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-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