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/archived/cerebras/btlm-ft.yml
+++ b/examples/archived/cerebras/btlm-ft.yml
@@ -66,7 +66,7 @@ flash_optimum:
 gptq_groupsize:
 gptq_model_v1:
-warmup_steps: 32
+warmup_ratio: 0.1
 evals_per_epoch: 4
 saves_per_epoch: 1
 save_total_limit:
--- a/examples/archived/cerebras/qlora.yml
+++ b/examples/archived/cerebras/qlora.yml
@@ -43,7 +43,7 @@ xformers_attention: true
 flash_attention:
 gptq_groupsize:
 gptq_model_v1:
-warmup_steps: 10
+warmup_ratio: 0.1
 evals_per_epoch: 4
 saves_per_epoch: 1
 weight_decay: 0.1
--- a/examples/archived/code-llama/13b/lora.yml
+++ b/examples/archived/code-llama/13b/lora.yml
@@ -47,7 +47,7 @@ resume_from_checkpoint:
 logging_steps: 1
 flash_attention: true
-warmup_steps: 10
+warmup_ratio: 0.1
 evals_per_epoch: 4
 saves_per_epoch: 1
 weight_decay: 0.0
--- a/examples/archived/code-llama/13b/qlora.yml
+++ b/examples/archived/code-llama/13b/qlora.yml
@@ -48,7 +48,7 @@ resume_from_checkpoint:
 logging_steps: 1
 flash_attention: true
-warmup_steps: 10
+warmup_ratio: 0.1
 evals_per_epoch: 4
 saves_per_epoch: 1
 weight_decay: 0.0
--- a/examples/archived/code-llama/34b/lora.yml
+++ b/examples/archived/code-llama/34b/lora.yml
@@ -47,7 +47,7 @@ resume_from_checkpoint:
 logging_steps: 1
 flash_attention: true
-warmup_steps: 10
+warmup_ratio: 0.1
 evals_per_epoch: 4
 saves_per_epoch: 1
 weight_decay: 0.0
--- a/examples/archived/code-llama/34b/qlora.yml
+++ b/examples/archived/code-llama/34b/qlora.yml
@@ -48,7 +48,7 @@ resume_from_checkpoint:
 logging_steps: 1
 flash_attention: true
-warmup_steps: 10
+warmup_ratio: 0.1
 evals_per_epoch: 4
 saves_per_epoch: 1
 weight_decay: 0.0
--- a/examples/archived/code-llama/7b/lora.yml
+++ b/examples/archived/code-llama/7b/lora.yml
@@ -47,7 +47,7 @@ resume_from_checkpoint:
 logging_steps: 1
 flash_attention: true
-warmup_steps: 10
+warmup_ratio: 0.1
 evals_per_epoch: 4
 saves_per_epoch: 1
 weight_decay: 0.0
--- a/examples/archived/code-llama/7b/qlora.yml
+++ b/examples/archived/code-llama/7b/qlora.yml
@@ -48,7 +48,7 @@ resume_from_checkpoint:
 logging_steps: 1
 flash_attention: true
-warmup_steps: 10
+warmup_ratio: 0.1
 evals_per_epoch: 4
 saves_per_epoch: 1
 weight_decay: 0.0
--- a/examples/archived/dbrx/16bit-lora.yaml
+++ b/examples/archived/dbrx/16bit-lora.yaml
@@ -54,7 +54,7 @@ resume_from_checkpoint:
 logging_steps: 1
 flash_attention: true
-warmup_steps: 10
+warmup_ratio: 0.1
 evals_per_epoch:
 saves_per_epoch: 1
--- a/examples/archived/dbrx/8bit-lora.yaml
+++ b/examples/archived/dbrx/8bit-lora.yaml
@@ -57,7 +57,7 @@ resume_from_checkpoint:
 logging_steps: 1
 flash_attention: true
-warmup_steps: 10
+warmup_ratio: 0.1
 evals_per_epoch:
 saves_per_epoch: 1
--- a/examples/archived/dbrx/fft-ds-zero3.yaml
+++ b/examples/archived/dbrx/fft-ds-zero3.yaml
@@ -41,7 +41,7 @@ resume_from_checkpoint:
 logging_steps: 1
 flash_attention: true
-warmup_steps: 10
+warmup_ratio: 0.1
 evals_per_epoch:
 saves_per_epoch: 1
--- a/examples/archived/deepcoder/deepcoder-14B-preview-lora.yml
+++ b/examples/archived/deepcoder/deepcoder-14B-preview-lora.yml
@@ -51,7 +51,7 @@ resume_from_checkpoint:
 logging_steps: 1
 flash_attention: true
