Conversation
|
|
||
| # Install Python dependencies from requirements.txt | ||
| # Note: Some packages may already be in the base image, skip torch/torchvision/torchaudio | ||
| RUN pip install --no-cache-dir \ |
There was a problem hiding this comment.
they have requirements.txt, so it is better if we use an opt out approach
reproduction/repositories/nvidia-cosmos/cosmos-predict1/Dockerfile
Lines 25 to 29 in 1010c19
| RUN wget -q --show-progress \ | ||
| "https://huggingface.co/KwaiVGI/VFRTok/resolve/main/vfrtok-l.bin" \ | ||
| -O vfrtok-l.bin |
There was a problem hiding this comment.
generally better to download weights on the machine, rather than inside the dockerfile (that way it always happens just once)
hf download KwaiVGI/VFRTokand when running the docker add
-v /shared/.cache:/root/.cache\(in the README, like we do for Wan2.2)
| echo "/workspace/VFRTok/sample_data/sample.mp4" >> /workspace/VFRTok/test.csv | ||
|
|
||
| # Add metadata to CSV | ||
| RUN python scripts/add_metadata_to_csv.py -i /workspace/VFRTok/test.csv -o /workspace/VFRTok/test.csv --data_column video_path || true |
There was a problem hiding this comment.
testing files should exist outside the dockerfile, and they can be mounted. see the example in Wan2.2
| deepspeed inference.py -i /workspace/VFRTok/test.csv -o /workspace/VFRTok/outputs \ | ||
| --config configs/vfrtok-l.yaml --ckpt vfrtok-l.bin \ | ||
| --enc_fps 24 --dec_fps 60 | ||
| ``` |
There was a problem hiding this comment.
can be generally more minimal. can ask for it to look at other READMEs, they are all shorter., the notes section is useless, etc
|
FYI, you can also tell claude:
|
| image: vfrtok:latest | ||
| environment: | ||
| - NVIDIA_VISIBLE_DEVICES=all | ||
| - HF_HUB_DISABLE_XET=1 |
| docker compose run --rm vfrtok \ | ||
| deepspeed inference.py -i test.csv -o outputs \ | ||
| --config configs/vfrtok-l.yaml --ckpt vfrtok-l.bin --enc_fps 24 | ||
| ``` |
There was a problem hiding this comment.
I would prefer having docker run command like we have for other repositories, with all the steps - for example i don't know where test.csv is from.
so:
Build:
docker build -t vfrtok:latest -f repositories/KlingTeam/VFRTok/Dockerfile .Run:
mkdir -p outputs
docker run --it --rm --gpus all \
--ipc=host --ulimit memlock=-1 --ulimit stack=67108864 \
-v /shared/.cache/huggingface:/root/.cache/huggingface \
-v "$(pwd)/outputs":/outputs \
-it vfrtok:latest bashUsage:
Missing - how did we create test.csv? where is the video?
deepspeed inference.py -i test.csv -o /outputs \
--config configs/vfrtok-l.yaml --ckpt vfrtok-l.bin --enc_fps 24| @@ -0,0 +1,27 @@ | |||
| # LTX-2 Video VAE Tokenizer | |||
| FROM gogamza/unsloth-vllm-gb10:latest | |||
| # Install system dependencies | ||
| RUN apt-get update && apt-get install -y --no-install-recommends \ | ||
| libgl1 \ | ||
| && rm -rf /var/lib/apt/lists/* | ||
|
|
||
| # Install ffmpeg 4 from source (required for decord) | ||
| COPY libraries/ffmpeg/install_from_source.sh /tmp/install_ffmpeg.sh | ||
| RUN bash /tmp/install_ffmpeg.sh | ||
|
|
||
| # Install decord from source | ||
| COPY libraries/decord/install_from_source.sh /tmp/install_decord.sh | ||
| RUN bash /tmp/install_decord.sh |
There was a problem hiding this comment.
you are using the unsloth image. are all these necessary? do they not come already with the unsloth image?
| RUN pip install --no-cache-dir \ | ||
| mediapy \ | ||
| simple-video-utils \ | ||
| git+https://github.com/huggingface/diffusers.git |
There was a problem hiding this comment.
do we have to install from git?
|
|
||
| ```bash | ||
| # Reconstruct video | ||
| docker compose run --rm ltx2 python -m video_tokenizer.bin \ |
There was a problem hiding this comment.
you have a run.sh script -
so either usage should use ./run.sh encode <input> <output> etc, or it should be a docker run command but then no run.sh needed
2 participants
No description provided.