# Problem loading saved neural network

**URL:** <https://discourse.julialang.org/t/problem-loading-saved-neural-network/98882>\
**Category:** Machine Learning\
**Tags:** error, flux, bson\
**Created:** [May 15, 2023, 2:40pm UTC](https://discourse.julialang.org/t/problem-loading-saved-neural-network/98882 "2023-05-15T14:40:38Z")\
**Posts on this page:** 10\
**Page:** 1

<div class="post-metadata">

**Author:** ![LucasMSpereira](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/lucasmspereira/32/26393_2.png) [@LucasMSpereira](https://discourse.julialang.org/u/LucasMSpereira)\
**Post date:** [May 15, 2023, 2:40pm UTC](https://discourse.julialang.org/t/problem-loading-saved-neural-network/98882/1 "2023-05-15T14:40:38Z")

</div>

When trying to load a saved Flux NN with  
`BSON.@load datasetPath * "trainedNetworks/convnextGen.bson" cpuGenerator`  
I’m getting a `invalid struct allocation` error.

Since doing this last time, the only thing I did was update julia to 1.9.0 when it launched. In the meantime, BSON.jl and Flux.jl versions didn’t change, and nothing was done to the BSON file. What is the problem here?

Stacktrace:

```julia
ERROR: invalid struct allocation
Stacktrace:
   [1] newstruct(::Type, ::Module, ::Vararg{Any})
     @ BSON C:\Users\kaoid\.julia\packages\BSON\73cTU\src\extensions.jl:135
   [2] (::BSON.var"#47#48")(d::Dict{Symbol, Any})
     @ BSON C:\Users\kaoid\.julia\packages\BSON\73cTU\src\extensions.jl:159
   [3] _raise_recursive(d::Dict{Symbol, Any}, cache::IdDict{Any, Any}, init::Module)    
     @ BSON C:\Users\kaoid\.julia\packages\BSON\73cTU\src\read.jl:82
   [4] (::BSON.var"#49#50")(d::Dict{Symbol, Any}, cache::IdDict{Any, Any}, init::Module)
     @ BSON C:\Users\kaoid\.julia\packages\BSON\73cTU\src\extensions.jl:168
   [5] raise_recursive(d::Dict{Symbol, Any}, cache::IdDict{Any, Any}, init::Module)
     @ BSON C:\Users\kaoid\.julia\packages\BSON\73cTU\src\read.jl:92
   [6] (::BSON.var"#23#24"{IdDict{Any, Any}, Module})(x::Dict{Symbol, Any})
     @ BSON C:\Users\kaoid\.julia\packages\BSON\73cTU\src\read.jl:98
   [7] applychildren!(f::BSON.var"#23#24"{IdDict{Any, Any}, Module}, x::Vector{Any})
     @ BSON C:\Users\kaoid\.julia\packages\BSON\73cTU\src\BSON.jl:26
   [8] raise_recursive
     @ C:\Users\kaoid\.julia\packages\BSON\73cTU\src\read.jl:98 [inlined]
   [9] (::BSON.var"#45#46"{IdDict{Any, Any}, Module})(x::Vector{Any})
     @ BSON C:\Users\kaoid\.julia\packages\BSON\73cTU\src\extensions.jl:153
  [10] iterate
     @ .\generator.jl:47 [inlined]
  [11] collect_to!(dest::Vector{Symbol}, itr::Base.Generator{Vector{Any}, BSON.var"#45#46"{IdDict{Any, Any}, Module}}, offs::Int64, st::Int64)
     @ Base .\array.jl:840
  [12] collect_to_with_first!(dest::Vector{Symbol}, v1::Symbol, itr::Base.Generator{Vector{Any}, BSON.var"#45#46"{IdDict{Any, Any}, Module}}, st::Int64)
