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

examples/llama-4/do-no-use-fa2/maverick-qlora-fsdp1.yaml

@@ -66,7 +66,7 @@ gradient_checkpointing: offload
 gradient_checkpointing_kwargs:
   use_reentrant: false
 
-warmup_steps: 20
+warmup_ratio: 0.1
 evals_per_epoch: 1
 saves_per_epoch: 1
 weight_decay: 0.0

examples/llama-4/do-no-use-fa2/scout-qlora-fsdp1.yaml

@@ -69,7 +69,7 @@ tf32: true
 logging_steps: 1
 flash_attention: true
 
-warmup_steps: 100
+warmup_ratio: 0.1
 evals_per_epoch: 1
 saves_per_epoch: 1
 weight_decay: 0.0

examples/llama-4/do-no-use-fa2/scout-qlora-single-h100.yaml

@@ -76,7 +76,7 @@ gradient_checkpointing: offload
 gradient_checkpointing_kwargs:
   use_reentrant: false
 
-warmup_steps: 20
+warmup_ratio: 0.1
 evals_per_epoch: 1
 saves_per_epoch: 1
 weight_decay: 0.0

examples/llama-4/do-no-use-fa2/scout-vision-qlora-fsdp.yaml

@@ -65,7 +65,7 @@ tf32: true
 logging_steps: 1
 flash_attention: true
 
-warmup_steps: 100
+warmup_ratio: 0.1
 evals_per_epoch: 1
 saves_per_epoch: 1
 weight_decay: 0.0

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

examples/llama-4/scout-qlora-single-h100-flex.yaml

@@ -74,7 +74,7 @@ gradient_checkpointing_kwargs:
   use_reentrant: false
 
 logging_steps: 1
-warmup_steps: 20
+warmup_ratio: 0.1
 evals_per_epoch: 1
 saves_per_epoch: 1
 

examples/llama-4/scout-vision-qlora-fsdp2-flex.yaml

@@ -67,7 +67,7 @@ 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