Abliterated models by parameter size

Abliterated models by parameter size groups releases by how much hardware they need. Ablation does not change a model's size, so these buckets map directly onto the VRAM you must have available to evaluate one.

• Under 4Bruns on a laptop CPU39
• 4B – 10Bsingle consumer GPU129
• 10B – 25B24GB VRAM class83
• 25B – 50B48GB VRAM class238
• 50B – 100Bmulti-GPU10
• 100B+server-class deployment9

Parameter counts are parsed from model names. For mixture-of-experts models the total count is used, not the active count, since total parameters determine memory footprint.