baidu · text

ERNIE-4.5-300B-A47B-PT

baidu/ERNIE-4.5-300B-A47B-PT

ERNIE-4.5-300B-A47B-PT at Q4_K_M is exactly 180,155,744,256 bytes (167.78 GiB / 180.16 GB) — an effective 4.797 bits per weight, not the nominal 4. Its KV cache at 32K is 6.75 GiB.

From the file· summed from 4 file(s)From the file· KV per layer
Parameters
300B
Architecture
ernie4_5-moe
54 layers
Context
131,072
native (config.json)
License
apache-2.0

Shipped quantizations

● exact bytes, summed from published files
QuantSize●Exact bytes●Effective bpw●Tensors●Publisher
UD-TQ1_071.48 GiB76,750,552,6402.043—unsloth
UD-IQ1_S2 shards80.54 GiB86,483,435,2962.303—unsloth
UD-IQ1_M2 shards86.78 GiB93,179,117,3122.481—unsloth
UD-IQ2_XXS3 shards94.32 GiB101,273,599,9042.696—unsloth
UD-IQ2_M3 shards98.47 GiB105,736,339,3602.815—unsloth
Q2_K3 shards101.58 GiB109,067,796,3522.904—unsloth
Q2_K_L3 shards101.76 GiB109,266,370,4322.909—unsloth
UD-IQ3_XXS3 shards116.56 GiB125,152,296,8323.332—unsloth
Q3_K_S3 shards120.25 GiB129,116,635,0083.438—unsloth
Q3_K_M3 shards132.78 GiB142,569,471,8723.796—unsloth
IQ4_XS4 shards148.58 GiB159,541,395,4884.248—unsloth
IQ4_NL4 shards157.24 GiB168,832,303,1044.495—unsloth
Q4_04 shards157.55 GiB169,172,041,7284.504—unsloth
Q4_K_S4 shards157.93 GiB169,572,597,7604.515—unsloth
Q4_K_M4 shards167.78 GiB180,155,744,2564.797—unsloth
Q4_14 shards174.57 GiB187,438,759,9364.990—unsloth
Q5_K_S5 shards191.95 GiB206,101,840,0325.487—unsloth
Q5_K_M5 shards197.41 GiB211,964,297,4085.644—unsloth
Q6_K5 shards228.88 GiB245,760,884,8326.543—unsloth
Q8_07 shards296.43 GiB318,285,513,1528.474—unsloth
BF1613 shards557.89 GiB599,026,009,31215.949—unsloth

KV cache by context

computed per layer
ContextKV cache (f16)●Flat formulaOverstated byFull / windowed / recurrent
4,0960.84 GiB0.84 GiB—54 / 0 / 0
8,1921.69 GiB1.69 GiB—54 / 0 / 0
16,3843.38 GiB3.38 GiB—54 / 0 / 0
32,7686.75 GiB6.75 GiB—54 / 0 / 0
65,53613.50 GiB13.50 GiB—54 / 0 / 0
131,07227.00 GiB27.00 GiB—54 / 0 / 0

Compare with

same modality, comparable size
vs Qwen3-Coder-30B-A3B-Instructvs Qwen3.6-27Bvs Qwen3.8-27Bvs DeepSeek-V4-Flashvs Qwen3-30B-A3B-Thinking-2507vs Laguna-XS-2.1All 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 157.41 GiB. The real file is 167.78 GiB, because a quantization is a mixture and some tensors are always kept at higher precision.

Architecture

from config.json
Layers
54
Attention heads
64
KV heads
8
Head dim
128
Hidden size
8192
Vocab
103,424
Sliding window
none
SWA period
—
MLA
no
Experts
—
Experts per token
—
use_sliding_window
—

Questions people ask

How much VRAM does ERNIE-4.5-300B-A47B-PT need?
Q4_K_M is exactly 180,155,744,256 bytes (167.78 GiB) in weights. Add the KV cache, which depends on your context length, plus roughly half a gigabyte of runtime overhead.
How large is ERNIE-4.5-300B-A47B-PT's KV cache?
6.75 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 ERNIE-4.5-300B-A47B-PT 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.