[Draft] Add minicpmv finetune script

ds_config_zero2.json

@@ -0,0 +1,52 @@
+{
+    "fp16": {
+        "enabled": "auto",
+        "loss_scale": 0,
+        "loss_scale_window": 1000,
+        "initial_scale_power": 16,
+        "hysteresis": 2,
+        "min_loss_scale": 1
+    },
+    "bf16": {
+        "enabled": "auto"
+    },
+    "optimizer": {
+        "type": "AdamW",
+        "params": {
+            "lr": "auto",
+            "betas": "auto",
+            "eps": "auto",
+            "weight_decay": "auto"
+        }
+    },
+
+    "scheduler": {
+        "type": "WarmupLR",
+        "params": {
+            "warmup_min_lr": "auto",
+            "warmup_max_lr": "auto",
+            "warmup_num_steps": "auto"
+        }
+    },
+
+    "zero_optimization": {
+        "stage": 2,
+        "offload_optimizer": {
+            "device": "none",
+            "pin_memory": true
+        },
+        "allgather_partitions": true,
+        "allgather_bucket_size": 2e8,
+        "overlap_comm": true,
+        "reduce_scatter": true,
+        "reduce_bucket_size": 2e8,
+        "contiguous_gradients": true
+    },
+
+    "gradient_accumulation_steps": "auto",
+    "gradient_clipping": "auto",
+    "steps_per_print": 100,
+    "train_batch_size": "auto",
+    "train_micro_batch_size_per_gpu": "auto",
+    "wall_clock_breakdown": false
+}

finetune_ds.sh

@@ -0,0 +1,48 @@
+#!/bin/bash
+export CUDA_DEVICE_MAX_CONNECTIONS=1
+DIR=`pwd`
+
+GPUS_PER_NODE=8
+NNODES=1
+NODE_RANK=0
+MASTER_ADDR=localhost
+MASTER_PORT=6001
+
+SAVE_PATH=/path/to/experiments/MiniCPM-V-FFT
+BASE_MODEL=/path/to/pretrained_model/MiniCPM-V
+TRAIN_DATASET=/path/to/train.json
+VAL_DATASET=/path/to/test.json
+
+
+DISTRIBUTED_ARGS="
+    --nproc_per_node $GPUS_PER_NODE \
+    --nnodes $NNODES \
+    --node_rank $NODE_RANK \
+    --master_addr $MASTER_ADDR \
+    --master_port $MASTER_PORT
+"
+
+torchrun $DISTRIBUTED_ARGS finetune_minicpmv.py \
+    --model_name_or_path $BASE_MODEL \
+    --data_path $TRAIN_DATASET \
+    --bf16 True \
+    --fix_vit True \
+    --output_dir $SAVE_PATH \
+    --num_train_epochs 2 \
+    --per_device_train_batch_size 12 \
+    --per_device_eval_batch_size 8 \
+    --gradient_accumulation_steps 4 \
+    --evaluation_strategy "no" \
+    --save_strategy "steps" \
+    --save_steps 100 \
+    --save_total_limit 2 \
+    --learning_rate 1e-5 \
+    --weight_decay 0.1 \
+    --adam_beta2 0.95 \
+    --warmup_ratio 0.01 \
+    --lr_scheduler_type "cosine" \
+    --logging_steps 2 \
+    --report_to "tensorboard" \
+    --model_max_length 512 \
+    --gradient_checkpointing True \
+    --deepspeed ds_config_zero2.json