upstage · text

solar-pro-preview-instruct

upstage/solar-pro-preview-instruct

solar-pro-preview-instruct at Q4_K_M is exactly 13,310,219,072 bytes (12.40 GiB / 13.31 GB) — an effective 4.809 bits per weight, not the nominal 4.

From the file· summed from 1 file(s)From the file· KV per layer
Parameters
22.1B
Architecture
llama
64 layers
Context
4,096
native (config.json)
License
—

Shipped quantizations

● exact bytes, summed from published files
QuantSize●Exact bytes●Effective bpw●Tensors●Publisher
IQ1_S4.46 GiB4,786,663,2321.730—MaziyarPanahi
IQ1_M4.87 GiB5,231,488,8321.890—MaziyarPanahi
IQ2_XS6.16 GiB6,618,394,4322.392—MaziyarPanahi
Q2_K7.65 GiB8,213,924,6722.968—MaziyarPanahi
IQ3_XS8.50 GiB9,125,197,6323.297—MaziyarPanahi
Q3_K_S8.92 GiB9,580,672,8323.462—MaziyarPanahi
Q3_K_M9.95 GiB10,686,592,8323.861579MaziyarPanahi
Q3_K_L10.84 GiB11,635,226,4324.204—MaziyarPanahi
IQ4_XS11.06 GiB11,878,032,1924.292579MaziyarPanahi
Q4_K_S11.73 GiB12,594,238,2724.551—MaziyarPanahi
Q4_K_M12.40 GiB13,310,219,0724.809579MaziyarPanahi
Q5_K_S14.20 GiB15,246,070,5925.509—MaziyarPanahi
Q5_K_M14.59 GiB15,663,862,5925.660579MaziyarPanahi
Q6_K16.92 GiB18,164,608,8326.564579MaziyarPanahi
Q8_021.91 GiB23,526,487,8728.501579MaziyarPanahi

KV cache by context

unresolved

This model declares a 2,047-token sliding window, but we could not establish which layers use it. Its architecture publishes the layout as a per-layer array inside the model file rather than as a period in config.json, and we have not yet ingested that array.

A flat context × layers × heads figure would be substantially too high, so we are not showing one. This is tracked as a known gap rather than filled with a guess.

Compare with

same modality, comparable size
vs Qwen3-Coder-30B-A3B-Instructvs gemma-4-12B-it-qat-q4_0-unquantizedvs Llama-3.2-1B-Instructvs Qwen-AgentWorld-35B-A3Bvs llama-3-youko-8bvs gpt-oss-20bAll alternatives →Will it fit on my card? →

Will it run on your card?

full quant x context sweep
By VRAM:8GB12GB16GB24GB32GB48GB
Radeon RX 6500 XT 4GBGeForce RTX 3050 6GBGeForce RTX 5050 8GBGeForce RTX 3080 10GBGeForce RTX 2080 Ti 11GBGeForce RTX 5070 12GBGeForce RTX 5060 Ti 16GBApple M3 Pro 18GBGeForce RTX 3080 Ti 20GBGeForce RTX 5090 D V2 24GBGeForce RTX 5090 32GBApple M5 Max 36GBApple M5 Max 48GBApple M5 Max 64GBApple M3 Ultra 96GBApple M5 Max 128GBApple M2 Ultra 192GBApple M3 Ultra 256GBApple M3 Ultra 512GB

Why other calculators give a different number

A parameters × bits ÷ 8 estimate puts Q4_K_M at roughly 11.60 GiB. The real file is 12.40 GiB, because a quantization is a mixture and some tensors are always kept at higher precision.

Architecture

from config.json
Layers
64
Attention heads
40
KV heads
10
Head dim
128
Hidden size
5120
Vocab
32,128
Sliding window
2047
SWA period
—
MLA
no
Experts
—
Experts per token
—
use_sliding_window
—

Questions people ask

How much VRAM does solar-pro-preview-instruct need?
Q4_K_M is exactly 13,310,219,072 bytes (12.40 GiB) in weights. Add the KV cache, which depends on your context length, plus roughly half a gigabyte of runtime overhead.
Which quantization of solar-pro-preview-instruct 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.