-warmup_steps: 10
+warmup_ratio: 0.1
 evals_per_epoch: 1
 saves_per_epoch: 1
 weight_decay: 0.0
--- a/examples/archived/falcon/config-7b-lora.yml
+++ b/examples/archived/falcon/config-7b-lora.yml
@@ -47,7 +47,7 @@ xformers_attention: true
 flash_attention:
 gptq_groupsize:
 gptq_model_v1:
-warmup_steps: 40
+warmup_ratio: 0.1
 evals_per_epoch: 4
 saves_per_epoch: 1
 weight_decay: 0.0
--- a/examples/archived/falcon/config-7b-qlora.yml
+++ b/examples/archived/falcon/config-7b-qlora.yml
@@ -77,7 +77,7 @@ xformers_attention: true
 flash_attention:
 gptq_groupsize:
 gptq_model_v1:
-warmup_steps: 10
+warmup_ratio: 0.1
 evals_per_epoch: 4
 saves_per_epoch: 1
 weight_decay: 0.000001
--- a/examples/archived/falcon/config-7b.yml
+++ b/examples/archived/falcon/config-7b.yml
@@ -44,7 +44,7 @@ xformers_attention: true
 flash_attention:
 gptq_groupsize:
 gptq_model_v1:
-warmup_steps: 40
+warmup_ratio: 0.1
 evals_per_epoch: 4
 saves_per_epoch: 1
 weight_decay: 0.0
--- a/examples/archived/gptj/qlora.yml
+++ b/examples/archived/gptj/qlora.yml
@@ -40,7 +40,7 @@ xformers_attention: true
 flash_attention:
 gptq_groupsize:
 gptq_model_v1:
-warmup_steps: 10
+warmup_ratio: 0.1
 evals_per_epoch: 4
 saves_per_epoch: 1
 weight_decay: 0.1
--- a/examples/archived/jeopardy-bot/config.yml
+++ b/examples/archived/jeopardy-bot/config.yml
@@ -41,7 +41,7 @@ xformers_attention: true
 flash_attention:
 gptq_groupsize:
 gptq_model_v1:
-warmup_steps: 20
+warmup_ratio: 0.1
 evals_per_epoch: 4
 saves_per_epoch: 1
 weight_decay: 0.1
--- a/examples/archived/mpt-7b/config.yml
+++ b/examples/archived/mpt-7b/config.yml
@@ -42,7 +42,7 @@ logging_steps: 5
 flash_attention:
 gptq_groupsize:
 gptq_model_v1:
-warmup_steps: 20
+warmup_ratio: 0.1
 evals_per_epoch: 4
 saves_per_epoch: 1
 weight_decay: 0.0001
--- a/examples/archived/openllama-3b/config.yml
+++ b/examples/archived/openllama-3b/config.yml
@@ -42,7 +42,7 @@ logging_steps: 1
 flash_attention: true
 gptq_groupsize:
 gptq_model_v1:
-warmup_steps: 20
+warmup_ratio: 0.1
 evals_per_epoch: 4
 saves_per_epoch: 1
 weight_decay: 0.1
--- a/examples/archived/openllama-3b/lora.yml
+++ b/examples/archived/openllama-3b/lora.yml
@@ -50,7 +50,7 @@ logging_steps: 1
 flash_attention: true
 gptq_groupsize:
 gptq_model_v1:
-warmup_steps: 20
+warmup_ratio: 0.1
 evals_per_epoch: 4
 saves_per_epoch: 1
 weight_decay: 0.1
--- a/examples/archived/openllama-3b/qlora.yml
+++ b/examples/archived/openllama-3b/qlora.yml
@@ -43,7 +43,7 @@ logging_steps: 1
 flash_attention: true
 gptq_groupsize:
 gptq_model_v1:
-warmup_steps: 20
+warmup_ratio: 0.1
 evals_per_epoch: 4
 saves_per_epoch: 1
 weight_decay: 0.1
--- a/examples/archived/qwen/lora.yml
+++ b/examples/archived/qwen/lora.yml
@@ -49,7 +49,7 @@ resume_from_checkpoint:
 logging_steps: 1
 flash_attention:
-warmup_steps: 10
+warmup_ratio: 0.1
 evals_per_epoch: 4
 saves_per_epoch: 1
 weight_decay: 0.0
--- a/examples/archived/qwen/qlora.yml
+++ b/examples/archived/qwen/qlora.yml
@@ -49,7 +49,7 @@ resume_from_checkpoint:
 logging_steps: 1
 flash_attention:
-warmup_steps: 10
+warmup_ratio: 0.1
 evals_per_epoch: 4
 saves_per_epoch: 1
 weight_decay: 0.0
--- a/examples/archived/qwen/qwen2-moe-lora.yaml
