DavidAU · text · mixture of experts

Mistral-MOE-4X7B-Dark-MultiVerse-Uncensored-Enhanced32-24B

DavidAU/Mistral-MOE-4X7B-Dark-MultiVerse-Uncensored-Enhanced32-24B

Mistral-MOE-4X7B-Dark-MultiVerse-Uncensored-Enhanced32-24B at Q4_K_M is exactly 14,609,885,184 bytes (13.61 GiB / 14.61 GB) — an effective 4.839 bits per weight, not the nominal 4. Its KV cache at 32K is 4.00 GiB.

From the file· summed from 1 file(s)From the file· KV per layer
Parameters
24.2B
total, not active
Architecture
llama
32 layers
Context
32,768
native (config.json)
License
apache-2.0

Shipped quantizations

exact bytes, summed from published files
QuantSizeExact bytesEffective bpwTensorsPublisher
Q2_K8.24 GiB8,842,930,1762.929DevQuasar-9
Q3_K_S9.72 GiB10,433,300,4803.456DevQuasar-9
Q3_K_S9.76 GiB10,482,452,2883.472DavidAU
Q3_K_S9.76 GiB10,482,452,2883.472ekampra
Q3_K_M10.79 GiB11,580,442,6243.836DevQuasar-9
Q3_K_M10.83 GiB11,629,594,4323.852DavidAU
Q3_K_M10.83 GiB11,629,594,4323.852ekampra
Q3_K_L11.68 GiB12,544,083,9684.155DevQuasar-9
Q3_K_L11.73 GiB12,593,235,7764.171ekampra
Q3_K_L11.73 GiB12,593,235,7764.171DavidAU
Q4_K_S12.80 GiB13,743,237,1204.552DevQuasar-9
Q4_K_S12.84 GiB13,791,364,9284.568DavidAU
Q4_K_S12.84 GiB13,791,364,9284.568ekampra
Q4_K_M13.61 GiB14,609,885,1844.839DevQuasar-9
Q4_K_M13.65 GiB14,658,012,9924.855ekampra
Q4_K_M13.65 GiB14,658,012,9924.855DavidAU
Q5_K_S15.48 GiB16,626,427,9045.507DevQuasar-9
Q5_K_S15.53 GiB16,675,579,7125.523DavidAU
Q5_K_S15.53 GiB16,675,579,7125.523ekampra
Q5_K_M15.96 GiB17,134,462,9765.675DevQuasar-9
Q5_K_M16.00 GiB17,183,614,7845.691ekampra
Q5_K_M16.00 GiB17,183,614,7845.691DavidAU
Q6_K18.46 GiB19,816,826,8806.564DevQuasar-9
Q8_023.90 GiB25,666,357,2488.501DevQuasar-9
Q2_K3 shards25.95 GiB27,863,061,9529.229ekampra
Q2_K3 shards25.95 GiB27,863,061,9529.229DavidAU
IQ4_XS3 shards37.66 GiB40,442,303,93613.395DavidAU
IQ4_XS3 shards37.66 GiB40,442,303,93613.395ekampra
F162 shards44.99 GiB48,309,700,83216.001DevQuasar-9
Q6_K3 shards56.49 GiB60,656,752,06420.090ekampra
Q6_K3 shards56.49 GiB60,656,752,06420.090DavidAU
Q8_03 shards72.66 GiB78,014,879,16825.840ekampra
Q8_03 shards72.66 GiB78,014,879,16825.840DavidAU

KV cache by context

computed per layer
ContextKV cache (f16)Flat formulaOverstated byFull / windowed / recurrent
4,0960.50 GiB0.50 GiB32 / 0 / 0
8,1921.00 GiB1.00 GiB32 / 0 / 0
16,3842.00 GiB2.00 GiB32 / 0 / 0
32,7684.00 GiB4.00 GiB32 / 0 / 0
65,5368.00 GiB8.00 GiB32 / 0 / 0
131,07216.00 GiB16.00 GiB32 / 0 / 0

Compare with

same modality, comparable size

Will it run on your card?

full quant x context sweep

Why other calculators give a different number

A parameters × bits ÷ 8 estimate puts Q4_K_M at roughly 12.65 GiB. The real file is 13.61 GiB, because a quantization is a mixture and some tensors are always kept at higher precision.

Architecture

from config.json
Layers
32
Attention heads
32
KV heads
8
Head dim
128
Hidden size
4096
Vocab
32,000
Sliding window
none
SWA period
MLA
no
Experts
4
Experts per token
4
use_sliding_window

Questions people ask

How much VRAM does Mistral-MOE-4X7B-Dark-MultiVerse-Uncensored-Enhanced32-24B need?
Q4_K_M is exactly 14,609,885,184 bytes (13.61 GiB) in weights. Add the KV cache, which depends on your context length, plus roughly half a gigabyte of runtime overhead.
How large is Mistral-MOE-4X7B-Dark-MultiVerse-Uncensored-Enhanced32-24B's KV cache?
4.00 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.
Is Mistral-MOE-4X7B-Dark-MultiVerse-Uncensored-Enhanced32-24B a mixture-of-experts model?
Yes — 4 experts, 4 routed per token. Every expert must be resident, but only the routed ones are read per token, which is why its memory requirement and its speed behave very differently.
Which quantization of Mistral-MOE-4X7B-Dark-MultiVerse-Uncensored-Enhanced32-24B 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.