chore: update for fine tune

unit-mesh · Dec 27, 2023 · f8f8d63 · f8f8d63
1 parent 217e258
commit f8f8d63
Showing 1 changed file with 3 additions and 3 deletions.
diff --git a/finetunes/deepseek/finetune.ipynb b/finetunes/deepseek/finetune.ipynb
@@ -87,14 +87,14 @@
     "    --output_dir $OUTPUT_PATH \\\n",
     "    --num_train_epochs 1 \\\n",
     "    --model_max_length 768 \\\n",
-    "    --per_device_train_batch_size 16 \\\n",
+    "    --per_device_train_batch_size 8 \\\n",
     "    --per_device_eval_batch_size 1 \\\n",
     "    --gradient_accumulation_steps 4 \\\n",
     "    --evaluation_strategy \"no\" \\\n",
     "    --save_strategy \"no\" \\\n",
-    "    --save_steps 50 \\\n",
+    "    --save_steps 375 \\\n",
     "    --save_total_limit 10 \\\n",
-    "    --learning_rate 4e-5 \\\n",
+    "    --learning_rate 1e-4 \\\n",
     "    --warmup_steps 10 \\\n",
     "    --logging_steps 1 \\\n",
     "    --lr_scheduler_type \"cosine\" \\\n",