+++ b/examples/archived/qwen/qwen2-moe-lora.yaml
@@ -45,7 +45,7 @@ resume_from_checkpoint:
 logging_steps: 1
 flash_attention: true
-warmup_steps: 10
+warmup_ratio: 0.1
 evals_per_epoch: 4
 saves_per_epoch: 1
 weight_decay: 0.0
--- a/examples/archived/qwen/qwen2-moe-qlora.yaml
+++ b/examples/archived/qwen/qwen2-moe-qlora.yaml
@@ -48,7 +48,7 @@ resume_from_checkpoint:
 logging_steps: 1
 flash_attention: true
-warmup_steps: 10
+warmup_ratio: 0.1
 evals_per_epoch: 4
 saves_per_epoch: 1
 weight_decay: 0.0
--- a/examples/archived/redpajama/config-3b.yml
+++ b/examples/archived/redpajama/config-3b.yml
@@ -43,7 +43,7 @@ logging_steps: 5
 flash_attention:
 gptq_groupsize:
 gptq_model_v1:
-warmup_steps: 20
+warmup_ratio: 0.1
 evals_per_epoch: 4
 saves_per_epoch: 1
 weight_decay: 0.0001
--- a/examples/archived/replit-3b/config-lora.yml
+++ b/examples/archived/replit-3b/config-lora.yml
@@ -41,7 +41,7 @@ logging_steps: 1
 flash_attention:
 gptq_groupsize:
 gptq_model_v1:
-warmup_steps: 20
+warmup_ratio: 0.1
 evals_per_epoch: 4
 saves_per_epoch: 1
 weight_decay: 0
--- a/examples/archived/stablelm-2/1.6b/fft.yml
+++ b/examples/archived/stablelm-2/1.6b/fft.yml
@@ -50,7 +50,7 @@ flash_attn_rms_norm: true
 flash_attn_fuse_qkv: false
 flash_attn_fuse_mlp: true
-warmup_steps: 100
+warmup_ratio: 0.1
 evals_per_epoch: 4
 saves_per_epoch: 1
--- a/examples/archived/stablelm-2/1.6b/lora.yml
+++ b/examples/archived/stablelm-2/1.6b/lora.yml
@@ -51,7 +51,7 @@ flash_attention: true
 flash_attn_cross_entropy: false
 flash_attn_rms_norm: true
-warmup_steps: 10
+warmup_ratio: 0.1
 evals_per_epoch: 4
 saves_per_epoch: 1
 weight_decay: 0.0
--- a/examples/archived/starcoder2/qlora.yml
+++ b/examples/archived/starcoder2/qlora.yml
@@ -48,7 +48,7 @@ resume_from_checkpoint:
 logging_steps: 1
 flash_attention: true
-warmup_steps: 20
+warmup_ratio: 0.1
 evals_per_epoch: 4
 eval_steps:
 saves_per_epoch: 4
--- a/examples/archived/tiny-llama/lora-mps.yml
+++ b/examples/archived/tiny-llama/lora-mps.yml
@@ -49,7 +49,7 @@ resume_from_checkpoint:
 logging_steps: 1
 flash_attention: false
-warmup_steps: 10
+warmup_ratio: 0.1
 evals_per_epoch: 0
 saves_per_epoch: 1
 weight_decay: 0.0
--- a/examples/archived/tiny-llama/lora.yml
+++ b/examples/archived/tiny-llama/lora.yml
@@ -47,7 +47,7 @@ resume_from_checkpoint:
 logging_steps: 1
 flash_attention: true
-warmup_steps: 10
+warmup_ratio: 0.1
 evals_per_epoch: 4
 saves_per_epoch: 1
 weight_decay: 0.0
--- a/examples/archived/tiny-llama/pretrain.yml
+++ b/examples/archived/tiny-llama/pretrain.yml
@@ -38,7 +38,7 @@ resume_from_checkpoint:
 logging_steps: 1
 flash_attention: true
-warmup_steps: 10
+warmup_ratio: 0.1
 evals_per_epoch:
 saves_per_epoch: 1
 weight_decay: 0.0
--- a/examples/archived/tiny-llama/qlora.yml
+++ b/examples/archived/tiny-llama/qlora.yml
@@ -49,7 +49,7 @@ resume_from_checkpoint:
 logging_steps: 1
 flash_attention: true
-warmup_steps: 10
+warmup_ratio: 0.1
 evals_per_epoch: 4
 saves_per_epoch: 1
 weight_decay: 0.0
--- a/examples/archived/xgen-7b/xgen-7b-8k-qlora.yml
+++ b/examples/archived/xgen-7b/xgen-7b-8k-qlora.yml
@@ -75,7 +75,7 @@ xformers_attention: true
 flash_attention:
 gptq_groupsize:
