forward pass (#2773)

CarloLucibello committed 5 days ago

Verified 2f21ab27

Don't copy BFloat16 input of a BFloat16 layer in `_match_eltype` (#2769)

CarloLucibello committed 6 days ago

Verified 04a9ac3c

Fix the gradient rule of `xlogx` (#2764)

CarloLucibello committed 6 days ago

Verified ef30e8d6

README: standard example first, install/GPU/feature overview; CI: nightly non-blocking (#2763)

CarloLucibello committed 6 days ago

Verified f6aa8297

Add a GPT-2 pretraining example (Reactant, HuggingFaceDatasets.jl) (#2754)

CarloLucibello committed 8 days ago

Verified d004f141

Compute losses in Float32 for Float16/BFloat16 inputs (#2760)

CarloLucibello committed 8 days ago

Verified ba4530fd

Integer class labels in crossentropy and logitcrossentropy (#2758)

CarloLucibello committed 8 days ago

Verified e0d7ea15

Bump version from 0.16.12 to 0.16.13

CarloLucibello committed 9 days ago

Verified ce864c76

CI: run the Reactant tests in their own job (#2753)

CarloLucibello committed 9 days ago

Verified 71df6fcd

CI: fix macOS stalls, cut test and precompile time (#2752)

CarloLucibello committed 9 days ago

Verified c8fac832

Resume a saved optimiser state on a Reactant device (#2749)

CarloLucibello committed 9 days ago

Verified 68872baa

[tests] Expand Reactant test coverage: layers, losses, optimisers, precision (#2750)

CarloLucibello committed 9 days ago

Verified 08ed6809

Older