tocmo0nlord/axolotl
1
0
Fork 0
Code Issues Pull Requests Actions 3 Packages Projects Releases Wiki Activity

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

Browse Source 
* 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
Download Patch File Download Diff File Expand all files Collapse all files

2
examples/archived/dbrx/16bit-lora.yaml
View File

@@ -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

2
examples/archived/dbrx/8bit-lora.yaml
View File

@@ -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

2
examples/archived/dbrx/fft-ds-zero3.yaml
View File

@@ -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