 gptq_model_v1:
-warmup_steps: 10
+warmup_ratio: 0.1
 evals_per_epoch: 4
 saves_per_epoch: 1
 weight_decay: 0.0
--- a/examples/archived/yi-34B-chat/qlora.yml
+++ b/examples/archived/yi-34B-chat/qlora.yml
@@ -20,7 +20,7 @@ special_tokens:
 datasets:
  - path: mhenrichsen/alpaca_2k_test
    type: alpaca
-warmup_steps: 10
+warmup_ratio: 0.1
 # Iterations
 num_epochs: 1
--- a/examples/deepcogito/cogito-v1-preview-llama-3B-lora.yml
+++ b/examples/deepcogito/cogito-v1-preview-llama-3B-lora.yml
@@ -51,7 +51,7 @@ resume_from_checkpoint:
 logging_steps: 1
 flash_attention: true
-warmup_steps: 10
+warmup_ratio: 0.1
 evals_per_epoch: 1
 saves_per_epoch: 1
 weight_decay: 0.0
--- a/examples/deepcogito/cogito-v1-preview-qwen-14B-lora.yml
+++ b/examples/deepcogito/cogito-v1-preview-qwen-14B-lora.yml
@@ -51,7 +51,7 @@ resume_from_checkpoint:
 logging_steps: 1
 flash_attention: true
-warmup_steps: 10
+warmup_ratio: 0.1
 evals_per_epoch: 1
 saves_per_epoch: 1
 weight_decay: 0.0
--- a/examples/deepseek-v2/fft-fsdp-16b.yaml
+++ b/examples/deepseek-v2/fft-fsdp-16b.yaml
@@ -37,7 +37,7 @@ resume_from_checkpoint:
 logging_steps: 1
 flash_attention: true
-warmup_steps: 100
+warmup_ratio: 0.1
 evals_per_epoch: 2
 saves_per_epoch: 1
 weight_decay: 0.0
--- a/examples/deepseek-v2/qlora-fsdp-2_5.yaml
+++ b/examples/deepseek-v2/qlora-fsdp-2_5.yaml
@@ -61,7 +61,7 @@ resume_from_checkpoint:
 logging_steps: 1
 flash_attention: true
-warmup_steps: 100
+warmup_ratio: 0.1
 evals_per_epoch: 2
 saves_per_epoch: 1
 weight_decay: 0.0
--- a/examples/glm4/qlora-32b.yaml
+++ b/examples/glm4/qlora-32b.yaml
@@ -55,7 +55,7 @@ flash_attention: true
 loss_watchdog_threshold: 5.0
 loss_watchdog_patience: 3
-warmup_steps: 10
+warmup_ratio: 0.1
 evals_per_epoch: 1
 saves_per_epoch: 1
 weight_decay: 0.0
--- a/examples/jamba/qlora.yaml
+++ b/examples/jamba/qlora.yaml
@@ -49,7 +49,7 @@ resume_from_checkpoint:
 logging_steps: 1
 flash_attention: true
-warmup_steps: 10
+warmup_ratio: 0.1
 evals_per_epoch:
 saves_per_epoch: 1
 weight_decay: 0.0
--- a/examples/jamba/qlora_deepspeed.yaml
+++ b/examples/jamba/qlora_deepspeed.yaml
@@ -48,7 +48,7 @@ resume_from_checkpoint:
 logging_steps: 1
 flash_attention: true
-warmup_steps: 10
+warmup_ratio: 0.1
 evals_per_epoch:
 saves_per_epoch: 1
--- a/examples/jamba/qlora_fsdp_large.yaml
+++ b/examples/jamba/qlora_fsdp_large.yaml
@@ -47,7 +47,7 @@ gradient_checkpointing_kwargs:
 logging_steps: 1
 flash_attention: true
-warmup_steps: 10
+warmup_ratio: 0.1
 evals_per_epoch: 1
 saves_per_epoch: 1
 weight_decay: 0.0
--- a/examples/llama-2/fft_optimized.yml
+++ b/examples/llama-2/fft_optimized.yml
@@ -48,7 +48,7 @@ flash_attn_rms_norm: true
 flash_attn_fuse_qkv: false
 flash_attn_fuse_mlp: true
-warmup_steps: 100
+warmup_ratio: 0.1
 evals_per_epoch: 4
 saves_per_epoch: 1
--- a/examples/llama-2/gptq-lora.yml
+++ b/examples/llama-2/gptq-lora.yml
@@ -56,7 +56,7 @@ logging_steps: 1
 flash_attention:
 sdp_attention:
 flash_optimum:
-warmup_steps: 100
+warmup_ratio: 0.1
 evals_per_epoch: 4
 saves_per_epoch: 1
 weight_decay: 0.1
--- a/examples/llama-2/lisa.yml
+++ b/examples/llama-2/lisa.yml
@@ -52,7 +52,7 @@ flash_attn_rms_norm: true
 flash_attn_fuse_qkv: false
