Carlo Lucibello

possibly related to #2508

tracking this in https://github.com/EnzymeAD/Enzyme.jl/issues/2837

Related to this (or maybe worth opening a new issue), the search bar at https://touying-typ.github.io/ is of little use, since it only retrieves previous docs from previous versions of the...

Thanks, this is a nice start. A few comments: - No need to introduce the `DotProductAttention` type, we can use the [MultiHeadAttention](https://fluxml.ai/Flux.jl/stable/models/layers/#Flux.MultiHeadAttention) from Flux. According to the table A.2-A.5 in...

Thanks for this PR. You need to include the file in `src/MLDatasets.jl`. Also needs some tests, a docstring and an entry in the documentation in `docs/src/datasets`

You mean the function `jl_gc_safe_enter`? I added it to the list and now I get: ``` ERROR: EnzymeNoDerivativeError: No create nofree of empty function (cuStreamCreate) cuStreamCreate) at context: call fastcc...

they keep coming at me :) so far I annotated "cuStreamCreate", "cuCtxGetCurrent", "cuStreamGetCaptureInfo", "cuStreamQuery", "cuStreamSynchronize", "cuMemGetInfo_v2", "cuMemPoolCreate", "cuDeviceSetMemPool", "cuMemPoolSetAttribute", "jl_gc_safe_enter", "jl_gc_safe_leave", "ijl_gc_is_in_finalizer", "ijl_get_task_threadpoolid", "jl_rand_ptls", "ijl_wakeup_thread", "ijl_safe_printf", "ijl_threadpoolid", "ijl_task_get_next", "ijl_switch", "ijl_process_events",...

Is there some way to speed up the process? Even if using Revise, have to restart the repl at each addition for the change to take place.

Finally something different: ```julia julia> Enzyme.autodiff(ad, Const(f), Active, dmodel, Const(x)) ERROR: AssertionError: non bits type MUT_REF of String has size 24 != sizeof(Int) from arg type LLVM.ArrayType([3 x i64]) Stacktrace:...

```julia using NNlib: gather, gather! using Enzyme: Const, Duplicated using EnzymeTestUtils src = Float64[3, 4, 5, 6, 7] idx = [ 1 2 3 4; 4 2 1 3; 3...