What are the hardware requirements for ERNIE 4.5 21B A3B PT?

ERNIE 4.5 21B A3B PT requires approximately 44 GB of VRAM and supports a context window of 128K tokens. Quantized variants may run on less VRAM; see the Quantized Models section on this page.

Who developed ERNIE 4.5 21B A3B PT and how large is it?

ERNIE 4.5 21B A3B PT is developed by unsloth, a model with 21b parameters. The model is published as open weights on Hugging Face and indexed on LLM Explorer with full benchmark history.

How does ERNIE 4.5 21B A3B PT perform on standard benchmarks?

ERNIE 4.5 21B A3B PT has the following published scores: LMSYS ELO 1468. Compare against reference models on this page or on the LLM Explorer leaderboards.

ERNIE 4.5 21B A3B PT by unsloth — VRAM 44GB, 128K context

Name: ERNIE 4.5 21B A3B PT
Author: unsloth

ERNIE 4.5 21B A3B PT is an open-source language model by unsloth. Features: 21b LLM, VRAM: 44GB, Context: 128K, License: apache-2.0, LLM Explorer Score: 0.34, ELO: 1468.

Base model:baidu/ernie-4.5-21b... Base model:finetune:baidu/erni... Conversational En Endpoints compatible Ernie4.5 Ernie4 5 moe Region:us Safetensors Sharded Tensorflow Unsloth Zh

Model Card on HF 🤗: https://huggingface.co/unsloth/ERNIE-4.5-21B-A3B-PT

ERNIE 4.5 21B A3B PT Benchmarks

LMSys ELO 1468 vs 1272 (so35) 15.4%

^nn.n% — How the model compares to the reference models: Anthropic Sonnet 3.5 ("so35"), GPT-4o ("gpt4o") or GPT-4 ("gpt4").

What is the LLM Explorer Rank (Score)

ERNIE 4.5 21B A3B PT Parameters and Internals

LLM Name	ERNIE 4.5 21B A3B PT
Repository 🤗	https://huggingface.co/unsloth/ERNIE-4.5-21B-A3B-PT
Base Model(s)	baidu/ERNIE-4.5-21B-A3B-PT baidu/ERNIE-4.5-21B-A3B-PT
Model Size	21b
Required VRAM	44 GB
Updated	2026-08-10
Maintainer	unsloth
Model Type	ernie4_5_moe
Model Files	5.0 GB: 1-of-9 5.0 GB: 2-of-9 5.0 GB: 3-of-9 5.0 GB: 4-of-9 5.0 GB: 5-of-9 5.0 GB: 6-of-9 5.0 GB: 7-of-9 5.0 GB: 8-of-9 4.0 GB: 9-of-9
Supported Languages	en zh
Model Architecture	Ernie4_5_MoeForCausalLM
License	apache-2.0
Context Length	131072
Model Max Length	131072
Transformers Version	4.57.6
Tokenizer Class	LlamaTokenizer
Padding Token	<unk>
Vocabulary Size	103424
Torch Data Type	bfloat16

Rank the ERNIE 4.5 21B A3B PT capabilities

Have you tried this model? Rate its performance — this feedback helps the ML community find the right model for their needs.

Instruction Following and Task Automation
Factuality and Completeness of Knowledge
Censorship and Alignment
Data Analysis and Insight Generation
Text Generation
Text Summarization and Feature Extraction
Code Generation
Multi-Language Support and Translation

Best Alternatives to ERNIE 4.5 21B A3B PT

Best Alternatives	Context / RAM	Downloads	Likes
ERNIE 4.5 21B A3B PT	128K / 44 GB	50007	175
... Claude 4.5 High OPUS Thinking	128K / 43.7 GB	24	3
ERNIE 4.5 21B A3B Thinking	128K / 43.7 GB	14750	786
ERNIE 4.5 21B A3B Base PT	128K / 43.9 GB	788	36
ERNIE 4.5 21B A3B Thinking	128K / 43.7 GB	110	11
ERNIE 4.5 21B A3B Paddle	128K / 44 GB	47	18
ERNIE 4.5 21B A3B Base Paddle	128K / 44.2 GB	38	15
ERNIE 4.5 21B A3B PT Bf16	128K / 43.7 GB	19	1
ERNIE 4.5 21B A3B PT 8bit	128K / 23.2 GB	36	2

Note: green Score (e.g. "73.2") means that the model is better than unsloth/ERNIE-4.5-21B-A3B-PT.