 flash_attn_fuse_mlp: true
-warmup_steps: 100
+warmup_ratio: 0.1
 evals_per_epoch: 4
 saves_per_epoch: 1
 weight_decay: 0.1
--- a/examples/llama-2/loftq.yml
+++ b/examples/llama-2/loftq.yml
@@ -47,7 +47,7 @@ resume_from_checkpoint:
 logging_steps: 1
 flash_attention: true
-warmup_steps: 10
+warmup_ratio: 0.1
 evals_per_epoch: 4
 saves_per_epoch: 1
 weight_decay: 0.0
--- a/examples/llama-2/lora.yml
+++ b/examples/llama-2/lora.yml
@@ -47,7 +47,7 @@ resume_from_checkpoint:
 logging_steps: 1
 flash_attention: true
-warmup_steps: 10
+warmup_ratio: 0.1
 evals_per_epoch: 4
 saves_per_epoch: 1
 weight_decay: 0.0
--- a/examples/llama-2/qlora-fsdp.yml
+++ b/examples/llama-2/qlora-fsdp.yml
@@ -50,7 +50,7 @@ resume_from_checkpoint:
 logging_steps: 1
 flash_attention: true
-warmup_steps: 10
+warmup_ratio: 0.1
 evals_per_epoch: 4
 saves_per_epoch: 1
 weight_decay: 0.0
--- a/examples/llama-2/qlora.yml
+++ b/examples/llama-2/qlora.yml
@@ -48,7 +48,7 @@ resume_from_checkpoint:
 logging_steps: 1
 flash_attention: true
-warmup_steps: 10
+warmup_ratio: 0.1
 evals_per_epoch: 4
 saves_per_epoch: 1
 weight_decay: 0.0
--- a/examples/llama-2/relora.yml
+++ b/examples/llama-2/relora.yml
@@ -26,7 +26,7 @@ lora_dropout: 0.05
 lora_target_linear: true
 relora_steps: 150
-relora_warmup_steps: 10
+relora_warmup_ratio: 0.1
 relora_cpu_offload: false
 wandb_project:
@@ -50,7 +50,7 @@ resume_from_checkpoint:
 logging_steps: 1
 flash_attention: true
-warmup_steps: 10
+warmup_ratio: 0.1
 evals_per_epoch: 4
 saves_per_epoch: 1
 weight_decay: 0.0
--- a/examples/llama-3/3b-qat-fsdp2.yaml
+++ b/examples/llama-3/3b-qat-fsdp2.yaml
@@ -58,7 +58,7 @@ logging_steps: 1
 evals_per_epoch: 1
 saves_per_epoch: 1
-warmup_steps: 10
+warmup_ratio: 0.1
 weight_decay: 0.0
 fsdp:
  - full_shard
--- a/examples/llama-3/fft-8b-liger-fsdp.yaml
+++ b/examples/llama-3/fft-8b-liger-fsdp.yaml
@@ -51,7 +51,7 @@ resume_from_checkpoint:
 logging_steps: 1
 flash_attention: true
-warmup_steps: 100
+warmup_ratio: 0.1
 evals_per_epoch: 2
 saves_per_epoch: 1
 weight_decay: 0.0
--- a/examples/llama-3/fft-8b.yaml
+++ b/examples/llama-3/fft-8b.yaml
@@ -36,7 +36,7 @@ resume_from_checkpoint:
 logging_steps: 1
 flash_attention: true
-warmup_steps: 100
+warmup_ratio: 0.1
 evals_per_epoch: 2
 saves_per_epoch: 1
 weight_decay: 0.0
--- a/examples/llama-3/instruct-dpo-lora-8b.yml
+++ b/examples/llama-3/instruct-dpo-lora-8b.yml
@@ -67,7 +67,7 @@ resume_from_checkpoint:
 logging_steps: 1
 flash_attention: true
-warmup_steps: 10
+warmup_ratio: 0.1
 evals_per_epoch: 4
 saves_per_epoch: 1
 weight_decay: 0.0
--- a/examples/llama-3/instruct-lora-8b.yml
+++ b/examples/llama-3/instruct-lora-8b.yml
@@ -58,7 +58,7 @@ resume_from_checkpoint:
 logging_steps: 1
 flash_attention: true
-warmup_steps: 10
+warmup_ratio: 0.1
 evals_per_epoch: 4
 saves_per_epoch: 1
 weight_decay: 0.0
--- a/examples/llama-3/lora-1b-deduplicate-dpo.yml
+++ b/examples/llama-3/lora-1b-deduplicate-dpo.yml
@@ -79,7 +79,7 @@ resume_from_checkpoint:
 logging_steps: 1
 flash_attention: true
-warmup_steps: 10
+warmup_ratio: 0.1
 evals_per_epoch: 4
 saves_per_epoch: 1
 weight_decay: 0.0
--- a/examples/llama-3/lora-1b-deduplicate-sft.yml
