InfoCapability
FSDPv2 with FP8 quantization in v1.8.0
AI Impact Summary
New FSDPv2 with FP8 quantization capabilities released in v1.8.0, enabling more memory-efficient training of large models.
Hugging Face Accelerate library training infrastructure
Risk domains
675%670%
Source text
v1.8.0 combines Fully Sharded Data Parallel v2 with FP8 precision, enabling memory-efficient large model training.
- Date
- 19 Jun 2025
- Change type
- capability
- Severity
- info