|
Mila
Deep Neural Network Library
|
What a deployment would cost, and how it would prefill. More...
Public Attributes | |
| MemoryStats | memory |
| PrefillChunking | prefill |
What a deployment would cost, and how it would prefill.
Fitting in memory and running well are different questions, and a caller choosing a context length needs both: the largest context that fits can be one where the prefill chunk has walked down to its floor. The two answers come from one graph construction because they are two readings of the same arithmetic.