ROCm/FastFlowLM v0.9.11

🚀 FastFlowLM v0.9.11: Faster Vision, Smarter Controls, Smoother Flow

on GitHub

✨ What’s New

⚡ 1. Faster Vision Models (~2× speedup on vision head)

Vision models just got a big boost:

• ~2× faster image processing/understanding
• TTFT (Time to First Token) cut from ~8s → ~4s

Try it:

flm run gemma3:4b
flm run medgemma

These commands will automatically download the updated models.

🔔 2. Auto Version Detection

Running an older FLM? The console will now notify you with a direct upgrade link
(activates starting with next version v0.9.12).

🌐 3. Change Server Port at Launch

Pick the port that fits your setup:

• Show current FLM port in PowerShell:

  flm port

• Set a custom port at runtime:

  flm serve llama3.2:1b --port 8000
  flm serve llama3.2:1b -p 8000

⚠️ Tip: --port (-p) only affects the current run; it won’t change the default port.

📊 4. Server Mode Performance Insights

Get real-time performance numbers (TTFT, decoding speed in TPS, etc.) where you need them:

• Shown in console while serving
• Open WebUI supports monitoring

🧠 5. Clear Model Handling

• Removed the “fallback to default model” behavior.
• Unsupported names now return a clear error message.

🌟 Summary

This release delivers faster vision inference, smarter server configuration,
and improved runtime stability.