ViperEkura
5244f1a8fc
feat: add te-style scaled fp8 training via fp8_autocast
- per-tensor scales applied inside cublasLt via A_SCALE/B_SCALE
- delayed scaling: weight amax history ring, refresh every 16 steps
- quantize kernels emit atomic amax, device-side scale updates
- fp8_autocast context toggles aten::linear dispatch like torch.autocast
- fallback to bf16 when M/N not 16-aligned (fp8 gemm constraint)
- x/g scales delayed one step, reuse free atomic amax (no abs/max reduce)
2026-08-14 12:14:04 +08:00
..
2026-08-09 23:12:53 +08:00
2026-08-09 20:23:58 +08:00
2026-08-09 20:52:06 +08:00
2026-08-09 20:52:06 +08:00
2026-08-09 23:12:53 +08:00
2026-08-09 23:12:53 +08:00
2026-08-10 08:40:18 +08:00
2026-08-09 20:23:58 +08:00
2026-08-09 20:52:06 +08:00
2026-08-09 23:12:53 +08:00
2026-08-10 08:40:18 +08:00
2026-08-10 08:40:18 +08:00
2026-08-09 20:52:06 +08:00
2026-07-27 00:35:34 +08:00
2026-08-14 12:14:04 +08:00
2026-08-02 13:20:14 +08:00