← CanIRunLocal · All hardware · All models

Can I run Hy-Memory 30B-A3B (MoE, memory) locally?

Tencent · 30.06B parameters (3B active) · memory.

Ollama weight files: Q4_K_M 18.24 GB (hf.co/tencent/Hy-Memory-30B-A3B-GGUF:Q4_K_M, size from the Hugging Face file listing); Q8_0 31.99 GB (hf.co/tencent/Hy-Memory-30B-A3B-GGUF:Q8_0, size from the Hugging Face file listing).

Hy-Memory-30B-A3B is a long-term-memory writer (~3B active) for Mem0, MemOS and Hy-Memory. It decides what to keep and how new evidence should update a memory bank. It is not a general chat model, so the Chat, Coding, Reasoning and Vision filters leave it out.

Yes on 64 machines in this list. The smallest is RTX 3060 12GB: Q4_K_M, about 19.6 GB, roughly 21 tokens per second, via Ollama. ollama run hf.co/tencent/Hy-Memory-30B-A3B-GGUF:Q4_K_M

• DGX Spark 128GB: Q8_0, ~33.3 GB, fits in fast memory, ~49 tok/s — ollama run hf.co/tencent/Hy-Memory-30B-A3B-GGUF:Q8_0
• Mac Studio M5 Max 128GB: Q8_0, ~33.3 GB, fits in fast memory, ~77 tok/s — ollama run hf.co/tencent/Hy-Memory-30B-A3B-GGUF:Q8_0
• Mac Studio M5 Ultra 256GB: Q8_0, ~33.3 GB, fits in fast memory, ~150 tok/s — ollama run hf.co/tencent/Hy-Memory-30B-A3B-GGUF:Q8_0

Setup on DGX Spark 128GB (NVIDIA, ASUS, Dell, Lenovo, HP, MSI, Acer, Gigabyte)

• Ollama: ollama run hf.co/tencent/Hy-Memory-30B-A3B-GGUF:Q8_0
• LM Studio: lms get tencent/Hy-Memory-30B-A3B-GGUF@Q8_0
• llama.cpp: llama-server -hf tencent/Hy-Memory-30B-A3B-GGUF:Q8_0 -c 8192
• vLLM: vllm serve tencent/Hy-Memory-30B-A3B --max-model-len 8192 --gpu-memory-utilization 0.9

Hardware that fits

Estimates, not benchmarks. Model sizes are ollama.com download sizes; memory needed adds headroom for ~8k tokens of context; speeds are rough figures from memory bandwidth. Data updated 2026-10-20.

A project by OnDevice Labs · llms.txt