ROCm/FastFlowLM v0.9.5

📦 FastFlowLM v0.9.5 — Small Gemma Models & CLI Improvements

on GitHub

FastFlowLM v0.9.5 adds lightweight Gemma model support, simplifies model usage, and introduces multiple CLI quality-of-life upgrades.

🆕 What’s New

• Gemma3 Small Variants
  • Added gemma3:1b and gemma3:270m
  • ⚠️ gemma3:270m is limited and not recommended for Q&A tasks
  • ⚠️ gemma3:270m may occasionally get stuck in an infinite loop on long answers (a limitation of the Unsloth-provided weights, also observed in LM studio).
• Model Simplification
  • Removed gemma3-text:4b
  • Use gemma3:4b for both text and vision
• Model weights Update
  • Updated gemma3:4b weights for better quality and stability
• Interrupt Generation
  • Stop token generation in CLI (flm run) with Ctrl+C
• Installer Options
  • Configure model storage directory and REST API port at install
• /show Command
  • Quickly check a model’s maximum context length
• Improved CLI Help
  • Shows active model directory and REST API port
• Legacy Input Removed
  • Old syntax:

    flm run <tag> <input_file> <length>

    ❌ Removed
  • New way:

    /input "path/to/book.txt" Please summarize.
    

• Stability Fixes
  • Added 1-hour download timeout for unstable/slow networks
  • Fixed broken CLI input file paths

🚀 Summary

This release improves support for smaller Gemma models, simplifies model selection, removes legacy formats, and delivers a smoother CLI experience with clearer commands, better control, and flexible installer options.