Huihui-Qwen3.5-4B-Claude-4.6-Opus-abliterated

24 of 32 layers use linear attention, which keeps a fixed-size recurrent state instead of a per-token cache. Those layers do not grow with context at all — treating them as ordinary attention, as a flat formula does, overstates this model's cache by roughly 4.0× at long context.

A parameters × bits ÷ 8 estimate puts Q4_K_M at roughly 2.44 GiB. The real file is 2.52 GiB, because a quantization is a mixture and some tensors are always kept at higher precision.

How much VRAM does Huihui-Qwen3.5-4B-Claude-4.6-Opus-abliterated need?

Q4_K_M is exactly 2,708,801,536 bytes (2.52 GiB) in weights. Add the KV cache, which depends on your context length, plus roughly half a gigabyte of runtime overhead.

How large is Huihui-Qwen3.5-4B-Claude-4.6-Opus-abliterated's KV cache?

1.00 GiB at 32K context with an f16 cache, computed per layer. Quantizing the cache to q8_0 roughly halves it, which is often the difference between a context length fitting and not.

Which quantization of Huihui-Qwen3.5-4B-Claude-4.6-Opus-abliterated should I use?

Q4_K_M is the usual default. Pick the largest quantization that fits your card at the context you actually need — the table above gives exact sizes for every one published.