Refresh base adapter to best eval_loss (profile=qwen3_5_9b, step=864, eval_loss=0.038223)

qwen__qwen3.5-9b/adapter_model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f54ee0e15865f2d52dea25af355336b225a01714f713a605b03a6286cfc1446d
+oid sha256:1249c92de35d5c27bfcec51c1fb87fc0ca48bf11da5369311e172aea490a4cdd
 size 80275624

qwen__qwen3.5-9b/best_eval_metrics.json

@@ -1,5 +1,5 @@
 {
-  "best_eval_loss": 0.045138414949178696,
-  "best_step": 288,
-  "epoch": 0.5008695652173913
+  "best_eval_loss": 0.03822297602891922,
+  "best_step": 864,
+  "epoch": 1.502608695652174
 }