Skip to content

Latest commit

 

History

History
89 lines (72 loc) · 4.35 KB

train_video_chatgpt.md

File metadata and controls

89 lines (72 loc) · 4.35 KB
Raw

Training Video-ChatGPT

We train Video-ChatGPT model on our 100K video instruction dataset. We initialize the training from LLaVA. Please follow the instructions below to train Video-ChatGPT-7B model.

Prepare LLaVA weights

Video-ChatGPT is build using LLaVA. Please follow the following instructions to get LLaVA weights.

  • Get the original LLaMA weights in the Hugging Face format by following the instructions here.
  • Use the following scripts to get LLaVA weights by applying our delta.
python scripts/apply_delta.py \ 
        --base-model-path <path to LLaMA 7B weights> \
        --target-model-path LLaVA-Lightning-7B-v1-1 \
        --delta-path liuhaotian/LLaVA-Lightning-7B-delta-v1-1

The above command will download the LLaVA-Lightening-7B-v1-1 delta from Hugging Face, apply it to the provided LLaMA weights and save the LLaVA-Lightening-7B-v1-1 weights in the current directory.

Alternatively you can download the ready LLaVA-Lightening-7B weights from mmaaz60/LLaVA-Lightening-7B-v1-1.

Prepare Dataset

1. Download our 100K video instruction dataset from this download link.

2. Convert the downloaded Json into the required format for training,

python scripts/convert_instruction_json_to_training_format.py \
        --input_json_file <path to json file downloaded in step 2> \
        --output_json_file video_chatgpt_training.json

The above script will generate video_chatgpt_training.json required to train our model.

3. Download ActivityNet videos

All the videos annotated in our work are taken from ActivityNet dataset. We provide the ids of all the required videos in the train_video_ids.txt file. Please follow the instructions on the official site to download the videos. Alternatively, you can download these from here.

4. Prepare Spatio-Temporal features using CLIP Note that for training efficiency, we pre-computed the video spatio-temporal features and use them directly during training. After downloading the videos, please use the following command to generate CLIP spatio-temporal features.

python scripts/save_spatio_temporal_clip_features.py \
        --video_dir_path <path to the directory containing all the videos> \
        --clip_feat_path <The output dir to save the features in.>

The script will generate the spatiotempral features for each video and save one pickle file per video in directory specified by --clip_feat_path argemunt. Alternatively, you can download the pre-computed spatiotemporal CLIP features from here.

Train Video-ChatGPT

Train on 8 A100 40GB GPUs using the following command,

torchrun --nproc_per_node=8 --master_port 29001 video_chatgpt/train/train_mem.py \
          --model_name_or_path <path to LLaVA-7B-Lightening-v-1-1 model> \
          --version v1 \
          --data_path <path to the video_chatgpt using `convert_instruction_json_to_training_format.py` script.> \
          --video_folder <path to the spatio-temporal features generated in step 4 using `save_spatio_temporal_clip_features.py` script> \
          --tune_mm_mlp_adapter True \
          --mm_use_vid_start_end \
          --bf16 True \
          --output_dir ./Video-ChatGPT_7B-1.1_Checkpoints \
          --num_train_epochs 3 \
          --per_device_train_batch_size 4 \
          --per_device_eval_batch_size 4 \
          --gradient_accumulation_steps 1 \
          --evaluation_strategy "no" \
          --save_strategy "steps" \
          --save_steps 3000 \
          --save_total_limit 3 \
          --learning_rate 2e-5 \
          --weight_decay 0. \
          --warmup_ratio 0.03 \
          --lr_scheduler_type "cosine" \
          --logging_steps 100 \
          --tf32 True \
          --model_max_length 2048 \
          --gradient_checkpointing True \
          --lazy_preprocess True