DeepSeek/ESFT
1
0
Fork 0
mirror of https://github.com/deepseek-ai/ESFT synced 2025-06-26 18:15:50 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

streamline code

Browse Source
This commit is contained in:
ZihanWang314
2025-05-22 07:47:10 +00:00
parent f38f67706c
commit 3746ca7441
5 changed files with 67 additions and 55 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
configs/base.yaml
View File

@@ -8,7 +8,7 @@ n_device: 8 # Number of devices
# Training settings
optim: adamw_torch_fused
steps: 1 # Number of training steps
steps: 500 # Number of training steps
learning_rate: 0.00001 # Learning rate
weight_decay: 0.1 # Weight decay for optimizer
warmup_steps: 0 # Number of warmup steps for learning rate scheduler
@@ -19,8 +19,8 @@ random_concat_ratio: 0.2 # Ratio of random concatenation
# Evaluation settings
eval_steps: 1 # Evaluate every X steps
save_steps: 1 # Save model every X steps
eval_steps: 100 # Evaluate every X steps
save_steps: 100 # Save model every X steps
# Tokenizer settings