From 7719d75c2044e300ef96f56c84fa9439ae7aaedd Mon Sep 17 00:00:00 2001
From: hyperzlib <hyperzlib@outlook.com>
Date: Thu, 25 Apr 2024 15:54:31 +0800
Subject: [PATCH] =?UTF-8?q?=E6=9B=B4=E6=94=B9=E5=BE=AE=E8=B0=83=E8=AE=AD?=
 =?UTF-8?q?=E7=BB=83=E5=8F=82=E6=95=B0?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

---
 .gitignore                           | 2 ++
 finetune/finetune_lora_single_gpu.sh | 4 ++--
 2 files changed, 4 insertions(+), 2 deletions(-)

diff --git a/.gitignore b/.gitignore
index 39e9066..9a7fbea 100644
--- a/.gitignore
+++ b/.gitignore
@@ -9,3 +9,5 @@ build
 .DS_Store
 
 /private/
+output_qwen/
+output_qwen*/
\ No newline at end of file
diff --git a/finetune/finetune_lora_single_gpu.sh b/finetune/finetune_lora_single_gpu.sh
index 972bcf0..7c99e4b 100644
--- a/finetune/finetune_lora_single_gpu.sh
+++ b/finetune/finetune_lora_single_gpu.sh
@@ -41,7 +41,7 @@ python finetune.py \
   --data_path $DATA \
   --bf16 True \
   --output_dir output_qwen \
-  --num_train_epochs 5 \
+  --num_train_epochs 16 \
   --per_device_train_batch_size 2 \
   --per_device_eval_batch_size 1 \
   --gradient_accumulation_steps 8 \
@@ -56,7 +56,7 @@ python finetune.py \
   --lr_scheduler_type "cosine" \
   --logging_steps 1 \
   --report_to "none" \
-  --model_max_length 512 \
+  --model_max_length 3072 \
   --lazy_preprocess True \
   --gradient_checkpointing \
   --use_lora