Viewing profile — craigacp

SIMT - Single Instruction, Multiple Threads

The same operations in the same order is a tough constraint in an environment where core count is increasing and clock speeds/IPC are not. It's hard to rewrite some of these algori…

Some of the gradients are present in the TF C API (and thus the C++ API), but it's hit & miss which are in C and which are in Python. There was an attempt years ago to port more gr…

The encoder's embedding is contextual, it depends on all the tokens. If you pull out the embedding layer from a decoder only model then that is a fixed embedding where each token's…

There are a bunch of these things in a word2vec space. I had a blog post years ago on my group's blog which trained word2vec on a bunch of wikias so we could find out who is the Ha…

Java has had the ability to run a single code file directly by running `java ` since Java 11 - https://openjdk.org/jeps/330 . And it's had a Java REPL since Java 9.

There's usually a two or three step training procedure, first training to predict the next word on a huge corpus of text (billions or trillions of words), then possibly some instru…

I did. It depends what you want, for an overview of how ONNX Runtime works then Microsoft have a bunch of things on https://onnxruntime.ai , but the Java content is a bit lacking o…

Both GPU support and SIMD CPU support are being worked on in OpenJDK projects, there are talks on them from last year's JVM language summit - https://youtube.com/playlist?list=PLX8…

ONNX Runtime needed a good demo application in Java and this was fun to do (I maintain the Java API for ONNX Runtime and wrote this SD implementation). I've added SDv2 and SDXL sup…

There's a correction to that tweet, larger vocab means fewer tokens for any given sequence (usually, assuming it's not to add other languages or character sets).

As ONNX models are protobufs you can edit them at a Python or Java REPL (or other language but I've personally used those two). Dumping them out as text seems like a lot more work,…

We built model & data provenance into our open source ML library, though it's admittedly not the W3C PROV standard. There were a few gaps in it until we built an automated reproduc…

A conventional neural network, i.e. one using a stack of dense layers, can't unroll across a sequence in the way the transformer does. So while it could compute the relative import…

Exactly so, there are a bunch of Java interfaces to native ML libraries that have some messy bit hacking to convert between fp16 and Java floats. This API along with the intrinsifi…

Probabilistic programming can be done via MCMC approaches, but you can also infer the necessary quantities by using variational inference (which approximates the distribution descr…

There is extensive discussion of that topic in the valhalla-dev and valhalla-spec-experts mailing lists, but I think the details are still being worked out.

There are JEPs out for parts of it now - Primitive Objects (which is the new name for value types) https://openjdk.java.net/jeps/401 and Unify the basic primitives with objects htt…

XGBoost also has a binary file format which is neither pickle nor json. It's shared across their Python, Java and R frontends and parsed by the C++ library.

The reviews for published NeurIPS papers for at least the past 5 years are available from each paper's page on http://papers.neurips.cc .

It's weird to me that people build libraries on top of the ML stack to track provenance, when it's really the ML library's job to do that for its inputs. However it is a right pain…

You can with some hoop jumping in the current release (e.g. using 1.14 you need to define the model and optimizer in Python - https://github.com/oracle/tribuo/blob/main/Interop/Ten…

The Tensorflow Java project is alive and well. The next version of the API based on TF 2 is coming out soon. (Full disclosure, I'm a member of the SIG that's building it). I'm a fi…

Actually my group in Oracle Labs launched a Java ML library last week - https://tribuo.org .

We're implementing the extra trees algorithm at the moment, an isolation forest is only a small amount of code from there.