+++ b/examples/llama-3/lora-1b-deduplicate-sft.yml
@@ -55,7 +55,7 @@ resume_from_checkpoint:
 logging_steps: 1
 flash_attention: true
-warmup_steps: 10
+warmup_ratio: 0.1
 evals_per_epoch: 4
 saves_per_epoch: 1
 weight_decay: 0.0
--- a/examples/llama-3/lora-1b-kernels.yml
+++ b/examples/llama-3/lora-1b-kernels.yml
@@ -59,7 +59,7 @@ flash_attention: true
 loss_watchdog_threshold: 5.0
 loss_watchdog_patience: 3
-warmup_steps: 10
+warmup_ratio: 0.1
 evals_per_epoch: 4
 saves_per_epoch: 1
 weight_decay: 0.0
--- a/examples/llama-3/lora-1b-ray.yml
+++ b/examples/llama-3/lora-1b-ray.yml
@@ -53,7 +53,7 @@ flash_attention: true
 loss_watchdog_threshold: 5.0
 loss_watchdog_patience: 3
-warmup_steps: 10
+warmup_ratio: 0.1
 evals_per_epoch: 4
 saves_per_epoch: 1
--- a/examples/llama-3/lora-1b-sample-packing-sequentially.yml
+++ b/examples/llama-3/lora-1b-sample-packing-sequentially.yml
@@ -57,7 +57,7 @@ resume_from_checkpoint:
 logging_steps: 1
 flash_attention: true
-warmup_steps: 10
+warmup_ratio: 0.1
 evals_per_epoch: 4
 saves_per_epoch: 1
 weight_decay: 0.0
--- a/examples/llama-3/lora-1b.yml
+++ b/examples/llama-3/lora-1b.yml
@@ -54,7 +54,7 @@ flash_attention: true
 loss_watchdog_threshold: 5.0
 loss_watchdog_patience: 3
-warmup_steps: 10
+warmup_ratio: 0.1
 evals_per_epoch: 4
 saves_per_epoch: 1
 weight_decay: 0.0
--- a/examples/llama-3/lora-8b.yml
+++ b/examples/llama-3/lora-8b.yml
@@ -51,7 +51,7 @@ resume_from_checkpoint:
 logging_steps: 1
 flash_attention: true
-warmup_steps: 10
+warmup_ratio: 0.1
 evals_per_epoch: 4
 saves_per_epoch: 1
 weight_decay: 0.0
--- a/examples/llama-3/qlora-1b-kto.yaml
+++ b/examples/llama-3/qlora-1b-kto.yaml
@@ -55,7 +55,7 @@ resume_from_checkpoint:
 logging_steps: 1
 flash_attention: true
-warmup_steps: 20
+warmup_ratio: 0.1
 evals_per_epoch: 4
 saves_per_epoch: 1
 weight_decay: 0.0
--- a/examples/llama-3/qlora-1b.yml
+++ b/examples/llama-3/qlora-1b.yml
@@ -56,7 +56,7 @@ flash_attention: true
 loss_watchdog_threshold: 5.0
 loss_watchdog_patience: 3
-warmup_steps: 10
+warmup_ratio: 0.1
 evals_per_epoch: 4
 saves_per_epoch: 1
 weight_decay: 0.0
--- a/examples/llama-3/qlora-fsdp-405b.yaml
+++ b/examples/llama-3/qlora-fsdp-405b.yaml
@@ -41,7 +41,7 @@ gradient_checkpointing_kwargs:
 logging_steps: 1
 flash_attention: true
-warmup_steps: 10
+warmup_ratio: 0.1
 evals_per_epoch: 4
 saves_per_epoch: 1
 weight_decay: 0.0
--- a/examples/llama-3/qlora-fsdp-70b.yaml
+++ b/examples/llama-3/qlora-fsdp-70b.yaml
@@ -50,7 +50,7 @@ resume_from_checkpoint:
 logging_steps: 1
 flash_attention: true
-warmup_steps: 10
+warmup_ratio: 0.1
 evals_per_epoch: 4
 saves_per_epoch: 1
 weight_decay: 0.0
--- a/examples/llama-3/qlora.yml
+++ b/examples/llama-3/qlora.yml
@@ -48,7 +48,7 @@ resume_from_checkpoint:
 logging_steps: 1
 flash_attention: true
-warmup_steps: 10
+warmup_ratio: 0.1
 evals_per_epoch: 4
 saves_per_epoch: 1
 weight_decay: 0.0
--- a/examples/llama-3/sparse-finetuning.yaml
+++ b/examples/llama-3/sparse-finetuning.yaml
@@ -47,7 +47,7 @@ logging_steps: 1
 xformers_attention:
 flash_attention: true
-warmup_steps: 100
+warmup_ratio: 0.1
 evals_per_epoch: 2
 eval_table_size:
 saves_per_epoch: 1
