FITMYLLM · JULY 28, 2026 · OPEN METHODOLOGY · COMMUNITY BENCHMARKS
LIVE · UPDATED DAILY
Quantization Options Select your GPU for speed estimates Context length: 4K 8K 16K 32K 64K 128K
Quant Bits VRAM @ 16K Quality IQ2_XXS 2.38 92.4 GB
89.9 + 2.5 KV
low IQ2_M 2.93 113.1 GB
110.5 + 2.5 KV
low Q2_K 3.16 121.7 GB
119.2 + 2.5 KV
low IQ3_XXS 3.25 125.1 GB
122.6 + 2.5 KV
low IQ3_XS 3.5 134.5 GB
132.0 + 2.5 KV
low Q3_K_S 3.64 139.7 GB
137.2 + 2.5 KV
low IQ3_M 3.76 144.3 GB
141.7 + 2.5 KV
low Q3_K_M 4 153.3 GB
150.7 + 2.5 KV
low Q3_K_L 4.3 164.5 GB
162.0 + 2.5 KV
moderate IQ4_XS 4.46 170.5 GB
168.0 + 2.5 KV
moderate Q4_K_S 4.67 178.4 GB
175.9 + 2.5 KV
moderate Q4_K_M 4.89 186.7 GB
184.2 + 2.5 KV
good Q5_K_S 5.57 212.2 GB
209.7 + 2.5 KV
good Q5_K_M 5.7 217.1 GB
214.6 + 2.5 KV
good Q6_K 6.56 249.4 GB
246.9 + 2.5 KV
excellent Q8_0 8.5 322.3 GB
319.8 + 2.5 KV
lossless FP16 16 604.0 GB
601.5 + 2.5 KV
lossless
Select your GPU above to see speed estimates and compatibility for each quantization.
Too big for a single GPU — plan a multi-GPU deployment
Even the lightest quant needs ~92 GB. Size GPUs, replicas, TCO and scaling for a production setup. Open in Enterprise →
▸ READY TO RUN THIS? RENT BY THE HOUR
RENT A GPU AND RUN ERNIE-4.5-300B-A47B-PADDLE NOW
Spin up an A100 / H100 / 4090 in ~60s. Pay by the second. Cancel anytime.
Community Ratings Chat Coding Reasoning Creative Vision Roleplay Agentic
Loading ratings...
Run this model IQ2_XXS — 89.9 GB VRAM IQ2_M — 110.5 GB VRAM Q2_K — 119.2 GB VRAM IQ3_XXS — 122.6 GB VRAM IQ3_XS — 132.0 GB VRAM Q3_K_S — 137.2 GB VRAM IQ3_M — 141.7 GB VRAM Q3_K_M — 150.7 GB VRAM Q3_K_L — 162.0 GB VRAM IQ4_XS — 168.0 GB VRAM Q4_K_S — 175.9 GB VRAM Q4_K_M — 184.2 GB VRAM Q5_K_S — 209.7 GB VRAM Q5_K_M — 214.6 GB VRAM Q6_K — 246.9 GB VRAM Q8_0 — 319.8 GB VRAM FP16 — 601.5 GB VRAM
Ollama llama.cpp vLLM LM Studio KoboldCpp Jan Docker
▸ Easiest way to get started · Beginners
DOCS ↗ curl -fsSL https://ollama.com/install.sh | shCOPY
$ ollama run ernie:301b-q4_K_MCOPY
Tag may need adjustment — check ollama.com/library/ernie for available tags.
▸ SETUP GUIDE >_
Auto-setup with fitmyllm CLI Detects your GPU, recommends the best model, downloads it, and starts chatting — zero config. Benchmarks your speed and contributes anonymous data to improve predictions.
Auto-detect GPU Live tok/s in chat Speed benchmarks 9 inference engines
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-Paddle
Build Hardware for ERNIE-4.5-300B-A47B-Paddle ▸ COLOPHON FITMYLLM · INDEPENDENT · DATA-DRIVEN
FITMYLLM · EST. 2025 · © 2026
RECOMMENDATIONS FROM PUBLISHED MATH, CORRECTED BY THE COMMUNITY — 30.
▸ SPEC SHEET
ERNIE-4.5-300B-A47B-Paddle — 300.5B Dense. ▸ SPECIFICATIONS
PARAMETERS 300.5B
ARCHITECTURE Dense Transformer
CONTEXT LENGTH 128K tokens
CAPABILITIES chat
RELEASE DATE 2025-06-28
PROVIDER Baidu
FAMILY ernie ▸ VRAM REQUIREMENTS
QUANT BPW VRAM QUALITY IQ2_XXS 2.38 89.9 GB 65% IQ2_M 2.93 110.5 GB 75% Q2_K 3.16 119.2 GB 78% IQ3_XXS 3.25 122.6 GB 82% IQ3_XS 3.5 132.0 GB 84% Q3_K_S 3.64 137.2 GB 85% IQ3_M 3.76 141.7 GB 86% Q3_K_M 4 150.7 GB 88% Q3_K_L 4.3 162.0 GB 90% IQ4_XS 4.46 168.0 GB 92% Q4_K_S 4.67 175.9 GB 93% Q4_K_M 4.89 184.2 GB 94% Q5_K_S 5.57 209.7 GB 96% Q5_K_M 5.7 214.6 GB 96% Q6_K 6.56 246.9 GB 97% Q8_0 8.5 319.8 GB 100% FP16 16 601.5 GB 100%
§ 01 BENCHMARK SCORES
MMLU-PRO 77.6
GPQA Diamond 74.0
aa_ifbench 39.1
aa_terminal_bench 6.1
aa_scicode 31.5
aa_lcr 2.3
LiveCodeBench 46.7
AIME 41.3
MATH-500 93.1
HLE 3.5
AA Intelligence 15.0
AA Coding 14.5
AA Math 41.3
§ 02 RUN COMMAND
Run ERNIE-4.5-300B-A47B-Paddle locally with Ollama — needs 184.2 GB VRAM at Q4_K_M:
$ ollama run ernie:300b
§ 03 COMPATIBLE GPUs
13 @ Q4_K_M Feedback