ernie/Mixture of Experts

EERNIE 4.5 300B-A47B

chatcodingreasoningmath
300B
Parameters (47B active)
128K
Context length
8
Benchmarks
17
Quantizations
Architecture
MoE
Released
2025-06-28
Layers
54
KV Heads
8
Head Dim
128
Family
ernie

Quantization Options

QuantBitsVRAMQuality
IQ2_XXS2.3889.7 GBlow
IQ2_M2.93110.4 GBlow
Q2_K3.16119.0 GBlow
IQ3_XXS3.25122.4 GBlow
IQ3_XS3.5131.7 GBlow
Q3_K_S3.64137.0 GBlow
IQ3_M3.76141.5 GBlow
Q3_K_M4150.5 GBlow
Q3_K_L4.3161.7 GBmoderate
IQ4_XS4.46167.7 GBmoderate
Q4_K_S4.67175.6 GBmoderate
Q4_K_M4.89183.9 GBgood
Q5_K_S5.57209.4 GBgood
Q5_K_M5.7214.2 GBgood
Q6_K6.56246.5 GBexcellent
Q8_08.5319.2 GBlossless
FP1616600.5 GBlossless

Select your GPU above to see speed estimates and compatibility for each quantization.

Benchmarks (8)

MATH-50096.4
BBH94.3
HumanEval92.1
IFEval88.0
MMLU-PRO78.4
MUSR69.9
AIME54.8
LiveCodeBench38.8

Run this model

Easiest way to get starteddocs →
curl -fsSL https://ollama.com/install.sh | sh
$ollama run ernie:300b-q4_K_M

Tag may need adjustment — check ollama.com/library/ernie for available tags.

Setup guide

GPUs that can run this model

At Q4_K_M quantization. Sorted by minimum VRAM.

Find the best GPU for ERNIE 4.5 300B-A47B

Build Hardware for ERNIE 4.5 300B-A47B

ERNIE 4.5 300B-A47B300B Parameter Mixture of Experts LLM

Model Specifications

Parameters
300B (47B active)
Architecture
Mixture of Experts
Context Length
128K tokens
Capabilities
chat, coding, reasoning, math
Release Date
2025-06-28
Family
ernie

VRAM Requirements

QuantizationBPWVRAMQuality
IQ2_XXS2.3889.7 GB65%
IQ2_M2.93110.4 GB75%
Q2_K3.16119.0 GB78%
IQ3_XXS3.25122.4 GB82%
IQ3_XS3.5131.7 GB84%
Q3_K_S3.64137.0 GB85%
IQ3_M3.76141.5 GB86%
Q3_K_M4150.5 GB88%
Q3_K_L4.3161.7 GB90%
IQ4_XS4.46167.7 GB92%
Q4_K_S4.67175.6 GB93%
Q4_K_M4.89183.9 GB94%
Q5_K_S5.57209.4 GB96%
Q5_K_M5.7214.2 GB96%
Q6_K6.56246.5 GB97%
Q8_08.5319.2 GB100%
FP1616600.5 GB100%

Benchmark Scores

HumanEval92.1
MMLU-PRO78.4
IFEval88.0
BBH94.3
MUSR69.9
LiveCodeBench38.8
AIME54.8
MATH-50096.4

Compatible GPUs (13)

GPUs that can run ERNIE 4.5 300B-A47B at Q4_K_M quantization: