No recommended model for 8GB M1

[Auditor08]

Problem

For my 32GB Mac M4 there are plenty of recommended models. But when I installed it on my Macbook M1 with only 8GB of RAM, there is no recommended (tweaked) model.
Shouldn't there be something like Qwen 4B model or a Gemma E4B/E2B?

Proposal

If you have a model optimised for 8GB of RAM, please included it in the list of recommended models.
Otherwise mention it somewhere (in the program, on X, github) which model you recommend for MTPLX.

[youssofal]

On an 8 GB M1, 2.12.0 still has no recommended model. M1 and M2 have no native BF16, so the app and the CLI only offer them FP16 builds, and the smallest one, Qwen 3.5 9B Optimized Speed FP16, peaks at about 10.5 GB. That leaves the list empty, and mtplx start says it needs at least 11 GB. An 8 GB M3, M4 or M5 gets Qwen 3.5 4B Optimized Speed, a 2.6 GB download that peaks near 2.9 GB, but that build is BF16 and there is no FP16 version yet. An FP16 build of the 4B would fix this, so I'm keeping the issue open. Thank you for raising it.

[pumano]

@youssofal did you have plans for support Ming-Image-0.1-design? That model only 6B