gabfssilva/lift-MLX-oQ4

Add oQ3 to variants table

Add oQ3.5 to variants table

Specify test hardware in card footnote

Add model card

Fix eos_token_id: add <|im_end|> (248046) so MLX servers stop cleanly

Add oQ4 (data-driven mixed ~4.6bpw) MLX quant of datalab-to/lift

initial commit