SvalTek · text

MQ-Coldbrew-Base

SvalTek/MQ-Coldbrew-Base

MQ-Coldbrew-Base at Q4_K_M is exactly 4,681,090,784 bytes (4.36 GiB / 4.68 GB) — an effective 4.919 bits per weight, not the nominal 4. Its KV cache at 32K is 1.75 GiB.

From the file· summed from 1 file(s)From the file· KV per layer
Parameters
7.6B
Architecture
qwen2
28 layers
Context
131,072
native (config.json)
License

Shipped quantizations

exact bytes, summed from published files
QuantSizeExact bytesEffective bpwTensorsPublisher
I1-IQ1_S1.77 GiB1,902,208,4801.999mradermacher
I1-IQ1_M1.90 GiB2,040,737,2482.144mradermacher
I1-IQ2_XXS2.12 GiB2,271,618,5282.387mradermacher
I1-IQ2_XS2.30 GiB2,467,562,9762.593mradermacher
I1-IQ2_S2.42 GiB2,594,033,6642.726mradermacher
I1-IQ2_M2.59 GiB2,778,738,6882.920mradermacher
I1-Q2_K_S2.64 GiB2,832,425,4402.977mradermacher
Q2_K2.81 GiB3,014,291,6803.168mradermacher
I1-Q2_K2.81 GiB3,014,291,9363.168mradermacher
I1-IQ3_XXS2.90 GiB3,112,910,8483.271mradermacher
I1-IQ3_XS3.11 GiB3,344,462,8483.515mradermacher
Q3_K_S3.25 GiB3,490,575,1043.668mradermacher
I1-Q3_K_S3.25 GiB3,490,575,3603.668mradermacher
I1-IQ3_S3.26 GiB3,497,399,2963.675mradermacher
I1-IQ3_M3.33 GiB3,572,218,8803.754mradermacher
Q3_K_M3.55 GiB3,806,597,8884.000mradermacher
I1-Q3_K_M3.55 GiB3,806,598,1444.000mradermacher
Q3_K_L3.81 GiB4,086,665,9844.295mradermacher
I1-Q3_K_L3.81 GiB4,086,666,2404.295mradermacher
I1-IQ4_XS3.93 GiB4,216,534,5284.431mradermacher
IQ4_XS3.96 GiB4,248,360,1924.465mradermacher
I1-IQ4_NL4.13 GiB4,435,830,7524.662mradermacher
I1-Q4_04.14 GiB4,442,138,5924.668mradermacher
Q4_K_S4.15 GiB4,455,786,2084.682mradermacher
I1-Q4_K_S4.15 GiB4,455,786,4644.682mradermacher
Q4_K_M4.36 GiB4,681,090,7844.919mradermacher
I1-Q4_K_M4.36 GiB4,681,091,0404.919mradermacher
I1-Q4_14.54 GiB4,871,211,9365.119mradermacher
Q5_K_S4.95 GiB5,313,015,3925.583mradermacher
I1-Q5_K_S4.95 GiB5,313,015,6485.583mradermacher
Q5_K_M5.07 GiB5,442,670,1765.720mradermacher
I1-Q5_K_M5.07 GiB5,442,670,4325.720mradermacher
Q6_K5.82 GiB6,251,848,2886.570mradermacher
I1-Q6_K5.82 GiB6,251,848,5446.570mradermacher
Q8_07.54 GiB8,095,483,7448.507mradermacher
F1614.19 GiB15,232,137,18416.007mradermacher

KV cache by context

computed per layer
ContextKV cache (f16)Flat formulaOverstated byFull / windowed / recurrent
4,0960.22 GiB0.22 GiB28 / 0 / 0
8,1920.44 GiB0.44 GiB28 / 0 / 0
16,3840.88 GiB0.88 GiB28 / 0 / 0
32,7681.75 GiB1.75 GiB28 / 0 / 0
65,5363.50 GiB3.50 GiB28 / 0 / 0
131,0727.00 GiB7.00 GiB28 / 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 3.99 GiB. The real file is 4.36 GiB, because a quantization is a mixture and some tensors are always kept at higher precision.

Architecture

from config.json
Layers
28
Attention heads
28
KV heads
4
Head dim
128
Hidden size
3584
Vocab
151,666
Sliding window
none
SWA period
MLA
no
Experts
Experts per token
use_sliding_window
false

Questions people ask

How much VRAM does MQ-Coldbrew-Base need?
Q4_K_M is exactly 4,681,090,784 bytes (4.36 GiB) in weights. Add the KV cache, which depends on your context length, plus roughly half a gigabyte of runtime overhead.
How large is MQ-Coldbrew-Base's KV cache?
1.75 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 MQ-Coldbrew-Base 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.