Skip to content

post-training_autoregressive_base.md

Latest commit

 

History

History
173 lines (129 loc) · 7.42 KB

post-training_autoregressive_base.md

File metadata and controls

173 lines (129 loc) · 7.42 KB

Post-training autoregressive-based base models

Model Support Matrix

We support the following Cosmos Autoregressive models for post-training. Review the available models and their compute requirements for post-training and inference to determine the best model for your use case.

Model Name Model Status Compute Requirements for Post-Training
Cosmos-Predict1-4B Supported 1 NVIDIA GPUs*

* H100-80GB or A100-80GB GPUs are recommended.

Environment setup

Please refer to the Post-training section of INSTALL.md for instructions on environment setup.

Download checkpoints

  1. Generate a Hugging Face access token (if you haven't done so already). Set the access token to Read permission (default is Fine-grained).

  2. Log in to Hugging Face with the access token:

    huggingface-cli login

  3. Accept the Llama-Guard-3-8B terms

  4. Download the Cosmos model weights from Hugging Face:

    CUDA_HOME=$CONDA_PREFIX PYTHONPATH=$(pwd) python scripts/download_autoregressive_checkpoints.py --model_sizes 4B --checkpoint_dir checkpoints

  5. For tensor parallel training, checkpoints need to be sharded to the target tensor model parallel size TP. Shard checkpoints to TP=4 with:

     python scripts/shard_autoregressive_base_checkpoints.py --checkpoint_path checkpoints/Cosmos-Predict1-4B/model.pt --model_size 4b --tensor_parallel_size 4

    This command will shard and save 4 TP checkpoint shards as checkpoints/Cosmos-Predict1-4B/model_model_mp_{rank}.pt

Examples

Post-training a Cosmos Autoregressive WFM enables you to train the model to generate videos that are more specific to your use case.

There are 2 steps to post-training: downloading a dataset and post-training the model.

1. Download a Dataset

The first step is to download a dataset with videos. You must provide a folder containing a collection of videos in MP4 format.

You can use nvidia/Cosmos-NeMo-Assets for post-training.

mkdir -p datasets/cosmos_nemo_assets/

# This command will download the videos for physical AI
huggingface-cli download nvidia/Cosmos-NeMo-Assets --repo-type dataset --local-dir datasets/cosmos_nemo_assets/ --include "*.mp4*"

mv datasets/cosmos_nemo_assets/nemo_diffusion_example_data datasets/cosmos_nemo_assets/videos

2. Post-train the model on a single GPU (TP = 1)

Run the following command to execute an example post-training job with the above data which is scaled to a lower resolution by the dataloader to fit on a single GPU

export OUTPUT_ROOT=checkpoints # default value
torchrun --nproc_per_node=1 -m cosmos_predict1.autoregressive.train --config=cosmos_predict1/autoregressive/configs/config.py -- experiment=base_4b_example_tealrobotsmall_tp1

The model will be post-trained using the above cosmos_nemo_assets dataset. See the VideoDataset defined in cosmos_predict1/autoregressive/datasets/video_dataset.py and register_training_data in cosmos_predict1/autoregressive/configs/registry.py to understand how the dataloader works and is registered.

The checkpoints will be saved to ${OUTPUT_ROOT}/PROJECT/GROUP/NAME. In the above example, PROJECT is posttraining, GROUP is autoregressive_base, NAME is base_4b_example_tealrobotsmall_tp1.

See the job config to understand how they are determined.

base_4b_example_tealrobotsmall_tp1= LazyDict(
    dict(
        ...
        job=dict(
            project="posttraining",
            group="autoregressive_base",
            name="base_4b_example_tealrobotsmall_tp1",
        ),
        ...
    )
)

During the training, the checkpoints will be saved in the below structure.

checkpoints/posttraining/autoregressive_base/base_4b_example_tealrobotsmall_tp1/checkpoints/
├── iter_{NUMBER}.pt

3. Post-train the model on 4 GPUs (TP = 4)

The model can also be post-trained on multiple GPUs using tensor parallelism. Run the following command to execute an example post-training job with the above data with higher resolution.

export OUTPUT_ROOT=checkpoints # default value
torchrun --nproc_per_node=4 -m cosmos_predict1.autoregressive.train --config=cosmos_predict1/autoregressive/configs/config.py -- experiment=base_4b_example_tealrobot_tp4

The checkpoints will be saved to ${OUTPUT_ROOT}/PROJECT/GROUP/NAME. In the above example, PROJECT is posttraining, GROUP is autoregressive_base, NAME is base_4b_example_tealrobot_tp4.

See the job config to understand how they are determined.

base_4b_example_tealrobotsmall_tp4= LazyDict(
    dict(
        ...
        job=dict(
            project="posttraining",
            group="autoregressive_base",
            name="base_4b_example_tealrobotsmall_tp4",
        ),
        ...
    )
)

During the training, the sharded checkpoints will be saved in the below structure.

checkpoints/posttraining/autoregressive_base/base_4b_example_tealrobot_tp4/checkpoints/
├── iter_{NUMBER}.pt
├── iter_{NUMBER}_model_mp_0.pt
├── iter_{NUMBER}_model_mp_1.pt
├── iter_{NUMBER}_model_mp_2.pt
├── iter_{NUMBER}_model_mp_3.pt

Inference with the Post-trained Model Checkpoint

The inference can be done with the same interface as described in examples/inference_autoregressive_base.md.

1. Copying checkpoint to Designated Location

The post-trained checkpoint needs to be copied to checkpoints/Cosmos-Predict1-4B-Base_post-trained/model.pt

For example, with TP=1 if a posttrained checkpoint with 1000 iterations is to be used,

# copy checkpoint to the designated location
mkdir checkpoints/Cosmos-Predict1-4B-Base_post-trained/
cp checkpoints/posttraining/autoregressive_base/base_4b_example_tealrobotsmall_tp1/checkpoints/iter_000001000.pt checkpoints/Cosmos-Predict1-4B-Base_post-trained/model.pt

With TP=4, the postrained checkpoints are sharded and should first be merged into a single checkpoint for inference

# merge tensor parallel model shards
mkdir checkpoints/Cosmos-Predict1-4B-Base_post-trained/
python scripts/merge_autoregressive_tp_checkpoints.py --checkpoint_path checkpoints/posttraining/autoregressive_base/base_4b_example_tealrobot_tp4/checkpoints/iter_000001000.pt --output_path checkpoints/Cosmos-Predict1-4B-Base_post-trained/model.pt --model_size 4b --tensor_parallel_size 4

2. Running the Inference

This is the basic example for running inference on the post-trained 4B model with a single video.

NUM_GPUS=<NUM_GPUS>
CUDA_HOME=$CONDA_PREFIX PYTHONPATH=$(pwd) torchrun --nproc_per_node=${NUM_GPUS} cosmos_predict1/autoregressive/inference/base.py \
    --num_gpus ${NUM_GPUS} \
    --checkpoint_dir checkpoints \
    --ar_model_dir Cosmos-Predict1-4B-Base_post-trained \
    --input_type video \
    --input_image_or_video_path datasets/cosmos_nemo_assets/videos/output_oige_render_view_sub.mp4  \
    --top_p 0.8 \
    --temperature 1.0 \
    --offload_diffusion_decoder \
    --offload_tokenizer \
    --video_save_name autoregressive-4b-post-train