But the things we try to make LLMs do post-pre-training are primarily achieved via reinforcement learning. Isn't reinforcement learning machine learning? Correct me if I'm misconstruing what you're trying to say here
Just to follow up on the above replies, you could also just build a single package. For example, you could build asynchronous_union_find with `bazel build //in_memory/connected_components:asynchronous_union_find`. (This isn't very useful outside of the context of a cc_binary rule.)
This in turn allows you to only build and use the 'package' you care about without having to build the whole repo in other projects. Continuing on the above example, if you only wanted to use the asynchronous_union_find.h header file in your project, somewhere in your WORKSPACE file, you add the graph-mining library using a git_repository rule (see WORKSPACE.bazel for examples), and in a cc_library rule in a BUILD file inside your project, you can add a `@graph-mining//in_memory/connected_components:asynchronous_union_find`. Then you can include it as a header elsewhere. Building your project then only builds that package and its dependencies, and not the entire graph-mining library.
Fundamentals of Music Theory course from University of Edinburgh[1] and its accompanying textbook[2]. It's a centered around western music but certainly a good resource.
They did; perhaps not in ways that were immediately obvious to the general public. Back in 2014, they created attention to improve how RNNs handled long sequences, an effort driven by google translate. Same goes for when they created the transformer architecture. The first product to see an improvement was also translate. They simply never exposed the bare language models to the public a la OpenAI style with GPT3