jgrusewski 2d16926d71 docs: add BF16 training design — pure BF16 for all 10 models ...

Dynamic dtype detection (Ampere+ → BF16, else F32), zero casts in
training hot path, cast only at data ingestion and loss scalar.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

2026-03-03 15:56:14 +01:00

..

infra

feat(infra): GPU training launcher with local/cloud routing

2026-02-23 11:19:58 +01:00

plans

docs: add BF16 training design — pure BF16 for all 10 models

2026-03-03 15:56:14 +01:00