Training in progress, step 50

adapter_config.json

@@ -20,13 +20,13 @@
   "rank_pattern": {},
   "revision": null,
   "target_modules": [
-    "q_proj",
-    "down_proj",
-    "v_proj",
-    "up_proj",
     "gate_proj",
     "k_proj",
-    "o_proj"
+    "q_proj",
+    "up_proj",
+    "o_proj",
+    "down_proj",
+    "v_proj"
   ],
   "task_type": "CAUSAL_LM",
   "use_dora": false,

adapter_model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9c88c95b04963e4f18b6ddc7c1720b05142b6273b0f530ccb48b4e79a5e48716
+oid sha256:ba6098ff0bb0074b01a084c6235684e66ae223c1f9e74d3e280c1f4006617018
 size 83945296

llama3_lora_sft.yaml

@@ -32,7 +32,7 @@ load_best_model_at_end: true
 per_device_train_batch_size: 1
 gradient_accumulation_steps: 16
 learning_rate: 1.0e-4
-num_train_epochs: 10.0
+num_train_epochs: 5.0
 lr_scheduler_type: cosine
 warmup_ratio: 0.1
 bf16: true

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a9ce98a38dadb51181607a02d76bcea0b82dd2f11cffdc127a4446f9ca4e5976
+oid sha256:13d33f5c6f17f9bf9724fe3d9566f9b350aa9673707489aa3ec437427e9b0249
 size 5624