ROCm/FastFlowLM v0.9.0

🏁 FastFlowLM v0.9.0 — First Official Release with Major Prefill Boost & Version Control

on GitHub

We’re excited to announce FastFlowLM v0.9.0, the first official release packed with critical performance and infrastructure updates.

⚠️ Important Notice

This version introduces breaking changes.
A fresh reinstall is required to use all models properly.
All models will be re-downloaded automatically after reinstalling.

✨ What's New

⚡ 1. Prefill Speed Optimization

• Significant prefill performance boost across all supported models.
• Reduces initial token latency, especially for large models and long prompts.

🗂️ 2. Model Version Control

• Introduced internal version tracking for both:
  • Model format compatibility
  • NPU accelerator updates
• This paves the way for future upgrades without breaking older model downloads.

✅ Usage Recap

Interactive CLI:

flm run llama3.2
flm run qwen3:4b prompt.txt

Server Mode:

flm serve

Model Management:

flm list
flm remove llama3.1:8b
flm version

This marks a major milestone in FastFlowLM’s development — delivering faster performance and laying the foundation for robust, future-proof model deployment.