     @ Base .\array.jl:818
  [13] _collect(c::Vector{Any}, itr::Base.Generator{Vector{Any}, BSON.var"#45#46"{IdDict{Any, Any}, Module}}, #unused#::Base.EltypeUnknown, isz::Base.HasShape{1})
     @ Base .\array.jl:812
  [14] collect_similar(cont::Vector{Any}, itr::Base.Generator{Vector{Any}, BSON.var"#45#46"{IdDict{Any, Any}, Module}})
     @ Base .\array.jl:711
  [15] map(f::Function, A::Vector{Any})
     @ Base .\abstractarray.jl:3261
  [16] newstruct_raw(cache::IdDict{Any, Any}, T::Type, d::Dict{Symbol, Any}, init::Module)
     @ BSON C:\Users\kaoid\.julia\packages\BSON\73cTU\src\extensions.jl:153
  [17] (::BSON.var"#49#50")(d::Dict{Symbol, Any}, cache::IdDict{Any, Any}, init::Module)
     @ BSON C:\Users\kaoid\.julia\packages\BSON\73cTU\src\extensions.jl:169
  [18] raise_recursive(d::Dict{Symbol, Any}, cache::IdDict{Any, Any}, init::Module)
     @ BSON C:\Users\kaoid\.julia\packages\BSON\73cTU\src\read.jl:92
  [19] (::BSON.var"#23#24"{IdDict{Any, Any}, Module})(x::Dict{Symbol, Any})
     @ BSON C:\Users\kaoid\.julia\packages\BSON\73cTU\src\read.jl:98
  [20] applychildren!(f::BSON.var"#23#24"{IdDict{Any, Any}, Module}, x::Vector{Any})
     @ BSON C:\Users\kaoid\.julia\packages\BSON\73cTU\src\BSON.jl:26
  [21] raise_recursive
     @ C:\Users\kaoid\.julia\packages\BSON\73cTU\src\read.jl:98 [inlined]
  [22] (::BSON.var"#45#46"{IdDict{Any, Any}, Module})(x::Vector{Any})
     @ BSON C:\Users\kaoid\.julia\packages\BSON\73cTU\src\extensions.jl:153
  [23] iterate
     @ .\generator.jl:47 [inlined]
  [24] _collect(c::Vector{Any}, itr::Base.Generator{Vector{Any}, BSON.var"#45#46"{IdDict{Any, Any}, Module}}, #unused#::Base.EltypeUnknown, isz::Base.HasShape{1})
     @ Base .\array.jl:802
  [25] collect_similar(cont::Vector{Any}, itr::Base.Generator{Vector{Any}, BSON.var"#45#46"{IdDict{Any, Any}, Module}})
     @ Base .\array.jl:711
  [26] map(f::Function, A::Vector{Any})
     @ Base .\abstractarray.jl:3261
  [27] newstruct_raw(cache::IdDict{Any, Any}, T::Type, d::Dict{Symbol, Any}, init::Module)
     @ BSON C:\Users\kaoid\.julia\packages\BSON\73cTU\src\extensions.jl:153
  [28] (::BSON.var"#49#50")(d::Dict{Symbol, Any}, cache::IdDict{Any, Any}, init::Module)
     @ BSON C:\Users\kaoid\.julia\packages\BSON\73cTU\src\extensions.jl:169
  [29] raise_recursive(d::Dict{Symbol, Any}, cache::IdDict{Any, Any}, init::Module)
     @ BSON C:\Users\kaoid\.julia\packages\BSON\73cTU\src\read.jl:92
  [30] (::BSON.var"#45#46"{IdDict{Any, Any}, Module})(x::Dict{Symbol, Any})
     @ BSON C:\Users\kaoid\.julia\packages\BSON\73cTU\src\extensions.jl:153
  [31] iterate
     @ .\generator.jl:47 [inlined]
  [32] collect_to!(dest::Vector{Any}, itr::Base.Generator{Vector{Any}, BSON.var"#45#46"{IdDict{Any, Any}, Module}}, offs::Int64, st::Int64)
     @ Base .\array.jl:840
