internlm · text

internlm2-math-plus-20b

internlm/internlm2-math-plus-20b

internlm2-math-plus-20b at Q4_K_M is exactly 11,984,469,440 bytes (11.16 GiB / 11.98 GB) — an effective 4.827 bits per weight, not the nominal 4. Its KV cache at 32K is 6.00 GiB.

From the file· summed from 1 file(s)From the file· KV per layer
Parameters
19.9B
Architecture
internlm2
48 layers
Context
8,192
native (config.json)
License
other

Shipped quantizations

● exact bytes, summed from published files
QuantSize●Exact bytes●Effective bpw●Tensors●Publisher
IQ1_S4.23 GiB4,543,268,2881.830—lmstudio-community
IQ1_S4.23 GiB4,543,268,2881.830—legraphista
I1-IQ1_S4.23 GiB4,543,269,0241.830—mradermacher
IQ1_M4.58 GiB4,918,396,3521.981—legraphista
IQ1_M4.58 GiB4,918,396,3521.981—lmstudio-community
I1-IQ1_M4.58 GiB4,918,397,0881.981—mradermacher
IQ2_XXS5.16 GiB5,543,609,7922.233—legraphista
IQ2_XXS5.16 GiB5,543,609,7922.233—lmstudio-community
I1-IQ2_XXS5.16 GiB5,543,610,5282.233—mradermacher
IQ2_XS5.68 GiB6,100,403,6482.457—lmstudio-community
IQ2_XS5.68 GiB6,100,403,6482.457—legraphista
I1-IQ2_XS5.68 GiB6,100,404,3842.457—mradermacher
IQ2_S6.03 GiB6,474,296,7682.608—legraphista
IQ2_S6.03 GiB6,474,296,7682.608—lmstudio-community
I1-IQ2_S6.03 GiB6,474,297,5042.608—mradermacher
IQ2_M6.50 GiB6,974,467,5202.809—legraphista
IQ2_M6.50 GiB6,974,467,5202.809—lmstudio-community
I1-IQ2_M6.50 GiB6,974,468,2562.809—mradermacher
Q2_K_S6.53 GiB7,013,469,6322.825—legraphista
Q2_K7.03 GiB7,546,670,5283.040—lmstudio-community
Q2_K7.03 GiB7,546,670,5283.040—legraphista
I1-Q2_K7.03 GiB7,546,671,2643.040—mradermacher
IQ3_XXS7.28 GiB7,814,376,8963.148—legraphista
IQ3_XXS7.28 GiB7,814,376,8963.148—lmstudio-community
I1-IQ3_XXS7.28 GiB7,814,377,6323.148—mradermacher
IQ3_XS7.79 GiB8,361,752,0003.368—lmstudio-community
IQ3_XS7.79 GiB8,361,752,0003.368—legraphista
I1-IQ3_XS7.79 GiB8,361,752,7363.368—mradermacher
Q3_K_S8.16 GiB8,760,473,0243.529—legraphista
Q3_K_S8.16 GiB8,760,473,0243.529—lmstudio-community
I1-Q3_K_S8.16 GiB8,760,473,7603.529—mradermacher
IQ3_S8.20 GiB8,800,581,0563.545—legraphista
IQ3_S8.20 GiB8,800,581,0563.545—lmstudio-community
I1-IQ3_S8.20 GiB8,800,581,7923.545—mradermacher
IQ3_M8.50 GiB9,121,445,3123.674—legraphista
IQ3_M8.50 GiB9,121,445,3123.674—lmstudio-community
I1-IQ3_M8.50 GiB9,121,446,0483.674—mradermacher
Q3_K9.05 GiB9,722,279,3603.916—legraphista
Q3_K_M9.05 GiB9,722,279,3603.916—lmstudio-community
I1-Q3_K_M9.05 GiB9,722,280,0963.916—mradermacher

KV cache by context

computed per layer
ContextKV cache (f16)●Flat formulaOverstated byFull / windowed / recurrent
4,0960.75 GiB0.75 GiB—48 / 0 / 0
8,1921.50 GiB1.50 GiB—48 / 0 / 0
16,3843.00 GiB3.00 GiB—48 / 0 / 0
32,7686.00 GiB6.00 GiB—48 / 0 / 0
65,53612.00 GiB12.00 GiB—48 / 0 / 0
131,07224.00 GiB24.00 GiB—48 / 0 / 0

Compare with

same modality, comparable size
vs Qwen3-Coder-30B-A3B-Instructvs Qwen3.6-27Bvs Qwen3.8-27Bvs gemma-4-12B-it-qat-q4_0-unquantizedvs gemma-4-E4B-itvs Qwen3-30B-A3B-Thinking-2507All 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 D 32GBApple M5 Max 36GBApple M5 Pro 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 10.40 GiB. The real file is 11.16 GiB, because a quantization is a mixture and some tensors are always kept at higher precision.

Architecture

from config.json
Layers
48
Attention heads
48
KV heads
8
Head dim
128
Hidden size
6144
Vocab
92,544
Sliding window
none
SWA period
—
MLA
no
Experts
—
Experts per token
—
use_sliding_window
—

Questions people ask

How much VRAM does internlm2-math-plus-20b need?
Q4_K_M is exactly 11,984,469,440 bytes (11.16 GiB) in weights. Add the KV cache, which depends on your context length, plus roughly half a gigabyte of runtime overhead.
How large is internlm2-math-plus-20b's KV cache?
6.00 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 internlm2-math-plus-20b 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.
internlm2-math-plus-20b — VRAM requirements, exact quant sizes — ossmodeldb