Does it support finetuning for InternLM-XComposer2-4KD?

[windar427]

only internlm-xcomposer-7b and internlm/internlm-xcomposer-vl-7b are supported. finetrun script

https://github.com/InternLM/InternLM-XComposer/issues/259#issue-2237149013

#!/bin/bash
export CUDA_DEVICE_MAX_CONNECTIONS=1
DIR=`pwd`

export MODEL="internlm/internlm-xcomposer2-4khd-7b"
export DATA="data.txt"

GPUS_PER_NODE=8
NNODES=1
NODE_RANK=0
MASTER_ADDR=localhost
MASTER_PORT=6001

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.py \
    --model_name_or_path $MODEL \
    --data_path $DATA \
    --img_size 490 \
    --given_num True \
    --bf16 True \
    --fix_vit False \
    --fix_sampler False \
    --use_lora False \
    --output_dir output/finetune \
    --num_train_epochs 1 \
    --batch_size 1 \
    --per_device_train_batch_size 1 \
    --per_device_eval_batch_size 1 \
    --gradient_accumulation_steps 8 \
    --evaluation_strategy "no" \
    --save_strategy "epoch" \
    --save_total_limit 1 \
    --learning_rate 1e-5 \
    --weight_decay 0.1 \
    --adam_beta2 0.95 \
    --warmup_ratio 0.01 \
    --lr_scheduler_type "cosine" \
    --logging_steps 1 \
    --report_to "none" \
    --max_length 16384 \
    --deepspeed ds_config_zero2.json \
    --gradient_checkpointing True

[windar427]

When conducting 4HD training, what is the default size?

[yuhangzang]

The default batch size is 1.