Compressed MLX builds of other people's LLMs so they fit a 24 GB Mac mini. The base models are upstream, not mine.
Michael Chen
mchen04
AI & ML interests
None yet
Recent Activity
updated a collection 8 days ago
Speech-to-speech on Apple Silicon (MLX) updated a model 11 days ago
mchen04/lfm2.5-audio-1.5b-mlx-delegate published a model 12 days ago
mchen04/lfm2.5-audio-1.5b-mlx-delegateOrganizations
None yet
Compressed MLX builds of other people's LLMs so they fit a 24 GB Mac mini. The base models are upstream, not mine.
A distilled Kokoro-derived TTS engine I built, plus bf16 and 4-bit MLX conversions of an upstream Breeze-TTS model.
Experimental LFM2.5-Audio LoRA adapter with learned delegation and streamed spoken replies. Liquid AI supplies the base model.
Side-project decision heads for a local MLX reader on a 24 GB Mac mini. Negative results included: they do not generalise past their training tasks.