GLM-5.2: use fp8 KV cache on Hopper (Blackwell/AMD keep fp8_e4m3) (#566)
Signed-off-by: Tiezhen WANG <38108242+xianbaoqian@users.noreply.github.com> Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
T
Tiezhen WANG committed
6d0eae529e75e657e10eb0e6eb33c4ec6f8e16ca
Parent: 490e06b
Committed by GitHub <noreply@github.com>
on 6/18/2026, 9:41:58 AM