From e274974822d387149d51007626c61868e6fc3d9e Mon Sep 17 00:00:00 2001
From: leoz9 <leoz@leozdeMacBook-Pro.local>
Date: Fri, 10 Jan 2025 16:09:12 +0800
Subject: [PATCH] add scripts

---
 .DS_Store             | Bin 0 -> 6148 bytes
 scripts/1-pretrain.sh |  34 ++++++++++++++++++++++++++++++++++
 scripts/2-full_sft.sh |   8 ++++++++
 scripts/3-lora.sh     |  13 +++++++++++++
 scripts/4-dpo         |   0
 5 files changed, 55 insertions(+)
 create mode 100644 .DS_Store
 create mode 100644 scripts/1-pretrain.sh
 create mode 100644 scripts/2-full_sft.sh
 create mode 100644 scripts/3-lora.sh
 create mode 100644 scripts/4-dpo

diff --git a/.DS_Store b/.DS_Store
new file mode 100644
index 0000000000000000000000000000000000000000..5f68b27b25f67ed393003b4f6ae5a625a93867cc
GIT binary patch
literal 6148
zcmeHKPfP1S5TC78H;B-Kg73EAwO}Hs2QRVK!@C;MgGy~o(O{YlO-c`?kc%hZ2Y6q_
zs|P=V;J0!1PeqM_Cl#52+28E!Ot$+=*y#WOYYih0paB3JRbkD+;T=)yN-I*co>@eu
zpOHZUc{s|mFsVe7<0mqp-rYRpa1H~A;l=xtA+FqpSrX@k>prmBOnr84o^!syPlHQ2
z3W{zq%v;^Tv6>x7nS@ij8y?2JQKzxAE7PJIr@etrh@&2coE*hzBu6bdOru=q24;af
zj?-!U8IRXD-KOxi+(}c6J=fi6icQa*OdP(vy0*RFxk~z}yfIud{Oq(WTAac^G~OI5
zx$uH>+0UjS)rh`qx5GpqoIn1ZySx5>zw=&H^ogkKva;KYjmabihyh~Ydl|5oCtvv9
zpFz751H`}&WI){y0#%`Ju{5Z!4jgm~fLKSj7PM6_p>l*p-(qPHS5TNqMKr0*wiwK$
zqhHuK-(qRdqyw|f2eUgf+o3SMJ3e3NaA3Yc8i@g7;57p?X4+8ofA8b||8)}0hyh~Y
zS24iq?V#PlBiY(oIh?As0`&}4iHa)?KBu6eOEJc(QoMqy1^q%9h`z<rAbL>fkAS9u
K24dh#8F&JRy=O!K

literal 0
HcmV?d00001

diff --git a/scripts/1-pretrain.sh b/scripts/1-pretrain.sh
new file mode 100644
index 0000000..930d370
--- /dev/null
+++ b/scripts/1-pretrain.sh
@@ -0,0 +1,34 @@
+#!/bin/bash
+
+# 环境变量配置
+export MASTER_PORT=29500      # 设置通信端口
+export CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7  # 设置使用的 GPU
+
+# 训练配置参数
+NUM_GPUS=8
+EPOCHS=3
+OUT_DIR="/home/jovyan/zlcode/minimind/out_dir"
+LOG_DIR="/home/jovyan/zlcode/minimind/log"
+LOG_FILE="${LOG_DIR}/pretrain.log"
+Batch_size=128
+# 创建输出目录和日志目录（如果不存在）
+mkdir -p "$OUT_DIR"
+mkdir -p "$LOG_DIR
+
+# 设置训练参数的选项
+TRAINING_OPTIONS="--use_wandb \
+                  --ddp \
+                  --epochs=$EPOCHS \
+                  --batch_size=${Batch_size} \
+                  --out_dir=$OUT_DIR"
+
+# 设置torchrun分布式训练参数
+DISTRIBUTED_OPTIONS="--nproc_per_node=$NUM_GPUS \
+                     --nnodes=1 \
+                     --rdzv_id=123 \
+                     --rdzv_backend=c10d"
+
+# 使用 nohup 启动训练任务，并且重定向输出
+nohup torchrun $DISTRIBUTED_OPTIONS 1-pretrain.py $TRAINING_OPTIONS \
+    2>&1 | tee "$LOG_FILE" &
+
diff --git a/scripts/2-full_sft.sh b/scripts/2-full_sft.sh
new file mode 100644
index 0000000..2973c44
--- /dev/null
+++ b/scripts/2-full_sft.sh
@@ -0,0 +1,8 @@
+nohup torchrun \
+    --nproc_per_node 8 \
+    3-full_sft.py \
+    --use_wandb \
+    --ddp \
+    --epochs 3 \
+    --out_dir /home/jovyan/zlcode/minimind/out_dir/sft \
+    2>&1 > /home/jovyan/zlcode/minimind/log/full_sft.log &
diff --git a/scripts/3-lora.sh b/scripts/3-lora.sh
new file mode 100644
index 0000000..70b9a15
--- /dev/null
+++ b/scripts/3-lora.sh
@@ -0,0 +1,13 @@
+#!/bin/bash
+
+# LoRA fine-tuning script
+# Run with distributed training and save logs
+
+nohup torchrun \
+  --nproc_per_node=8 \
+  4-lora_sft.py \
+  --use_wandb \
+  --epochs 3 \
+  --out_dir /home/jovyan/zlcode/minimind/out_dir \
+  2>&1 > /home/jovyan/zlcode/minimind/log/lora_sft.log &
+  
diff --git a/scripts/4-dpo b/scripts/4-dpo
new file mode 100644
index 0000000..e69de29