3 ms·
It sounds like Cerebras would be perfect for models with Mamba architecture, as those don't need a large KV cache for long contexts.
by cubefox 1y ago
It sounds like Cerebras would be perfect for models with Mamba architecture, as those don't need a large KV cache for long contexts.