Roadmap¶
Future development plans for Flux.
Current Status: v0.1¶
- Core training loop
- Adaptive async controller
- GRPO, PPO, DPO, REINFORCE algorithms
- SGLang integration
- CUDA IPC weight sync
- Basic checkpointing
v0.2 (Planned)¶
- Multi-node training
- vLLM backend support
- Advanced curriculum learning
- Reward model training
- RLHF evaluation suite
v0.3 (Future)¶
- Online DPO
- Rejection sampling fine-tuning
- Constitutional AI support
- Automated hyperparameter tuning
v1.0 (Long-term)¶
- Production-ready stability
- Comprehensive documentation
- Large-scale benchmarks
- Enterprise features
Contributing¶
We welcome contributions! See Contributing Guide.
Feedback¶
Share your ideas: - GitHub Issues - Discord