Upload folder using huggingface_hub

pretrain_all.sh

@@ -0,0 +1,42 @@
+CUDA_VISIBLE_DEVICES=0,1 deepspeed llava/train/train_mem.py \
+    --deepspeed ./scripts/zero2.json \
+    --model_name_or_path ./Meta-Llama-3.1-8B-Instruct \
+    --version plain \
+    --data_path train_json/video_image_asr_caption_pre.json \
+    --audio_asr_folder /home/yu/Downloads \
+    --audio_caption_folder /media/yu/33da5c9f-d06a-4b18-acc4-b4e1c20292471/0818 \
+    --video_folder video_data \
+    --image_folder /media/yu/33da5c9f-d06a-4b18-acc4-b4e1c20292471 \
+    --X "Audio_asr" "Audio_caption" "Video" "Image" \
+    --audio_tower ./LanguageBind_Audio_Asr \
+    --audio_caption_tower LanguageBind/LanguageBind_Audio \
+    --video_tower LanguageBind/LanguageBind_Video_merge \
+    --image_tower LanguageBind/LanguageBind_Image \
+    --mm_projector_type mlp2x_gelu \
+    --tune_mm_mlp_adapter True \
+    --mm_vision_select_layer -2 \
+    --mm_use_x_start_end False \
+    --mm_use_x_patch_token False \
+    --bf16 True \
+    --output_dir ./checkpoints/Video-LLaVA-Pretrain-7B \
+    --num_train_epochs 1 \
+    --per_device_train_batch_size 1 \
+    --per_device_eval_batch_size 16 \
+    --gradient_accumulation_steps 32 \
+    --evaluation_strategy "no" \
+    --save_strategy "steps" \
+    --save_steps 2000 \
+    --save_total_limit 20 \
+    --learning_rate 1e-3 \
+    --weight_decay 0. \
+    --warmup_ratio 0.03 \
+    --lr_scheduler_type "cosine" \
+    --logging_steps 1 \
+    --tf32 True \
+    --model_max_length 2048 \
+    --tokenizer_model_max_length 3072 \
+    --gradient_checkpointing True \
+    --dataloader_num_workers 8 \
+    --lazy_preprocess True \
+    --report_to tensorboard \
+    --cache_dir "./cache_dir"