  [33] collect_to!(dest::Vector{Module}, itr::Base.Generator{Vector{Any}, BSON.var"#45#46"{IdDict{Any, Any}, Module}}, offs::Int64, st::Int64)
     @ Base .\array.jl:848
  [34] collect_to_with_first!(dest::Vector{Module}, v1::Module, itr::Base.Generator{Vector{Any}, BSON.var"#45#46"{IdDict{Any, Any}, Module}}, st::Int64)
     @ Base .\array.jl:818
  [35] _collect(c::Vector{Any}, itr::Base.Generator{Vector{Any}, BSON.var"#45#46"{IdDict{Any, Any}, Module}}, #unused#::Base.EltypeUnknown, isz::Base.HasShape{1})
     @ Base .\array.jl:812
  [36] collect_similar(cont::Vector{Any}, itr::Base.Generator{Vector{Any}, BSON.var"#45#46"{IdDict{Any, Any}, Module}})
     @ Base .\array.jl:711
  [37] map(f::Function, A::Vector{Any})
     @ Base .\abstractarray.jl:3261
  [38] newstruct_raw(cache::IdDict{Any, Any}, T::Type, d::Dict{Symbol, Any}, init::Module)
     @ BSON C:\Users\kaoid\.julia\packages\BSON\73cTU\src\extensions.jl:153
  [39] (::BSON.var"#49#50")(d::Dict{Symbol, Any}, cache::IdDict{Any, Any}, init::Module)
     @ BSON C:\Users\kaoid\.julia\packages\BSON\73cTU\src\extensions.jl:169
  [40] raise_recursive(d::Dict{Symbol, Any}, cache::IdDict{Any, Any}, init::Module)
     @ BSON C:\Users\kaoid\.julia\packages\BSON\73cTU\src\read.jl:92
  [41] (::BSON.var"#23#24"{IdDict{Any, Any}, Module})(x::Dict{Symbol, Any})
     @ BSON C:\Users\kaoid\.julia\packages\BSON\73cTU\src\read.jl:98
  [42] applychildren!(f::BSON.var"#23#24"{IdDict{Any, Any}, Module}, x::Vector{Any})
     @ BSON C:\Users\kaoid\.julia\packages\BSON\73cTU\src\BSON.jl:26
  [43] raise_recursive
     @ C:\Users\kaoid\.julia\packages\BSON\73cTU\src\read.jl:98 [inlined]
  [44] (::BSON.var"#23#24"{IdDict{Any, Any}, Module})(x::Vector{Any})
     @ BSON C:\Users\kaoid\.julia\packages\BSON\73cTU\src\read.jl:98
  [45] applychildren!(f::BSON.var"#23#24"{IdDict{Any, Any}, Module}, x::Vector{Any})
     @ BSON C:\Users\kaoid\.julia\packages\BSON\73cTU\src\BSON.jl:26
  [46] raise_recursive(v::Vector{Any}, cache::IdDict{Any, Any}, init::Module)
     @ BSON C:\Users\kaoid\.julia\packages\BSON\73cTU\src\read.jl:98
  [47] newstruct_raw(cache::IdDict{Any, Any}, #unused#::Type{Core.TypeName}, d::Dict{Symbol, Any}, init::Module)
     @ BSON C:\Users\kaoid\.julia\packages\BSON\73cTU\src\anonymous.jl:146
  [48] (::BSON.var"#49#50")(d::Dict{Symbol, Any}, cache::IdDict{Any, Any}, init::Module)
     @ BSON C:\Users\kaoid\.julia\packages\BSON\73cTU\src\extensions.jl:169
  [49] raise_recursive(d::Dict{Symbol, Any}, cache::IdDict{Any, Any}, init::Module)
     @ BSON C:\Users\kaoid\.julia\packages\BSON\73cTU\src\read.jl:92
  [50] (::BSON.var"#18#21"{IdDict{Any, Any}, Module})(x::Dict{Symbol, Any})
     @ BSON C:\Users\kaoid\.julia\packages\BSON\73cTU\src\read.jl:82
  [51] applychildren!(f::BSON.var"#18#21"{IdDict{Any, Any}, Module}, x::Dict{Symbol, Any})
