🔌
New: connect Claude & other AIs to GenAIList over MCP — research the catalog and contribute to the shared knowledge base. Learn how →
// RUN — FITS
Can the Apple M3 Max 36GB run DeBERTa?
Yes — here's how fast and up to what context, with real measured numbers.
DeBERTa on the Apple M3 Max 36GB
| Quant | Fits? | Max context | Speed | Beyond context |
|---|---|---|---|---|
| FP16 | yes | 426k | 68-82 t/s | ~2.3-2.7 (slow) |
| Q8 | yes | 449k | 136-164 t/s | ~4.5-5.5 (slow) |
| Q4 | yes | 459k | 242.9-292.9 t/s | ~8.1-9.8 (slow) |
Estimate (memory-bound). Beyond "max context" the KV cache spills to SSD swap and speed drops to the "beyond" figure.
FAQ
Can the Apple M3 Max 36GB run DeBERTa?▶
Yes. At Q4 it generates ~242.9-292.9 tok/s and fits up to 459k context; beyond that it spills to SSD swap and slows to ~8.1-9.8 tok/s.
How much context fits?▶
At Q4, up to about 459k tokens stay in fast memory; longer context spills to SSD swap and throughput collapses.