-
Ideally, we have everything to support FP16,
While trying to train Inception-7, I find out it is necessary to support FP16 now, as it is so memory consuming.
Let's set a milestone.
-
I am trying to create a simple FCN using dl4j:
```
ComputationGraphConfiguration.GraphBuilder graph = new NeuralNetConfiguration.Builder().seed(seed)
.optimizationAlgo(OptimizationA…
-
#### Issue Description
```
18:38:13.730 [main] INFO org.nd4j.linalg.factory.Nd4jBackend - Loaded [JCublasBackend] backend
18:38:13.777 [main] WARN org.reflections.Reflections - given scan urls ar…
-
As CRF loss function is used frequently in NLP, maybe it`s good to officially support CRF layer(include Viterbi decoder) under C++ level in Dynet.
-
This plan captures our work from early August to mid September. We will ship around September 15th. Major work items of this iteration include Volta 16bit support and C#/.NET API. There will also be n…
-
Besides the #3 , I have a question on matrix multiply. What is its performance compared to the hardcoded library functions? If it is something like a third, then it might be possible to speed it up by…
-
@mvukov suggested automatically initializing Function instances on first use. I'm not fond of that idea, but I could imagine having initialization be done as part of the constructor, by default. Then …
-
Hi all,
I'm not quite clear on the semantics of the R2L [ForAllRule](https://github.com/opencog/opencog/blob/master/opencog/nlp/scm/relex-to-logic.scm#L263). For the sentence "All Canadians are right-…