ERNIE 4.5 21B-A3B on LattePanda Sigma (32 GB)

Can the LattePanda Sigma (32 GB) run ERNIE 4.5 21B-A3B?

Yes, it runs well (tight fit)

Runs well (tight fit). At Q4_K_M it needs about 15.7 GB of 16.0 GB usable, with an estimated speed of 28.9 tok/s.

Best quant: Q4_K_M
Generation speed: 28.9 tok/s
Prompt processing: 678 tok/s
First token (1K prompt): 1.5 s
Largest safe context: 12K

Best app to run it on the LattePanda Sigma (32 GB)

On Windows

llama.cpp

Command line + local API · runs on llama.cpp

Supports Intel Arc through SYCL and Vulkan.
Usually the first to support new model architectures.

llama-server -hf unsloth/ERNIE-4.5-21B-A3B-PT-GGUF:Q4_K_M

LM Studio Point-and-click: search, download and chat in one window.
Jan Point-and-click: search, download and chat in one window.

On Linux

llama.cpp

Command line + local API · runs on llama.cpp

Supports Intel Arc through SYCL and Vulkan.
Usually the first to support new model architectures.

llama-server -hf unsloth/ERNIE-4.5-21B-A3B-PT-GGUF:Q4_K_M

LM Studio Point-and-click: search, download and chat in one window.
Jan Point-and-click: search, download and chat in one window.

How we choose: the files published for this model, your operating system and GPU, whether it fits fully in memory, and published speed tests (MLX vs llama.cpp on Apple silicon, vLLM vs llama.cpp vs Ollama on NVIDIA). Details.

Download ERNIE 4.5 21B-A3B

Recommended for the LattePanda Sigma (32 GB): Q4_K_M. Every other quantization has its own download link in the table below.

Quantized files · GGUF

Works in LM Studio, llama.cpp, Jan and Ollama. No sign-in needed.

Download the Q4_K_M file (opens in a new tab)

huggingface.co/unsloth/ERNIE-4.5-21B-A3B-PT-GGUF (opens in a new tab)

ollama run hf.co/unsloth/ERNIE-4.5-21B-A3B-PT-GGUF:Q4_K_M

Original weights · Hugging Face

Full-precision files from Baidu, for fine-tuning, vLLM, Transformers.

huggingface.co/baidu/ERNIE-4.5-21B-A3B-PT (opens in a new tab)

No sign-in needed.

Official repo · GitHub

Reference code, inference scripts and docs from Baidu. The weights themselves download from Hugging Face.

github.com/PaddlePaddle/ERNIE (opens in a new tab)

git clone https://github.com/PaddlePaddle/ERNIE

Links checked 3 days ago against Hugging Face, Ollama and GitHub.

How it’s rated

Quality 60Good quality (B) · 60 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 79/100 (quality, fit, speed and memory use) sets the ranking.: How capable the model itself is at Q4_K_M, from its editorial quality tier scaled by the quantization or precision. It’s the same on any hardware that runs this variant. Scale: A+ 85+ · A 70+ · B 55+ · C 40+ · D under 40.
Runs Runs welltight fit: How well it runs on the LattePanda Sigma (32 GB): speed, capped by how well it fits in memory. Runs great: 30+ tokens/s (or near the image, video or decision speed target) · Runs well: 10+ tokens/s · Slow: 3+ tokens/s · Very slow: under 3 · Won’t run: doesn’t fit even with offload. A tight fit or RAM offload caps it at Runs well; heavy offload caps it at Slow.

Match score

To order lists, each result also gets a match score out of 100 from four parts. It favors the most capable model that still runs well.

Model qualityEditorial quality tier of the model, scaled by the quantization or precision used.	27.4 / 45
Memory fitFull marks when the model fits in the share of unified memory the GPU can use; fewer when it goes past that limit.	24.0 / 25
SpeedGeneration speed against a 40 tokens/s target for LLMs, or render time against a target for image and video.	12.7 / 15
Memory useRewards using the memory you have, so a bigger, better variant beats a tiny one that leaves most memory idle.	15.0 / 15
Match score	79.1 / 100

The match score orders results; it is not shown as a grade. Quality and Runs are the two ratings in the tables.