--- a/examples/llama-4/do-no-use-fa2/maverick-qlora-fsdp1.yaml
+++ b/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
--- a/examples/llama-4/do-no-use-fa2/scout-qlora-fsdp1.yaml
+++ b/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
--- a/examples/llama-4/do-no-use-fa2/scout-qlora-single-h100.yaml
+++ b/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
--- a/examples/llama-4/do-no-use-fa2/scout-vision-qlora-fsdp.yaml
+++ b/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
--- 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
--- a/examples/llama-4/scout-qlora-single-h100-flex.yaml
+++ b/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
--- a/examples/llama-4/scout-vision-qlora-fsdp2-flex.yaml
+++ b/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
--- a/examples/mamba/config.yml
+++ b/examples/mamba/config.yml
@@ -41,7 +41,7 @@ resume_from_checkpoint:
 logging_steps: 1
 flash_attention:
-warmup_steps: 10
+warmup_ratio: 0.1
 evals_per_epoch: 4
 saves_per_epoch: 1
 weight_decay: 0.0
--- a/examples/mistral/config.yml
+++ b/examples/mistral/config.yml
@@ -38,7 +38,7 @@ resume_from_checkpoint:
 logging_steps: 1
 flash_attention: true
-warmup_steps: 10
+warmup_ratio: 0.1
 evals_per_epoch: 4
 saves_per_epoch: 1
 weight_decay: 0.0
--- a/examples/mistral/lora-mps.yml
+++ b/examples/mistral/lora-mps.yml
@@ -59,7 +59,7 @@ sdp_attention: true
 loss_watchdog_threshold: 5.0
 loss_watchdog_patience: 3
-warmup_steps: 10
+warmup_ratio: 0.1
 evals_per_epoch: 4
 saves_per_epoch: 1
 weight_decay: 0.0
--- a/examples/mistral/lora.yml
+++ b/examples/mistral/lora.yml
@@ -59,7 +59,7 @@ flash_attention: true
 loss_watchdog_threshold: 5.0
 loss_watchdog_patience: 3
-warmup_steps: 10
+warmup_ratio: 0.1
 evals_per_epoch: 4
 saves_per_epoch: 1
 weight_decay: 0.0
--- a/examples/mistral/mistral-dpo-qlora.yml
+++ b/examples/mistral/mistral-dpo-qlora.yml
@@ -73,7 +73,7 @@ resume_from_checkpoint:
 logging_steps: 1
 flash_attention: false
-warmup_steps: 10
+warmup_ratio: 0.1
 evals_per_epoch: 4
 saves_per_epoch: 1
 weight_decay: 0.0
--- a/examples/mistral/mistral-qlora-fsdp.yml
+++ b/examples/mistral/mistral-qlora-fsdp.yml
@@ -56,7 +56,7 @@ flash_attention: true
 loss_watchdog_threshold: 5.0
 loss_watchdog_patience: 3
-warmup_steps: 10
+warmup_ratio: 0.1
 evals_per_epoch: 4
 saves_per_epoch: 1
--- a/examples/mistral/mistral-qlora-orpo.yml
+++ b/examples/mistral/mistral-qlora-orpo.yml
@@ -64,7 +64,7 @@ flash_attention: true
 loss_watchdog_threshold: 5.0
 loss_watchdog_patience: 3
-warmup_steps: 10
+warmup_ratio: 0.1
 evals_per_epoch: 4
 saves_per_epoch: 1
 weight_decay: 0.0
--- a/examples/mistral/mixtral-8x22b-qlora-fsdp.yml
+++ b/examples/mistral/mixtral-8x22b-qlora-fsdp.yml
@@ -54,7 +54,7 @@ flash_attention: true
 loss_watchdog_threshold: 5.0
 loss_watchdog_patience: 3
-warmup_steps: 10
+warmup_ratio: 0.1
 evals_per_epoch: 4
 saves_per_epoch: 1
--- a/examples/mistral/mixtral-qlora-fsdp.yml
+++ b/examples/mistral/mixtral-qlora-fsdp.yml
@@ -56,7 +56,7 @@ flash_attention: true
 loss_watchdog_threshold: 5.0
 loss_watchdog_patience: 3
-warmup_steps: 10
+warmup_ratio: 0.1
 evals_per_epoch: 4
 saves_per_epoch: 1
--- a/examples/mistral/mixtral.yml
