Qwen3.8-27B-Heretic-Abliterated-Uncensored requirements — can you run it?

Yes, if combined RAM+VRAM ≥ file size — but generation speed is limited by memory bandwidth. Expect roughly: DDR5 dual channel ~10-30 tok/s for small MoE actives, single digits for big ones, 0.1-1 tok/s when experts page from disk. CPU-only is fine for batch jobs, painful for chat.

Most pages compute weights from a formula (params × bits/8). We use the actual file sizes published in GGUF repos, which include the embedding table, unquantized tensors, and container overhead — that is why our numbers can differ from a naive calculation.

The measured answer: 24.6 GB at the Q5_K_M quant with 8K context. The smallest published build needs 13.8 GB; the lossless (F16/BF16) build needs 139.8 GB. These are real GGUF file sizes from Hugging Face, not formula estimates.

Not at Q5_K_M (24.6 GB). Quants that fit 24GB: Q2_K_S, Q2_K, Q3_K_S, Q3_K_M, Q3_K_L, Q4_K_S, Q5_K_S.

Yes, if your usable RAM is at least the file size. At Q5_K_M you need ~24.6 GB of RAM. CPU generation is memory-bandwidth-bound: expect single-digit tok/s for large MoE models, 10-30 tok/s for small active params, and below 1 tok/s when experts page from disk.

Most sites compute weights from a formula (params x bits/8). ModelFit uses the actual GGUF file sizes published on Hugging Face, which include the embedding table, unquantized tensors, and container overhead — so our numbers reflect what you really download and load.