meta-llama · text

Llama-2-13b-chat-hf

meta-llama/Llama-2-13b-chat-hf

Llama-2-13b-chat-hf at Q4_K_M is exactly 7,865,956,224 bytes (7.33 GiB / 7.87 GB) — an effective 4.835 bits per weight, not the nominal 4. Its KV cache at 32K is 25.00 GiB.

From the file· summed from 1 file(s)From the file· KV from mirror (mirror:NousResearch/Llama-2-13b-chat-hf)
Parameters
13.0B
Architecture
llama
40 layers
Context
4,096
native (config.json)
License
llama2

Shipped quantizations

● exact bytes, summed from published files
QuantSize●Exact bytes●Effective bpw●Tensors●Publisher
Q2_K5.06 GiB5,429,348,2243.337—TheBloke
Q2_K5.06 GiB5,429,349,1843.337—second-state
Q3_K_S5.27 GiB5,658,980,2243.478—TheBloke
Q3_K_S5.27 GiB5,658,981,1843.478—second-state
Q3_K_M5.90 GiB6,337,769,3443.895—TheBloke
Q3_K_M5.90 GiB6,337,770,3043.895—second-state
Q3_K_L6.45 GiB6,929,559,4244.259—TheBloke
Q3_K_L6.45 GiB6,929,560,3844.259—second-state
Q4_06.86 GiB7,365,834,6244.527—TheBloke
Q4_06.86 GiB7,365,835,5844.527—second-state
Q4_K_S6.91 GiB7,414,331,2644.557—TheBloke
Q4_K_S6.91 GiB7,414,332,2244.557—second-state
Q4_K_M7.33 GiB7,865,956,2244.835—TheBloke
Q4_K_M7.33 GiB7,865,957,1844.835—second-state
Q5_K_S8.36 GiB8,972,285,8245.515—TheBloke
Q5_08.36 GiB8,972,285,8245.515—TheBloke
Q5_08.36 GiB8,972,286,7845.515—second-state
Q5_K_S8.36 GiB8,972,286,7845.515—second-state
Q5_K_M8.60 GiB9,229,924,2245.673—TheBloke
Q6_K9.95 GiB10,679,140,2246.564—TheBloke
Q6_K9.95 GiB10,679,141,1846.564—second-state
Q8_012.88 GiB13,831,319,4248.501—TheBloke
Q8_012.88 GiB13,831,320,3848.501—second-state
Q5_K_M2 shards17.19 GiB18,459,849,40811.346—second-state
F1624.25 GiB26,033,304,35216.001—second-state

KV cache by context

computed per layer
ContextKV cache (f16)●Flat formulaOverstated byFull / windowed / recurrent
4,0963.13 GiB3.13 GiB—40 / 0 / 0
8,1926.25 GiB6.25 GiB—40 / 0 / 0
16,38412.50 GiB12.50 GiB—40 / 0 / 0
32,76825.00 GiB25.00 GiB—40 / 0 / 0
65,53650.00 GiB50.00 GiB—40 / 0 / 0
131,072100.00 GiB100.00 GiB—40 / 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 6.82 GiB. The real file is 7.33 GiB, because a quantization is a mixture and some tensors are always kept at higher precision.

Architecture

from mirror:NousResearch/Llama-2-13b-chat-hf
Layers
40
Attention heads
40
KV heads
40
Head dim
128
Hidden size
5120
Vocab
32,000
Sliding window
none
SWA period
—
MLA
no
Experts
—
Experts per token
—
use_sliding_window
—

Questions people ask

How much VRAM does Llama-2-13b-chat-hf need?
Q4_K_M is exactly 7,865,956,224 bytes (7.33 GiB) in weights. Add the KV cache, which depends on your context length, plus roughly half a gigabyte of runtime overhead.
How large is Llama-2-13b-chat-hf's KV cache?
25.00 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 Llama-2-13b-chat-hf 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.