Can it run?
Too heavy
Using Q4_K_M in vLLM
Fit status
Too heavy
Decode
3.4 tok/s
TTFT
56613 ms
Safe context
4K
Memory
472.9 GB / 69.1 GB
| Workload | Grade | Fit | Decode | TTFT | Context |
|---|---|---|---|---|---|
| Agentic Coding | F | Too heavy | 3.6 tok/s | 78714 ms | 5K |
| Chat | F | Too heavy | 3.6 tok/s | 29518 ms | 4K |
| Coding | F | Too heavy | 3.4 tok/s | 56613 ms | 4K |
| RAG | F | Too heavy | 3.6 tok/s | 98392 ms | 5K |
| Reasoning | F | Too heavy | 3.6 tok/s | 63955 ms | 4K |
How GLM-5 (744B params) fits at each quantization level on Mac Studio M3 Ultra 96GB (69.1 GB usable).
| Quant | Bits | VRAM | Quality | Fit |
|---|---|---|---|---|
Q2_K | 2 | 290.2 GB | Low | F0 |
Q3_K_S | 3 | 364.6 GB | Low | F0 |
NVFP4 | 4 |
| Medium |
| F0 |
Q4_K_M | 4 | 453.8 GB | Medium | F0 |
Q5_K_M | 5 | 535.7 GB | High | F0 |
Q6_K | 6 | 610.1 GB | High | F0 |
Q8_0 | 8 | 796.1 GB | Very High | F0 |
F16 | 16 | 1525.2 GB | Maximum | F0 |
