use warmup_ratio as a better default than warmup steps since it's data dependent (#2897) [skip ci]

* use warmup_ratio as a better default than warmup steps since it's data dependent * replace remainder of warmup_steps
2025-07-30 06:44:06 -04:00
parent 2eb7ff95af
commit 22810c97b7
99 changed files with 100 additions and 100 deletions
--- a/examples/llama-4/scout-qlora-flexattn-fsdp2.yaml
+++ b/examples/llama-4/scout-qlora-flexattn-fsdp2.yaml
@@ -64,7 +64,7 @@ flex_attn_compile_kwargs:
  dynamic: false
  mode: max-autotune-no-cudagraphs

-warmup_steps: 10
+warmup_ratio: 0.1
 evals_per_epoch: 1
 saves_per_epoch: 1
 weight_decay: 0.0