Refresh base adapter to best eval_loss (profile=llama3_1_8b_base_lessdata, step=1464, eval_loss=0.055296)

meta-llama__meta-llama-3.1-8b/sft/adapter_config.json

@@ -33,13 +33,13 @@
   "rank_pattern": {},
   "revision": null,
   "target_modules": [
-    "down_proj",
-    "v_proj",
-    "gate_proj",
+    "q_proj",
+    "k_proj",
     "up_proj",
+    "gate_proj",
+    "down_proj",
     "o_proj",
-    "q_proj",
-    "k_proj"
+    "v_proj"
   ],
   "target_parameters": null,
   "task_type": "CAUSAL_LM",

meta-llama__meta-llama-3.1-8b/sft/adapter_model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:962411903164374dd2a6a040c2423ac147ef552ddb0ecb206c3e2a97d648b430
+oid sha256:251f115f1399311d664d30a65f3f69edd39ca1066ff9de11cf1af58abc53789c
 size 83945296

meta-llama__meta-llama-3.1-8b/sft/best_eval_metrics.json

@@ -1,5 +1,5 @@
 {
-  "best_eval_loss": 0.03369956836104393,
-  "best_step": 5864,
-  "epoch": 0.9998294970161978
+  "best_eval_loss": 0.05529617890715599,
+  "best_step": 1464,
+  "epoch": 0.9986357435197817
 }