Skip to content

🖼️ FastFlowLM v0.9.23 — Faster Vision Prefill for Gemma3 & MedGemma

Choose a tag to compare

View all tags
@ngdxzy ngdxzy released this 11 Dec 16:52
· 906 commits to main since this release
v0.9.23
5e162ff
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

FastFlowLM v0.9.23 introduces a 35% speed boost for image understanding in vision-enabled Gemma models.

⚡ Vision Prefill Optimization

Models Improved:

  • gemma3:4b
  • medgemma:4b

What’s new:

  • Reduced Time to First Token (TTFT) from ~4.5s → ~3.4s for single-image cases (35% speedup).
  • Noticeably faster responses in visual chat and medical imaging scenarios.

🛠️ FLM Runtime Improvements

  • Model downloads now print both the total download size and per-file sizes — thanks to @jeremyfowers for the suggestion!

🤖 Flm-Companion (Independent Project)

Flm-Companion, created by @julienM77, is a modern GUI designed to complement FastFlowLM.
It provides an intuitive way to run local models, monitor the server, and manage configurations.

This release continues our push to accelerate on-device multimodal inference, especially for latency-sensitive workflows.

Contributors

julienM77 and jeremyfowers
Assets 3
Loading