WIP: CUDNN RNNs #161
Merged
WIP: CUDNN RNNs #161
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
MikeInnes
force-pushed
the
curnn
branch
2 times, most recently
from
10145c8
to
c3589c2
Compare
|
Eesh, this isn't even done yet (we still need to expose sequences properly). The basic RNN API is working though, so it's ready to go for basic stuff. |
|
I cannot get tests passed. INFO: Testing Flux/CUDNN
R = Flux.RNN: Error During Test
Got an exception of type CuArrays.CUDNN.CUDNNError outside of a @test
CUDNNError(code 3, CUDNN_STATUS_BAD_PARAM)
Stacktrace:
[1] macro expansion at /home/iblis/.julia/v0.6/CuArrays/src/dnn/error.jl:19 [inlined]
[2] cudnnRNNForward(::Flux.CUDA.RNNDesc{Float32}, ::Int64, ::Array{CuArrays.CUDNN.TensorDesc,1}, ::CuArray{Float32,2}, ::CuArrays.CUDNN.TensorDe
sc, ::CuArray{Float32,2}, ::Ptr{Void}, ::Ptr{Void}, ::CuArrays.CUDNN.FilterDesc, ::CuArray{Float32,1}, ::Array{CuArrays.CUDNN.TensorDesc,1}, ::CuAr
ray{Float32,2}, ::CuArrays.CUDNN.TensorDesc, ::CuArray{Float32,2}, ::Ptr{Void}, ::Ptr{Void}, ::CuArray{UInt8,1}, ::CuArray{UInt8,1}) at /home/iblis
/.julia/v0.6/Flux/src/cuda/cudnn.jl:131Test Summary: | Pass Error Total
Flux | 109 1 110
Throttle | 11 11
Jacobian | 1 1
Initialization | 14 14
Tracker | 29 29
Dropout | 8 8
BatchNorm | 10 10
losses | 5 5
Optimise | 8 8
Training Loop | 1 1
CuArrays | 4 4
RNN | 15 1 16
R = Flux.RNN | 1 1
R = Flux.GRU | 7 7
R = Flux.LSTM | 8 8 |
|
I'm on 6 so probably just a version issue. Will debug when I can get 7 handy. |
|
@iblis17 Not sure what's happening here, but this passes on my CUDNN 7 machine. julia> Pkg.test("Flux")
INFO: Testing Flux
INFO: Testing Flux/GPU
INFO: Testing Flux/CUDNN
Test Summary: | Pass Total
Flux | 116 116
INFO: Flux tests passed
julia> CuArrays.CUDNN.CUDNN_VERSION
7005 |
|
😱 julia> using CuArrays
julia> CuArrays.CUDNN.CUDNN_VERSION
7005 |
|
Hmmm. If you want to debug it a bit, it might be worth trying out Knet's implementation and seeing if that works. If it does, it should be straightforward (if tedious) to figure out what we're calling differently. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Mostly there now. Once the back passes are wrapped, the main item will be to rewrite our existing RNNs as one big matmul. This should have some CPU perf benefits anyway.