Massive update for fine-tuning @GoogleDeepMind Gemma 3! 3.6x faster fine-tuning and 50% memory reduction with Axolotl v0.8.0!! - Multimodal support Gemma3 models - Support for Sequence Parallelism, Cut-cross entropy & liger kernels - Up to 3.6x speedups over vanilla HF+FA2 - https://t.co/OkTmtZmq3f
— Philipp Schmid (@_philschmid) Apr 15, 2025
from Twitter https://twitter.com/_philschmid
April 15, 2025 at 07:23AM
via IFTTT
No comments:
Post a Comment