- 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
Kimi-Linear-48B-A3B-Instruct
Moonshot AI
MIT
Open language model for chat and text. 3 sizes from 49.1B to 2.8T parameters, partly mixture-of-experts (only a part of the parameters is active per token).
- Mixture-of-experts: 49.1B parameters in total, of which 3.0B 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
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 31.2 GB of memory. Hopper, FP8 support.
- Weights
- 29.5 GB
- KV cache
- 0.3 GB
- Speed
- ~530.8 Tok./s
More suitable configurations
| Card | Quantity | Tok./s | Status |
|---|---|---|---|
| Tesla T4 16 GB | 4× | 74.1 | 2-3 weeks delivery time |
| Tesla P100 16 GB | 4× | 93.3 | 2-3 weeks delivery time |
| Tesla P40 24 GB | 2× | 100.4 | 2-3 weeks delivery time |
| Tesla V100 16 GB | 4× | 206.3 | only 1 in stock |
| Tesla V100 32 GB | 2× | 236.8 | On order, delivery time approx. 1 week |
| AMD Instinct MI50 16 GB | 4× | 158.2 | only 2 in stock |
| AMD Instinct MI50 32 GB | 2× | 181.6 | in stock |
| AMD Instinct MI60 32 GB | 2× | 167.7 | 2-3 weeks delivery time |
| AMD Instinct MI100 32 GB | 2× | 202.7 | 2-3 weeks delivery time |
| A30 24 GB | 2× | 229.2 | 2-3 weeks delivery time |
| RTX A5000 24 GB | 2× | 238.8 | 2-3 weeks delivery time |
| A40 48 GB | 1× | 245.9 | 2-3 weeks delivery time |
