tensorflow / haskell

Haskell bindings for TensorFlow
https://tensorflow.github.io/haskell/haddock/
Apache License 2.0
1.58k stars 196 forks source link

tensorflow-core-ops compilation error #270

Open svenssonjoel opened 3 years ago

svenssonjoel commented 3 years ago

Used these commands from your docker file to obtain tensorflow: curl -O https://storage.googleapis.com/tensorflow/libtensorflow/libtensorflow-cpu-linux-x86_64-2.3.0.tar.gz && \ tar zxf libtensorflow-cpu-linux-x86_64-2.3.0.tar.gz -C /usr/local && \ ldconfig && \

But when doing stack build:

tensorflow-core-ops> Configuring tensorflow-core-ops-0.2.0.1... tensorflow-core-ops> build tensorflow-core-ops> parseAttrType: unrecognized type "func" for op "LegacyParallelInterleaveDatasetV2" tensorflow-core-ops> CallStack (from HasCallStack): tensorflow-core-ops> error, called at src/TensorFlow/OpGen/ParsedOp.hs:345:10 in tensorflow-opgen-0.2.0.1-BoVt9892svW5DvHSRs0Z2o:TensorFlow.OpGen.ParsedOp

blackgnezdo commented 3 years ago

We only support TF-2.3 with 0.3.0.0: https://github.com/tensorflow/haskell/blob/master/tensorflow-core-ops/tensorflow-core-ops.cabal#L2

Someone needs to push a new version to hackage. How hard is it for you to use the HEAD from this repo?

svenssonjoel commented 3 years ago

I'd rather grab from hackage or if it was integrated well with stack. Is that stackage? My stack.yaml file is full of those odd dependency lines with ugly hashes for it to find tensorflow.

Btw how come this project is split up into so many subpackages when so many of them are needed for very basic examples?