Update the README with latest instructions on how to build Habitat #5
Conversation
|
|
||
| # Download the pre-trained models | ||
| cd analyzer | ||
| curl -O https://zenodo.org/record/4876277/files/habitat-models.tar.gz\?download\=1 |
There was a problem hiding this comment.
Should use the newer trained models here.
There was a problem hiding this comment.
Originally, I had the new link. However, given that you're hosting it under University resources, that could potentially cause legal problems for us. So I'll update the repo with Git LFS maybe after this change is approved.
| 1. Once inside the container, run `install-dev.sh` under `analyzer/` to build and install the Habitat package. | ||
| 1. In your scripts, `import habitat` to get access to Habitat. See `experiments/run_experiment.py` for an example showing how to use Habitat. | ||
|
|
||
| **Note:** Habitat needs access to your GPU's performance counters, which requires special permissions if you are running with a recent driver (418.43 or later). If you encounter a `CUPTI_ERROR_INSUFFICIENT_PRIVILEGES` error when running Habitat, please follow the instructions [here](https://developer.nvidia.com/ERR_NVGPUCTRPERM) and in [issue #5](https://github.com/geoffxy/habitat/issues/5). |
There was a problem hiding this comment.
This is not only a problem when building with Docker. Would suggest moving this to the top.
| curl -O https://zenodo.org/record/4876277/files/habitat-models.tar.gz\?download\=1 | ||
|
|
||
| # Install the models | ||
| ./extract-models.sh |
There was a problem hiding this comment.
Doesn't need to be addressed in this PR, but something to think about - is it possible to wrap these installation steps inside a setup.py so that the user can automate the installation via pip or conda?
README.md
Outdated
|
|
||
| ### Building with Docker | ||
|
|
||
| 1. Run `setup.sh` under `docker/` to build the Habitat container image. |
There was a problem hiding this comment.
Again, something to think about - I took a look at the Dockerfile under docker/. Would it be easier to use NGC PyTorch containers as your base image? This way you don't have to manually align the version (of CUDA and any other NVIDIA tools) in the image with the version that PyTorch needs going forward.
NGC PyTorch containers have a fair amount of enterprise users. It'd be nice if we have a list in the readme that says "Habitat is tested on NGC PyTorch containers from version X to version Y".
There was a problem hiding this comment.
That's a good idea. I'll try building Habitat using these containers.
There was a problem hiding this comment.
I've tried Habitat with the NGC containers and it works out of the box. The user should either:
- Clone the repository, initialize the submodules, run
install-dev.shthen import the saved models, or - Install the correct wheel file (in the case of the latest NGC container, CUDA 11.7 + Python 3.8)
With that said, we should still use nvidia/cuda images for automated builds since would allow us to specify the version of CUDA that's linked to habitat_cuda.
README.md
Outdated
|
|
||
| 1. Run `setup.sh` under `docker/` to build the Habitat container image. | ||
| 1. Run `start.sh` to start a new container. By default, your home directory will be mounted inside the container under `~/home`. | ||
| 1. Once inside the container, run `install-dev.sh` under `analyzer/` to build and install the Habitat package. |
There was a problem hiding this comment.
Include the actual commands of how you run setup.sh or start.sh and what bash arguments they expect. E.g.,
bash docker/setup.sh <X> <Y> <Z>where <X> is a string that represents XXX, and its options are {a, b, c}.
Something like that. The same goes for the other places that says "Run xxx".
README.md
Outdated
| chmod +x cuda_11.7.1_515.65.01_linux.run | ||
|
|
||
| # Make sure to only install the toolkit | ||
| ./cuda_11.7.1_515.65.01_linux.run |
There was a problem hiding this comment.
Do you have to install CUPTI from the runfile? Is CUPTI not available from rmp or deb? Similarly for cmake?
I just feel like the users could get lost easily with such a long installation instruction. From a user experience point of view, it'd be better if we have a list of scripts and a multiplexer to select between them depending on the user's distro. E.g., for Ubuntu 22.02, run this command. For RedHat, run that command, etc. And we don't need to support that many distros. I personally feel like Ubuntu is 90% of the Linux for DL market right now.
Removed the INSTALL.md, since most of it has been incorporated into the main README.md. Some sections need further updates as we refine how we build and release the habitat.