Hy3-preview: hardware requirements
Hy3-preview needs about 185 GB of memory at Q4 quantization — a 180 GB download plus room for context and overhead. That is beyond consumer hardware; use it through an API instead.
This is a mixture-of-experts model. Only a fraction of its 299B parameters are used for each token, so it runs faster than its size suggests — but all of it still has to fit in memory. The speed is that of a small model; the memory requirement is that of a large one. More on this →
Will it run on your machine?
Download sizes by quantization
Lower quantization means a smaller file and less memory, at some cost to quality. Q4_K_M is the usual starting point. What is this?
| Quant | Download | RAM @ 4K | RAM @ 32K | Source |
|---|---|---|---|---|
| Q4_K_Mrecommended | 180 GB | 185 GB | 194 GB | estimated |
| Q5_K_M | 212 GB | 216 GB | 226 GB | estimated |
| Q6_K | 245 GB | 250 GB | 259 GB | estimated |
| Q8_0 | 317 GB | 322 GB | 331 GB | estimated |
| BF16 | 598 GB | 602 GB | 612 GB | estimated |
“Measured” sizes are read from published GGUF files. “Estimated” sizes are derived from the parameter count and are typically within a few percent. The RAM columns add the context cache and runtime overhead to the download size.