Is ERNIE-4.5-300B-A47B-PT or DeepSeek-V4-Flash smaller?

Both publish quantizations; compare the table for the exact sizes.

Which uses less memory at long context?

DeepSeek-V4-Flash — its KV cache at 32K context is 107.2x smaller, because of how its attention layers cache.

Model comparison

ERNIE-4.5-300B-A47B-PT vs DeepSeek-V4-Flash

These two publish different quantization sets; the table below has the exact sizes. At long context the gap widens: DeepSeek-V4-Flash's KV cache at 32K is 107.2× smaller, which usually matters more than the difference in weights.

From the file· summed bytes, KV per layer

Side by side

	ERNIE-4.5-300B-A47B-PT	DeepSeek-V4-Flash
Parameters	300B	291B
Architecture	ernie4_5-moe	deepseek4
Layers	54	43
Native context	131,072	1,048,576
Mixture of experts	no	yes, 256 experts
Quantizations published	21	12
Smallest quantization	71.48 GiB	76.87 GiB
Q4_K_M	167.78 GiB	—
Licence	apache-2.0	mit

KV cache by context

the term that decides long-context viability

Context	ERNIE-4.5-300B-A47B-PT	DeepSeek-V4-Flash	Ratio
4,096	0.84 GiB	0.06 GiB	13.40×
8,192	1.69 GiB	0.06 GiB	26.79×
16,384	3.38 GiB	0.06 GiB	53.58×
32,768	6.75 GiB	0.06 GiB	107.16×
65,536	13.50 GiB	0.06 GiB	214.33×
131,072	27.00 GiB	0.06 GiB	428.65×