Refresh base adapter to best eval_loss (profile=qwen3_5_9b, step=1152, eval_loss=0.037669)

qwen__qwen3.5-9b/adapter_model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1249c92de35d5c27bfcec51c1fb87fc0ca48bf11da5369311e172aea490a4cdd
+oid sha256:486ed260350c203aee8100eac07c25e4103d918c3f5186932be4241c6b2d800f
 size 80275624

qwen__qwen3.5-9b/best_eval_metrics.json

@@ -1,5 +1,5 @@
 {
-  "best_eval_loss": 0.03822297602891922,
-  "best_step": 864,
-  "epoch": 1.502608695652174
+  "best_eval_loss": 0.037669453769922256,
+  "best_step": 1152,
+  "epoch": 2.0034782608695654
 }