Can OneJev 9B run on MacBook Neo A18 Pro 8GB?
What this means
The checked Q4_K_M build needs about 7.3GB, above this Mac's conservative 5GB working budget.
OneJev scores typed decisions in one forward pass. Generic decode estimates are not decision latency; image inputs need additional projector memory. Use the publisher's qev server and System One client.
Where the memory goes
| Component | Detail | GB |
|---|---|---|
| Model weights | Q4_K_M GGUF (5.6 GB) + mmap overhead | 5.9 |
| KV cache | 4K context window | 0.6 |
| Runtime | macOS inference app + compute buffers | 0.8 |
| Total needed | at Q4_K_M, 4K context | 7.3 |
| Working budget | 8 GB unified memory − conservative macOS reserve | 5 |
| Headroom | memory shortfall | −2.3 |
Pick your quant
| Quant | Download | Memory | Estimated speed | Verdict |
|---|---|---|---|---|
| Q2_K | 3.8 GB | 5.4 GB | — | ✕ Won't fit |
| Q3_K_M | 4.6 GB | 6.3 GB | — | ✕ Won't fit |
| IQ4_XS | 5.2 GB | 6.9 GB | — | ✕ Won't fit |
| Q4_K_S | 5.4 GB | 7.1 GB | — | ✕ Won't fit |
| Q4_K_M ★ | 5.6 GB | 7.3 GB | — | ✕ Won't fit |
| Q5_K_M | 6.5 GB | 8.3 GB | — | ✕ Won't fit |
| Q6_K | 7.4 GB | 9.2 GB | — | ✕ Won't fit |
| Q8_0 | 9.5 GB | 11.4 GB | — | ✕ Won't fit |
Other models on this Mac
FAQ
Can the MacBook Neo A18 Pro · 8GB run OneJev 9B?
The tracked Q4_K_M working memory is estimated at 7.3GB, excluding the vision projector. Weight fit is not verified workflow support; use the publisher's qev server and System One client.
Which OneJev 9B quant should I use on this Mac?
None of the tracked quants fit this configuration safely. Choose a smaller model or a Mac with more unified memory.
Which app should I use for OneJev 9B on this Mac?
OneJev 9B is a task-specific model, not a normal local chat download. The selected weights do not fit this Mac, and the model also needs its publisher's intended workflow. This page links that repository and recommends Llama 3.2 3B if you just want local chat.
Are these speeds measured on a MacBook Neo A18 Pro?
Decision latency has not been measured. OneJev scores typed choices in one forward pass; generic decode estimates do not measure this workflow.