How much VRAM does MusaCoder-27B need?

Q4_K_M is exactly 16,547,399,264 bytes (15.41 GiB) in weights. Add the KV cache, which depends on context length, plus roughly half a gigabyte of runtime overhead.

How large is MusaCoder-27B's KV cache?

2.00 GiB at 32K context with an f16 cache, computed per layer from the model's architecture. Quantizing the cache to q8_0 roughly halves it.

MooreThreads · text

MusaCoder-27B

MooreThreads/MusaCoder-27B

MusaCoder-27B at Q4_K_M is exactly 16,547,399,264 bytes (15.41 GiB / 16.55 GB) Its KV cache at 32K is 2.00 GiB.

From the file· summed from 1 file(s)From the file· KV per layer

Parameters

—

Architecture

qwen35

64 layers

Context

262,144

native (config.json)

License

apache-2.0

Shipped quantizations

● exact bytes, summed from published files

Quant	Size●	Exact bytes●	Effective bpw●	Tensors●	Publisher
Q3_K_M	12.39 GiB	13,301,442,144	—	—	uxix
Q4_K_M	15.41 GiB	16,547,399,264	—	—	uxix
Q5_K_M	17.91 GiB	19,231,098,464	—	—	uxix
Q6_K	20.57 GiB	22,082,528,864	—	—	uxix
Q8_0	26.63 GiB	28,595,762,784	—	—	uxix

KV cache by context

computed per layer

Context	KV cache (f16)●	Flat formula	Overstated by	Full / windowed / recurrent
4,096	0.25 GiB	1.00 GiB	4.00×	16 / 0 / 48
8,192	0.50 GiB	2.00 GiB	4.00×	16 / 0 / 48
16,384	1.00 GiB	4.00 GiB	4.00×	16 / 0 / 48
32,768	2.00 GiB	8.00 GiB	4.00×	16 / 0 / 48
65,536	4.00 GiB	16.00 GiB	4.00×	16 / 0 / 48
131,072	8.00 GiB	32.00 GiB	4.00×	16 / 0 / 48

48 of 64 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.

Compare with

same modality, comparable size

vs Qwen3-Coder-30B-A3B-Instruct vs Qwen3.6-27B vs Qwen3.8-27B vs gemma-4-12B-it-qat-q4_0-unquantized vs gemma-4-E4B-it vs Qwen3-30B-A3B-Thinking-2507 All alternatives →Will it fit on my card? →

Why other calculators give a different number

A parameters × bits ÷ 8 estimate puts Q4_K_M at roughly 0.00 GiB. The real file is 15.41 GiB, because a quantization is a mixture and some tensors are always kept at higher precision.

Architecture

from config.json

Layers

Attention heads

KV heads

Head dim

256

Hidden size

5120

Vocab

248,320

Sliding window

none

SWA period

—

MLA

Experts

—

Experts per token

—

use_sliding_window

—

Questions people ask

How much VRAM does MusaCoder-27B need?: Q4_K_M is exactly 16,547,399,264 bytes (15.41 GiB) in weights. Add the KV cache, which depends on your context length, plus roughly half a gigabyte of runtime overhead.
How large is MusaCoder-27B's KV cache?: 2.00 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 MusaCoder-27B 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.