+++ b/examples/mistral/mixtral.yml
@@ -74,7 +74,7 @@ flash_attention: true
 loss_watchdog_threshold: 5.0
 loss_watchdog_patience: 3
-warmup_steps: 10
+warmup_ratio: 0.1
 evals_per_epoch: 4
 saves_per_epoch: 1
--- a/examples/mistral/qlora.yml
+++ b/examples/mistral/qlora.yml
@@ -59,7 +59,7 @@ flash_attention: true
 loss_watchdog_threshold: 5.0
 loss_watchdog_patience: 3
-warmup_steps: 10
+warmup_ratio: 0.1
 evals_per_epoch: 4
 saves_per_epoch: 1
 weight_decay: 0.0
--- a/examples/orpheus/finetune.yml
+++ b/examples/orpheus/finetune.yml
@@ -43,7 +43,7 @@ resume_from_checkpoint:
 logging_steps: 1
 flash_attention: true
-warmup_steps: 20
+warmup_ratio: 0.1
 evals_per_epoch: 5
 saves_per_epoch: 5
 weight_decay: 0.05
--- a/examples/phi/lora-3.5.yaml
+++ b/examples/phi/lora-3.5.yaml
@@ -59,7 +59,7 @@ gradient_checkpointing: true
 resume_from_checkpoint:
 logging_steps: 1
-warmup_steps: 10
+warmup_ratio: 0.1
 evals_per_epoch: 4
 saves_per_epoch: 4
 weight_decay: 0.0
--- a/examples/phi/phi-ft.yml
+++ b/examples/phi/phi-ft.yml
@@ -50,7 +50,7 @@ resume_from_checkpoint:
 logging_steps: 1
 flash_attention: true
-warmup_steps: 100
+warmup_ratio: 0.1
 evals_per_epoch: 4
 saves_per_epoch: 1
 weight_decay: 0.1
--- a/examples/phi/phi-qlora.yml
+++ b/examples/phi/phi-qlora.yml
@@ -53,7 +53,7 @@ resume_from_checkpoint:
 logging_steps: 1
 flash_attention: true
-warmup_steps: 100
+warmup_ratio: 0.1
 evals_per_epoch: 4
 saves_per_epoch: 1
 weight_decay: 0.1
--- a/examples/phi/phi2-ft.yml
+++ b/examples/phi/phi2-ft.yml
@@ -50,7 +50,7 @@ resume_from_checkpoint:
 logging_steps: 1
 flash_attention: true
-warmup_steps: 100
+warmup_ratio: 0.1
 evals_per_epoch: 4
 saves_per_epoch: 1
 weight_decay: 0.1
--- a/examples/phi/phi3-ft-fsdp.yml
+++ b/examples/phi/phi3-ft-fsdp.yml
@@ -51,7 +51,7 @@ resume_from_checkpoint:
 logging_steps: 1
 flash_attention: true
-warmup_steps: 100
+warmup_ratio: 0.1
 evals_per_epoch: 4
 saves_per_epoch: 1
 weight_decay: 0.1
--- a/examples/qwen2/dpo.yaml
+++ b/examples/qwen2/dpo.yaml
@@ -50,7 +50,7 @@ resume_from_checkpoint:
 logging_steps: 1
 flash_attention: true
-warmup_steps: 10
+warmup_ratio: 0.1
 evals_per_epoch: 4
 saves_per_epoch: 1
 weight_decay: 0.0
--- a/examples/qwen2/qlora-fsdp.yaml
+++ b/examples/qwen2/qlora-fsdp.yaml
@@ -49,7 +49,7 @@ resume_from_checkpoint:
 logging_steps: 1
 flash_attention: true
-warmup_steps: 10
+warmup_ratio: 0.1
 evals_per_epoch: 4
 saves_per_epoch: 1
 weight_decay: 0.0
--- a/examples/qwen3/32b-qlora.yaml
+++ b/examples/qwen3/32b-qlora.yaml
@@ -62,7 +62,7 @@ resume_from_checkpoint:
 logging_steps: 1
 flash_attention: true
-warmup_steps: 10
+warmup_ratio: 0.1
 evals_per_epoch: 4
 saves_per_epoch: 1
 weight_decay: 0.0
--- a/examples/qwen3/8b-qat-fsdp2.yml
+++ b/examples/qwen3/8b-qat-fsdp2.yml
@@ -58,7 +58,7 @@ logging_steps: 1
 evals_per_epoch: 1
 saves_per_epoch: 1
-warmup_steps: 10
+warmup_ratio: 0.1
 weight_decay: 0.0
 fsdp:
  - full_shard
--- a/examples/qwen3/qlora-fsdp.yaml
+++ b/examples/qwen3/qlora-fsdp.yaml
@@ -48,7 +48,7 @@ resume_from_checkpoint:
 logging_steps: 1
 flash_attention: true
-warmup_steps: 10
+warmup_ratio: 0.1
 evals_per_epoch: 4
 saves_per_epoch: 1
 weight_decay: 0.0