Every quantization on the LattePanda Sigma (32 GB)

Quant	Memory needed	Available	Speed	Runs	Quality	Download
`Q2_K`	11.1 GB	16.0 GB	45.1 tok/s	Runs great	44Fair quality (C) · 44 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 75/100 (quality, fit, speed and memory use) sets the ranking.	GGUF (opens in a new tab)
`Q3_K_M`	13.1 GB	16.0 GB	36.5 tok/s	Runs great	53Fair quality (C) · 53 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 78/100 (quality, fit, speed and memory use) sets the ranking.	GGUF (opens in a new tab)
`Q4_K_M`	15.7 GB	16.0 GB	28.9 tok/s	Runs welltight fit	60Good quality (B) · 60 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 79/100 (quality, fit, speed and memory use) sets the ranking.	GGUF (opens in a new tab)
`Q5_K_M`	17.9 GB	16.0 GB	20.6 tok/s	Slowfar over GPU memory limit	62Good quality (B) · 62 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 57/100 (quality, fit, speed and memory use) sets the ranking.	GGUF (opens in a new tab)
`Q6_K`	20.3 GB	16.0 GB	—	Won't run	63Good quality (B) · 63 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 8/100 (quality, fit, speed and memory use) sets the ranking.	GGUF (opens in a new tab)
`Q8_0`	25.6 GB	16.0 GB	—	Won't run	64Good quality (B) · 64 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 8/100 (quality, fit, speed and memory use) sets the ranking.	GGUF (opens in a new tab)
`F16`	46.2 GB	16.0 GB	—	Won't run	64Good quality (B) · 64 of 100: how capable the model is at F16, whatever it runs on. Match score 8/100 (quality, fit, speed and memory use) sets the ranking.	GGUF (opens in a new tab) Original weights (opens in a new tab)

Memory figures assume 8K of context. Longer conversations or documents grow the KV cache.

Other hardware that runs it

64Good quality (B) · 64 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 83/100 (quality, fit, speed and memory use) sets the ranking. Apple M4 Pro (48 GB) Q8_0 · 41.3 tok/s · Runs great
64Good quality (B) · 64 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 83/100 (quality, fit, speed and memory use) sets the ranking. NVIDIA GeForce RTX 5090 Q8_0 · 154 tok/s · Runs great
63Good quality (B) · 63 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 83/100 (quality, fit, speed and memory use) sets the ranking. AMD Radeon RX 7900 XTX Q6_K · 127 tok/s · Runs great
63Good quality (B) · 63 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 83/100 (quality, fit, speed and memory use) sets the ranking. NVIDIA GeForce RTX 3090 Q6_K · 136 tok/s · Runs great
63Good quality (B) · 63 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 83/100 (quality, fit, speed and memory use) sets the ranking. NVIDIA GeForce RTX 4090 Q6_K · 141 tok/s · Runs great
64Good quality (B) · 64 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 82/100 (quality, fit, speed and memory use) sets the ranking. Apple M1 Max (64 GB) Q8_0 · 56.4 tok/s · Runs great

Quick answers

How much memory does ERNIE 4.5 21B-A3B need on the LattePanda Sigma 32GB?

At Q4_K_M it needs about 15.7 GB. The LattePanda Sigma 32GB has 16.0 GB usable.

How fast is ERNIE 4.5 21B-A3B on the LattePanda Sigma 32GB?

About 28.9 tok/s at Q4_K_M, an estimate from the hardware’s memory bandwidth and the size of the model.

Which quantization of ERNIE 4.5 21B-A3B should I use on the LattePanda Sigma 32GB?

Q4_K_M. Of the versions that fit, it is the one that runs best; when two run equally well, the check prefers the higher-rated one.

Estimates come from published hardware specs and our open formulas, not from running the model on your machine. Real speed varies with runtime, drivers and settings. Numbers marked “Measured” come from a linked benchmark.