Project Roadmap #52
Description
This tracking issue aims to document the project's upcoming features. We ship frequent updates because the LLM ecosystem evolves very rapidly, and a lot of our dependent frameworks & libraries are still being developed.
Completed
- Separate evaluation & inference service + new features Improved evaluation and inference service #53 Support vLLM for inference & eval engine #27
- Deploy on your own GCP project with Terraform IaC to facilitate custom cloud deployment #17
- New separate export service for better flexibility Separate export feature that merges the model or exports it in GGUF format #43 Convert HF models to GGUF with llama.cpp #31
- Documentation Setting up documentation! #61
Work in progress
- Loading fine tuned models to continue training, and with a different trainer [feature] Load saved model to fine tune #48
- Ways to integrate deployment of fine tuned models Deploy fine tuned model to Cloud Run #28
- Fixing a few technical debt with GCP architecture: Replace GCS + Jobs API with a proper Pub/Sub queue #69 Integrate training and export service #67
- Fixing datasets technical debt [question] Change vision dataset formatting in format converter #50
Planned features
- Sharing datasets (and training jobs) on cloud Sharing datasets and training jobs #40
- Instruction tuning explicit support Support instruction tuning datasets #37 Add custom chat template support for instructing tuning #62
- New log streaming system to avoid dependency on WandB [feature] built-in tracker #58
- Benchmark feature in evaluations Add benchmarking in evaluation service #60