- 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
gemma-4-26B-A4B-it
Google
Apache 2.0
Open language model for chat and text. 5 sizes from 5.1B to 31.3B parameters, partly mixture-of-experts (only a part of the parameters is active per token).
- Mixture-of-experts: 25.8B parameters in total, of which 4.0B 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
1× H100 80 GB
The model needs about 17.4 GB of memory. Hopper, FP8 support.
- Weights
- 15.5 GB
- KV cache
- 0.4 GB
- Speed
- ~395.2 Tok./s
More suitable configurations
| Card | Quantity | Tok./s | Status |
|---|---|---|---|
| Tesla T4 16 GB | 2× | 63.3 | 2-3 weeks delivery time |
| Tesla P100 16 GB | 2× | 79.8 | 2-3 weeks delivery time |
| Tesla P40 24 GB | 1× | 80.7 | 2-3 weeks delivery time |
| Tesla V100 16 GB | 2× | 176.3 | only 1 in stock |
| Tesla V100 32 GB | 1× | 190.4 | On order, delivery time approx. 1 week |
| AMD Instinct MI50 16 GB | 2× | 135.2 | in stock |
| AMD Instinct MI50 32 GB | 1× | 146.0 | in stock |
| AMD Instinct MI60 32 GB | 1× | 134.8 | 2-3 weeks delivery time |
| AMD Instinct MI100 32 GB | 1× | 163.0 | 2-3 weeks delivery time |
| A30 24 GB | 1× | 184.3 | 2-3 weeks delivery time |
| RTX A5000 24 GB | 1× | 192.0 | 2-3 weeks delivery time |
| A40 48 GB | 1× | 183.1 | 2-3 weeks delivery time |