     @ BSON C:\Users\kaoid\.julia\packages\BSON\73cTU\src\BSON.jl:19
  [52] _raise_recursive(d::Dict{Symbol, Any}, cache::IdDict{Any, Any}, init::Module)
     @ BSON C:\Users\kaoid\.julia\packages\BSON\73cTU\src\read.jl:82
  [53] raise_recursive(d::Dict{Symbol, Any}, cache::IdDict{Any, Any}, init::Module)
     @ BSON C:\Users\kaoid\.julia\packages\BSON\73cTU\src\read.jl:93
  [54] (::BSON.var"#23#24"{IdDict{Any, Any}, Module})(x::Dict{Symbol, Any})
     @ BSON C:\Users\kaoid\.julia\packages\BSON\73cTU\src\read.jl:98
  [55] applychildren!(f::BSON.var"#23#24"{IdDict{Any, Any}, Module}, x::Vector{Any})
     @ BSON C:\Users\kaoid\.julia\packages\BSON\73cTU\src\BSON.jl:26
  [56] raise_recursive
     @ C:\Users\kaoid\.julia\packages\BSON\73cTU\src\read.jl:98 [inlined]
  [57] (::BSON.var"#17#20"{IdDict{Any, Any}, Module})(x::Vector{Any})
     @ BSON C:\Users\kaoid\.julia\packages\BSON\73cTU\src\read.jl:80
  [58] applychildren!(f::BSON.var"#17#20"{IdDict{Any, Any}, Module}, x::Dict{Symbol, Any})
     @ BSON C:\Users\kaoid\.julia\packages\BSON\73cTU\src\BSON.jl:19
  [59] _raise_recursive(d::Dict{Symbol, Any}, cache::IdDict{Any, Any}, init::Module)
     @ BSON C:\Users\kaoid\.julia\packages\BSON\73cTU\src\read.jl:80
--- the last 7 lines are repeated 6 more times ---
 [102] raise_recursive(d::Dict{Symbol, Any}, cache::IdDict{Any, Any}, init::Module)
     @ BSON C:\Users\kaoid\.julia\packages\BSON\73cTU\src\read.jl:93
 [103] (::BSON.var"#49#50")(d::Dict{Symbol, Any}, cache::IdDict{Any, Any}, init::Module)
     @ BSON C:\Users\kaoid\.julia\packages\BSON\73cTU\src\extensions.jl:167
 [104] raise_recursive(d::Dict{Symbol, Any}, cache::IdDict{Any, Any}, init::Module)
     @ BSON C:\Users\kaoid\.julia\packages\BSON\73cTU\src\read.jl:92
 [105] (::BSON.var"#19#22"{IdDict{Any, Any}, Module})(x::Dict{Symbol, Any})
     @ BSON C:\Users\kaoid\.julia\packages\BSON\73cTU\src\read.jl:86
 [106] applychildren!(f::BSON.var"#19#22"{IdDict{Any, Any}, Module}, x::Dict{Symbol, Any})
     @ BSON C:\Users\kaoid\.julia\packages\BSON\73cTU\src\BSON.jl:19
 [107] _raise_recursive(d::Dict{Symbol, Any}, cache::IdDict{Any, Any}, init::Module)
     @ BSON C:\Users\kaoid\.julia\packages\BSON\73cTU\src\read.jl:86
 [108] raise_recursive(d::Dict{Symbol, Any}, cache::IdDict{Any, Any}, init::Module)
     @ BSON C:\Users\kaoid\.julia\packages\BSON\73cTU\src\read.jl:93
 [109] raise_recursive
     @ C:\Users\kaoid\.julia\packages\BSON\73cTU\src\read.jl:103 [inlined]
 [110] load (repeats 2 times)
     @ C:\Users\kaoid\.julia\packages\BSON\73cTU\src\read.jl:108 [inlined]
 [111] macro expansion
     @ C:\Users\kaoid\.julia\packages\BSON\73cTU\src\BSON.jl:50 [inlined]
 [112] loadTrainedGANs(model::Symbol)
     @ Main c:\Users\kaoid\My Drive\Estudo\Poli\Pesquisa\Programas\QuickTO\QuickTO\utilities\IO\readIO.jl:156

