coaster
coaster copied to clipboard
fff-rs
Extendable HPC-Framework for CUDA, OpenCL and common CPU
I just notice this repo as a new active development fork of leaf, probably it will be great to provide a quick info for what has changed and what's on...
I noticed that the Documentation pointed to . However, it seems it was intended for this to be .
This would allow the individual backend to decide what accuracy would fit best. On CPU i.e. there is no gain in using fp16 over fp32, on the GPU - if...
Combind with blas to parallelize recursive operations efficiently.
This would potentially allow gains when learning since dual numers/autodiff could be employed more effectively.
Starting points: https://leonardoaraujosantos.gitbooks.io/artificial-inteligence/content/batch_norm_layer.html Google Paper: https://arxiv.org/abs/1502.03167
This would ease backend implementation and remove some code duplication.
https://ci.spearow.io/teams/spearow/pipelines/juice/jobs/test-coaster/builds/69 ``` # RUST_BACKTRACE=1 cargo t --verbose -- --nocapture backend_spec::opencl Fresh rustc-serialize v0.3.24 Fresh num-traits v0.1.40 Fresh libc v0.2.31 Fresh lazy_static v0.1.16 Fresh bitflags v0.3.3 Fresh utf8-ranges v1.0.0 Fresh byteorder...
Split out all backend specific functionalitity and make this interface only crate