@jsdevtom prototyped a training pipeline for a small code completion model targeted for defuss. In this story, we will improve of the dataset generation with an agentic synthetic dataset generator, document the training pipeline, run it on a decent GPU cluster, upload the model to HuggingFace, implement a small inference server (defuss-fim-inference) that runs cross-platform and document how to use it in both major IDEs (VS Code platform base, IntelliJ base).
@jsdevtom prototyped a training pipeline for a small code completion model targeted for defuss. In this story, we will improve of the dataset generation with an agentic synthetic dataset generator, document the training pipeline, run it on a decent GPU cluster, upload the model to HuggingFace, implement a small inference server (defuss-fim-inference) that runs cross-platform and document how to use it in both major IDEs (VS Code platform base, IntelliJ base).