• 3 % discount on prepayment
  • Insured shipping from Switzerland
  • Sustainable: refurbished instead of new
  • Warranty on all cards
← Back to overview

LFM2.5-8B-A1B

Liquid AI
GermanLfm1.0

Open language model for chat and text. 2 sizes from 2.7B to 8.5B parameters, partly mixture-of-experts (only a part of the parameters is active per token).

  • Mixture-of-experts: 8.5B parameters in total, of which 1.0B 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
4k64k
How much text the model keeps in view at once. More context needs more memory.
Recommendation
1× H100 80 GB

The model needs about 6.7 GB of memory. Hopper, FP8 support.

Weights
5.1 GB
KV cache
0.1 GB
Speed
~1'573.0 Tok./s

More suitable configurations

CardQuantityTok./sStatus
Tesla T4 16 GB1×272.22-3 weeks delivery time
Tesla P100 16 GB1×342.92-3 weeks delivery time
Tesla P40 24 GB1×321.42-3 weeks delivery time
Tesla V100 16 GB1×758.1in stock
Tesla V100 32 GB1×758.1On order, delivery time approx. 1 week
AMD Instinct MI50 16 GB1×581.2in stock
AMD Instinct MI50 32 GB1×581.2in stock
AMD Instinct MI60 32 GB1×536.62-3 weeks delivery time
AMD Instinct MI100 32 GB1×648.92-3 weeks delivery time
A30 24 GB1×733.52-3 weeks delivery time
RTX A5000 24 GB1×764.22-3 weeks delivery time
A40 48 GB1×728.92-3 weeks delivery time