TurboFieldfare
AI
local llm inference engine
SHOW HN2026
Legend
TurboFieldfare
Open-source engine running Gemma 4 26B in 2 GB RAM on any M-series Mac
github.com#gemma 4#local llm#apple silicon#mixture of experts#ssd streaming
How HN took it
23 positive, 63 neutral or asking, 0 mixed, 11 critical, out of 97 comments about the project. Thread score +0.18.
Performance: 5 for, 9 againstIdea: 8 for, 1 againstExecution: 3 for, 0 againstDocs: 2 for, 1 against
The thread, comment by comment
Each row is a top-level comment and its replies, in HN's order. Select a cell to open that comment on HN.
positive neutral or mixed critical about another commenter or something else the maker
What HN said
What people liked
Impressive low memory footprint and performance for a Mac
From 2 comments
It does exactly what it says it does. On my Mac mini M4 with 16GB of ram it is running at just over 5 tok/s. That jump from M4 to M5 is crazy.
greggh on HNRan this on a 64 GB M4 Max MacBook. I figured having Gemma available with a small footprint would be a nice setup. No more unloading models when I need more RAM for work? Hell…
pwython on HN
Useful for running local models on small-memory Macs
From 2 comments
Rushing to try it!
weras on HNYou’re a mad man - thank you! Do I understand correctly that Ollama doesnt do that, and that’s why responses hang forever on a M3 running the same model through Ollama?
cyanregiment on HN
Objections
Slow token speed on older 8 GB and 16 GB machines
From 2 comments
> It currently generates 5–6 tok/s on an 8 GB M2 MacBook Air and 31–35 tok/s on an M5 MacBook Pro. Where does this big a performance spread come from? I wouldn't naïvely expect…
addaon on HNIt does exactly what it says it does. On my Mac mini M4 with 16GB of ram it is running at just over 5 tok/s. That jump from M4 to M5 is crazy.
greggh on HN
Gemma is weak for coding tasks
From 1 comment
Would be awesome if it ran Qwen (the MoE probably won't squeeze that low, but...). This because I have hardly been able to use Gemma for any sort of useful coding.
rcarmo on HN
Requests
Support for Qwen MoE models
From 2 comments
Would be awesome if it ran Qwen (the MoE probably won't squeeze that low, but...). This because I have hardly been able to use Gemma for any sort of useful coding.
rcarmo on HNCan the same be done with qwen3.6-35b-a3b?
febed on HN
Multi-turn chat and a terminal client for the server
From 2 comments
hm. just open repo, copy commands into your terminal and you will get app installed (if you have swift toolchain installed) after that download 14gb of weights and enjoy offline…
gitpusher42 on HNThank you for testing and sharing results! I think I might understand your use case. You ssh the Mac and want something like `ollama run` with an interactive chat in terminal. Am…
gitpusher42 on HN
Questions
How does it compare to plain mmap and llama.cpp
From 1 comment
I'm curious how your project compares to plain mmap! Because llama.cpp will already run 26B in 2GB of RAM if you really want to (mmap enabled, repacking disabled). It seems like…
tredre3 on HN
Why the big speed gap between M2 and M5
From 1 comment
> It currently generates 5–6 tok/s on an 8 GB M2 MacBook Air and 31–35 tok/s on an M5 MacBook Pro. Where does this big a performance spread come from? I wouldn't naïvely expect…
addaon on HN
Alternatives named
Flash-MoE
From 2 comments
Yeah! Check the Colibri and Flash-MoE projects. They’re already doing that. https://github.com/danveloper/flash-moe https://github.com/JustVugg/colibri
gitpusher42 on HNI think I first saw Flash-MoE (https://github.com/danveloper/flash-moe) in April. Huge respect to them, it was a big inspiration for this project!
gitpusher42 on HN
Colibri
From 2 comments
Yeah! Check the Colibri and Flash-MoE projects. They’re already doing that. https://github.com/danveloper/flash-moe https://github.com/JustVugg/colibri
gitpusher42 on HNCheck for colibri, dwarf star and flash-moe. they do similar things with bigger models https://github.com/JustVugg/colibri https://github.com/antirez/ds4…
gitpusher42 on HN
MLX
From 1 comment
It depends on the use case. I measured this exact model with a 4k context on the mlx engine. It runs at 75 tok/s on my M5 Mac Pro and using 14 GB of RAM. For my engine the same…
gitpusher42 on HN
llama.cpp and Ollama
From 2 comments
I'm curious how your project compares to plain mmap! Because llama.cpp will already run 26B in 2GB of RAM if you really want to (mmap enabled, repacking disabled). It seems like…
tredre3 on HNThank you! afaik ollama relies on llama.cpp and mmap. mmap loads pages on demand and doesn't use the same explicit cache or parallel reads like my engine. Most likely…
gitpusher42 on HN