The ocaml-tensorboard library provides functions to write TensorBoard events from ocaml without relying on TensorFlow. TensorBoard is a collection of web applications that makes it easy to visualize and understand what happens during the training of a TensorFlow based model. TensorBoard events are logged for some specified step (which is an increasing integer) and can embed scalars, text, images, histograms...

Installation can be made via opam.

opam install tensorboard

Here is an example of code writing some TensorBoard values.

module Sw = Tensorboard.Summary_writer

let sw = Sw.create "/tmp/tensorboard-logs" in
Sw.write_value sw ~step:14 ~name:"value/3" ~value:42.0;
Sw.write_text sw ~step:28 ~name:"txt" ~text:"foobar";
Sw.close sw

Part of the code comes from ocaml-protoc and ppx_deriving_protobuf.