search

microsoft / BitNet

Official inference framework for 1-bit LLMs
MIT License
11.39k stars 768 forks source link

Memory leak in quantize_i2_s function #70

Closed asalaria-cisco closed 2 weeks ago

asalaria-cisco commented 1 month ago

Looking at the source code of quantize_i2_s, it seems that the buffer q8 is allocated on the heap but never freed. This would results in leaking memory each time the function is called.

Mujahid087 commented 1 month ago

yes you are right

paaspaas00 commented 1 month ago

54