For finetuning workflow, Linkedin's just launched Liger Kernel will immediately give you - increase multi-GPU training throughput by 20% and - reduces memory usage by 60%. And you need to a single line of code change `apply_liger_kernel_to_llama()` https://t.co/MTIYvyZcBu
— Rohan Paul (@rohanpaul_ai) Aug 24, 2024
from Twitter https://twitter.com/rohanpaul_ai
August 24, 2024 at 03:04AM
via IFTTT
No comments:
Post a Comment