- 3 % discount on prepayment
- Insured shipping from Switzerland
- Sustainable: refurbished instead of new
- Warranty on all cards
- 3 % discount on prepayment
- Insured shipping from Switzerland
- Sustainable: refurbished instead of new
- Warranty on all cards
← Back to overview
Phi-4-reasoning-plus
Microsoft
MIT
Open language model for chat and text. 2 sizes from 3.8B to 14.7B parameters.
Settings
Quality
4 bit (Q4_K_M) is the most common choice: good quality with low memory requirements.
Context length: 8k tokens
4k32k
How much text the model keeps in view at once. More context needs more memory.
Recommendation
1× H100 80 GB
The model needs about 12.0 GB of memory. Hopper, FP8 support.
- Weights
- 8.8 GB
- KV cache
- 1.7 GB
- Speed
- ~105.9 Tok./s
More suitable configurations
| Card | Quantity | Tok./s | Status |
|---|---|---|---|
| Tesla T4 16 GB | 1× | 18.3 | 2-3 weeks delivery time |
| Tesla P100 16 GB | 1× | 23.1 | 2-3 weeks delivery time |
| Tesla P40 24 GB | 1× | 21.6 | 2-3 weeks delivery time |
| Tesla V100 16 GB | 1× | 51.0 | in stock |
| Tesla V100 32 GB | 1× | 51.0 | On order, delivery time approx. 1 week |
| AMD Instinct MI50 16 GB | 1× | 39.1 | in stock |
| AMD Instinct MI50 32 GB | 1× | 39.1 | in stock |
| AMD Instinct MI60 32 GB | 1× | 36.1 | 2-3 weeks delivery time |
| AMD Instinct MI100 32 GB | 1× | 43.7 | 2-3 weeks delivery time |
| A30 24 GB | 1× | 49.4 | 2-3 weeks delivery time |
| RTX A5000 24 GB | 1× | 51.5 | 2-3 weeks delivery time |
| A40 48 GB | 1× | 49.1 | 2-3 weeks delivery time |
