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
This commit is contained in:
Wing Lian
2025-07-30 06:44:06 -04:00
committed by GitHub
parent 2eb7ff95af
commit 22810c97b7
99 changed files with 100 additions and 100 deletions

View File

@@ -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

View File

@@ -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

View File

@@ -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

View File

@@ -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

View File

@@ -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