Segmentation fault (core dumped)
Opened this issue · 3 comments
SachinVashisth commented
The bug
Hi! I am getting the message Segmentation fault (core dumped)
while running the following code.
To Reproduce
from guidance import models, gen
llama3 = models.Transformers("meta-llama/Meta-Llama-3-8B-Instruct", device_map = 'auto')
llama3 + f'Do you want a joke or a poem? ' + gen(stop='.')
System info (please complete the following information):
guidance version is 0.1.15
riedgar-ms commented
What version of LlamaCpp are you on? And does this happen if you run on the CPU rather than the GPU?
SachinVashisth commented
What version of LlamaCpp are you on? And does this happen if you run on the CPU rather than the GPU?
Hi
The version of llama_cpp_python
is 0.2.77.
The same issue occurs when I use the CPU rather than the GPU.
riedgar-ms commented
And the prompt is fine when run directly from LlamaCpp (sorry, we see a lot of segfaults from LlamaCpp, and a segfault can't be from our code, which is 100% Python).