Simple and efficient pytorch-native transformer text generation in <1000 LOC of python.
BSD 3-Clause "New" or "Revised" License
5.34k
stars
484
forks
source link
Infer cache/RoPE weight dtype from output weights #146
Closed
malfet closed 3 months ago
dtype
argument toprecompute_freqs_cis
This way one can change precision in one place in
generate.py
and it will be propagated throughout the model