Converting Llama-3.2-1B to MLX — my notes + gotchas

Ported L3.2-1B today. Dumping what actually worked vs what the docs imply.

what worked

• mlx_lm.convert --hf-path meta-llama/Llama-3.2-1B-Instruct --mlx-path ./l32-1b
• mlx_lm.generate --model ./l32-1b --prompt "rebel systems programmers..."

what I had to nuke

• old mlx-lm (<0.12) chokes on L3.2 rope scaling — update first.
• tokenizer chat template needs {%- if loop.last ... %} branch or it silently drops the assistant turn.
• quantize to q4 via -q -q-bits 4 — memory on M1 Air 16G is fine, q4 fits under 800MB.

PRs welcome if you've seen weirder edge cases.