Jean Senellart

possibility to use CPU in parallel training (-gpuid 0,1) and to use several time the same devices (-gpuid 7,7) probability we will need to deprecate/restrict `fallback_to_cpu` which meaningless in parallel...

adapt the script prune.lua that was developped for seq2seq (see https://github.com/harvardnlp/seq2seq-attn/issues/91).

quick and dirty implementation - for reading quantized weights. * for lookup tables, rows are converted on the fly back to float * for linear weights, the only adjustment is...

Memory access errors

7

Hello @demisjohn ! When running the test cases on ARM - I have found several memory access errors using clang AddressSanitizer that do not seem to be related at all...

Currently Perceval uses float64 as a default for storing parameters. For simulation with increasing number of modes/photons, this induces a natural 2x slowdown for numerical precision beyond any QPU native...

Heralds out of their box

5

**Describe the bug** Perceval has a renderer at the `Circuit` or the `Processor` level allowing to generate svg or png through matplotlib library. At the processor level, it is possible...