fixes

2025-08-18 19:09:09 +00:00
parent 556a69118f
commit b210db2d15
7 changed files with 33 additions and 44 deletions
--- a/examples/llama-3/diffusion-3.2-1b-sft.yaml
+++ b/examples/llama-3/diffusion-3.2-1b-sft.yaml
@@ -9,7 +9,7 @@ val_set_size: 0.05

 plugins:
  - diffusion.DiffusionPlugin
-noise_schedule: "linear"
+noise_schedule: cosine
 min_mask_ratio: 0.1
 max_mask_ratio: 0.9
 num_diffusion_steps: 128
@@ -39,6 +39,8 @@ resume_from_checkpoint:
 logging_steps: 1
 sdp_attention: true

+warmup_steps: 1000
+
 save_strategy: steps
 eval_strategy: steps
 save_steps: 500