Can it run?
Runs well
Using Q4_K_M in vLLM
Fit status
Runs well
Decode
77.8 tok/s
TTFT
2487 ms
Safe context
37K
Memory
125.4 GB / 288.0 GB
| Workload | Grade | Fit | Decode | TTFT | Context |
|---|---|---|---|---|---|
| Agentic Coding | C | Runs well | 77.8 tok/s | 3618 ms | 64K |
| Chat | C | Runs well | 77.8 tok/s | 1357 ms | 20K |
| Coding | C | Runs well | 77.8 tok/s | 2487 ms | 37K |
| RAG | C | Runs well | 77.8 tok/s | 4522 ms | 64K |
| Reasoning | C | Runs well | 77.8 tok/s | 2940 ms | 37K |
How Devstral 2 123B Instruct (123B params) fits at each quantization level on AMD Instinct MI350X 288GB (288.0 GB usable).
| Quant | Bits | VRAM | Quality | Fit |
|---|---|---|---|---|
Q2_K | 2 | 48.0 GB | Low | D33 |
Q3_K_S | 3 | 60.3 GB | Low | D34 |
NVFP4 | 4 | 68.9 GB | Medium | D35 |
Q4_K_M | 4 | 75.0 GB | Medium | D35 |
Q5_K_M | 5 | 88.6 GB | High | D36 |
Q6_K | 6 | 100.9 GB | High | D37 |
Q8_0Best for your GPU | 8 | 131.6 GB | Very High | D39 |
F16 | 16 | 252.2 GB | Maximum | C45 |
huggingface-cli download devstral-2-123b