- 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
Llama-4-Maverick-17B-128E-Instruct
Meta
GermanLlama 4 Community
Open language model for chat and text. 2 sizes from 108.6B to 401.6B parameters, partly mixture-of-experts (only a part of the parameters is active per token).
- Mixture-of-experts: 401.6B parameters in total, of which 17.0B active per token. Memory requirement depends on all parameters, speed on the active ones.
Settings
Quality
4 bit (Q4_K_M) is the most common choice: good quality with low memory requirements.
Context length: 8k tokens
4k256k
How much text the model keeps in view at once. More context needs more memory.
Recommendation
4× H100 80 GB
The model needs about 244.1 GB of memory. Hopper, FP8 support.
- Weights
- 241.0 GB
- KV cache
- 1.6 GB
- Speed
- ~75.0 Tok./s
More suitable configurations
| Card | Quantity | Tok./s | Status |
|---|---|---|---|
| A40 48 GB | 8× | 27.6 | 2-3 weeks delivery time |
| RTX A6000 48 GB | 8× | 30.9 | 2-3 weeks delivery time |
| Quadro RTX 8000 48 GB | 8× | 22.8 | 2-3 weeks delivery time |
| L40 48 GB | 8× | 33.8 | 2-3 weeks delivery time |
| A100 40 GB | 8× | 34.1 | 2-3 weeks delivery time |
| A100 80 GB | 4× | 53.4 | 2-3 weeks delivery time |
| H100 80 GB | 4× | 75.0 | only 1 in stock |
