ROCm/FastFlowLM v0.9.2

## 🌟 FastFlowLM v0.9.2 — Gemma3 Support, Smarter File Input, and Stability Fixes

on GitHub

FastFlowLM v0.9.2 introduces support for the latest Gemma3 model and brings several improvements to file-based input handling, generation control, and CLI interaction.

✨ What's New

🧠 1. Gemma3:4b Support (Text Only)

• Added support for gemma3:4b (text-only) model.
• 🔭 Vision-capable version (VLM) is coming soon!

📄 2. Max Generation Length in File Mode

• When using input file mode:

  flm run llama3.2 input.txt 1024

  You can now specify the maximum generation length directly after the file name.

🖋️ 3. Added /input Command in CLI Mode

• In flm run interactive CLI, you can use:

  /input "<file_path>" What do you think?
  

  • This reads the file content as part of the prompt.
  • If the path contains special characters or spaces, make sure it is wrapped in quotes.
  • Currently, the file must be a pure text file that can be opened correctly by text editors such as Notepad.

🐞 4. Prefill Error Handling Fix

• Previously, if a file contained too many tokens (exceeding model's context limit), FastFlowLM would still attempt generation.
• Now, it exits early with a clear message if prefill is not possible.

🎨 5. Improved App Icon for Dark Mode

• Updated the FastFlowLM icon to look better on dark screen backgrounds.

⚠️ Known Issue

When using flm run <file_name>, ensure that the filename contains only ASCII characters.
If your filename contain non-ASCII characters, use the /input method instead.

✅ Usage Recap

Run CLI Mode:

flm run gemma3

Text File Input:

flm run qwen3:1.7b notes.txt 512

CLI File Prompting:

/input "data/input.txt" Summarize this.

Server Mode:

flm serve

This release strengthens multi-model support and improves input handling to give users more control and clarity in CLI and file-based usage.