```

---

<div class="post-metadata">

**Author:** ![jling](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/jling/32/212909_2.png) [@jling](https://discourse.julialang.org/u/jling)\
**Post date:** [May 15, 2023, 6:15pm UTC](https://discourse.julialang.org/t/problem-loading-saved-neural-network/98882/2 "2023-05-15T18:15:56Z")

</div>

did you remember to move model back to CPU before saving it?

---

<div class="post-metadata">

**Author:** ![LucasMSpereira](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/lucasmspereira/32/26393_2.png) [@LucasMSpereira](https://discourse.julialang.org/u/LucasMSpereira)\
**Post date:** [May 15, 2023, 6:17pm UTC](https://discourse.julialang.org/t/problem-loading-saved-neural-network/98882/3 "2023-05-15T18:17:51Z")

</div>

Yes

---

<div class="post-metadata">

**Author:** ![jling](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/jling/32/212909_2.png) [@jling](https://discourse.julialang.org/u/jling)\
**Post date:** [May 15, 2023, 6:20pm UTC](https://discourse.julialang.org/t/problem-loading-saved-neural-network/98882/4 "2023-05-15T18:20:06Z")

</div>

> [@LucasMSpereira](#):
>
> the only thing I did was update julia to 1.9.0

oh, from docs

> Saving models this way could lead to compatibility issues across julia versions and across Flux versions if some of the Flux layers’ internals are changed. It is therefore not recommended for long term storage, use [`Flux.state`](https://fluxml.ai/Flux.jl/stable/destructure/#Flux.state) instead.

---

<div class="post-metadata">

**Author:** ![LucasMSpereira](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/lucasmspereira/32/26393_2.png) [@LucasMSpereira](https://discourse.julialang.org/u/LucasMSpereira)\
**Post date:** [May 15, 2023, 6:22pm UTC](https://discourse.julialang.org/t/problem-loading-saved-neural-network/98882/5 "2023-05-15T18:22:59Z")

</div>

Oof. After so much time reading Flux’s docs, I forgot that 😅. I’ll look in to it. thanks

---

<div class="post-metadata">

**Author:** ![ToucheSir](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/touchesir/32/14411_2.png) [@ToucheSir](https://discourse.julialang.org/u/ToucheSir)\
**Post date:** [May 16, 2023, 12:42am UTC](https://discourse.julialang.org/t/problem-loading-saved-neural-network/98882/6 "2023-05-16T00:42:21Z")

</div>

In fairness, those docs are brand new!

@LucasMSpereira, as a stopgap you can also load your model on 1.8, save them in another format (JLD2, native Julia `Serialization`, etc) and load them back in on 1.9. Those two will have a better chance of working than BSON.jl because the latter is unmaintained and often broken by new Julia versions.

---

<div class="post-metadata">

**Author:** ![LucasMSpereira](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/lucasmspereira/32/26393_2.png) [@LucasMSpereira](https://discourse.julialang.org/u/LucasMSpereira)\
**Post date:** [May 16, 2023, 1:20am UTC](https://discourse.julialang.org/t/problem-loading-saved-neural-network/98882/7 "2023-05-16T01:20:41Z")

</div>

That’s great to know. I was going to just go back to 1.8.5 for this project, but I would miss the (much) faster package load times of 1.9.0. I’ll try those other methods then. Thanks

---

<div class="post-metadata">

**Author:** ![HenriDeh](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/henrideh/32/8316_2.png) [@HenriDeh](https://discourse.julialang.org/u/HenriDeh)\
**Post date:** [May 16, 2023, 3:31pm UTC](https://discourse.julialang.org/t/problem-loading-saved-neural-network/98882/8 "2023-05-16T15:31:50Z")

</div>

Had the same issue, I can confirm that reverting to 1.8.5, as frustrating as it is, works.

---

<div class="post-metadata">

**Author:** ![ToucheSir](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/touchesir/32/14411_2.png) [@ToucheSir](https://discourse.julialang.org/u/ToucheSir)\
**Post date:** [May 16, 2023, 3:42pm UTC](https://discourse.julialang.org/t/problem-loading-saved-neural-network/98882/9 "2023-05-16T15:42:17Z")

</div>

Why not try one of the two approaches above? It’s the way things are moving anyhow (because so many of us have been bitten by BSON.jl issues), and it should only take a few seconds.

---

<div class="post-metadata">

**Author:** ![HenriDeh](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/henrideh/32/8316_2.png) [@HenriDeh](https://discourse.julialang.org/u/HenriDeh)\
**Post date:** [May 16, 2023, 4:09pm UTC](https://discourse.julialang.org/t/problem-loading-saved-neural-network/98882/10 "2023-05-16T16:09:46Z")

</div>

Will do, thanks for the suggestion btw.