Making LLMs faster without sacrificing accuracy
Imported from official source
A new scaling law that relates particular architectural choices to loss helps identify models that improve throughput by up to 47% with no loss of accuracy.
This version
- Version
- 1 of 1
- Recorded
- September 20, 2026 19:52
- Change
- Initial
- Content hash
96eb62c7507ca3d989da30cef81caf4d- All versions
- Revision history