upgrade to cuda 12.4

Author: quy-ng

.condarc

@@ -1,4 +1,5 @@
-channel_priority: strict
 channels:
-  - conda-forge
-  - defaults
+  - pytorch
+  - nvidia
+  - pyg
+  - conda-forge

Dockerfile

@@ -1,5 +1,5 @@
-FROM mambaorg/micromamba:1.5.1 as micromamba
-FROM nvcr.io/nvidia/pytorch:23.09-py3
+FROM mambaorg/micromamba:2.0 as micromamba
+FROM nvcr.io/nvidia/pytorch:24.10-py3
 
 ENV LANG=C.UTF-8 LC_ALL=C.UTF-8
 ENV ENV_NAME=ml-venv
@@ -30,10 +30,11 @@ USER $MAMBA_USER
 SHELL ["/usr/local/bin/_dockerfile_shell.sh"]
 ENTRYPOINT ["/usr/local/bin/_entrypoint.sh"]
 
-RUN micromamba install -y -n base -c conda-forge python=3.11 \
+RUN micromamba install -y -n base -c conda-forge python=3.12 \
     && micromamba clean --all --yes
 
 WORKDIR /tmp
+
 # Install any needed packages specified in environment.yml
 COPY environment.yml environment.yml
 RUN micromamba create -n $ENV_NAME -f environment.yml -y \

README.md

@@ -1,5 +1,5 @@
 # Project Templete to Kickoff Machine Learning Project
-> This template currently uses `Python 3.11`, and either `conda`, `docker`, `poetry`, or `micromamba`
+> This template currently uses `Python 3.11`, `Python 3.12`, and either `conda`, `poetry`, or `conda` or `micromamba`
 
 Template project aims to promote *versioning library*, *environment isolation* practice and help all ML practitioners quickly start a project. Using this template, practitioners will have below libraries
 + Pytorch
@@ -17,15 +17,30 @@ Template project aims to promote *versioning library*, *environment isolation* p
 
 Those libraries of course aren't enough, but it's easy to update other libraries that support your project. 
 
-> Using `poetry` is highly recommended. If you are using `conda` or `micromamba`, make sure that you use package hashes to ensure package selection is reproducible via `conda-lock` or `micromamba`
+> Using `poetry` is highly recommended. If you are using `conda` or `micromamba`, make sure that you use package hashes to ensure package selection is reproducible via `conda-lock`.
 
-## Install library dependencies
+## Isolate Environment with Docker
+- This tutorial is for those who have NVIDIA GPU (hereafter GPU), and you must have `docker`
+- Install Nvidia driver
+- Then install [Nvidia docker container toolkit](https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/latest/index.html).
+
+### Up and run (fast but not recommended)
+- `docker run -v "$PWD:/workspace" --gpus all  --rm -it pytorch/pytorch:2.4.1-cuda12.4-cudnn9-runtime bash`
+- `poetry install`
+- `poetry shell`
+
+### Using Conda-based environment
+- Note: Docker will use `micromamba` instead of `miniconda`. Replace `conda` with `micromamba` in your usual commands
+- Edit `.env` locates the same level with `run_docker.sh`, to add environment variables to the prospective docker container
+- There is a file named `run_docker.sh`, allow to execute it by `chmod +x run_docker.sh` and run `run_docker.sh`
+- Enjoy Jupyter lab at localhost:8888 as usual. Notebook token is shown after `run_docker.sh` runs successfully
+
+## Isolate Environment without Docker
 ### With `poetry`
 - `pip install poetry`
 - Create, install, activate environment
 ```console
-poetry install --with cpu # cpu
-poetry install --with cu117 # cuda 11.7
+poetry install
 poetry shell
 ```
 - Need to update environment after `poetry add a_lib`
@@ -60,21 +75,15 @@ conda env update --file binder/environment.yml --prune
 conda env export --from-history -f binder/environment.yml
 ```
 
-### With `docker`
-- This tutorial is for those who have NVIDIA GPU (hereafter GPU). For CPU case, this should be similar but need to adjust `Dockerfile` and `run_docker.sh`
-- Note: Docker will use `micromamba` instead of `miniconda`. Replace `conda` with `micromamba` in your usual commands
-- You need to install `docker`
-- Install Nvidia driver (ignore if you don't have GPU)
-- Then install [Nvidia docker container toolkit](https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/latest/index.html), ignore if you aim to use CPU.
-- Edit `.env` locates the same level with `run_docker.sh`, to add environment variables to the prospective docker container
-- There is a file named `run_docker.sh`, allow to execute it by `chmod +x run_docker.sh` and run `run_docker.sh`
-- Enjoy Jupyter lab at localhost:8888 as usual. Notebook token is shown after `run_docker.sh` runs successfully
-
 ## After installing libraries, verify
+
 try this in `ipython`
 ```python
 import torch
 from torch_geometric.data import Data
+from transformers import pipeline
+
+print(torch.cuda.is_available())
 
 edge_index = torch.tensor([[0, 1],
                            [1, 0],
@@ -83,21 +92,9 @@ edge_index = torch.tensor([[0, 1],
 x = torch.tensor([[-1], [0], [1]], dtype=torch.float)
 
 data = Data(x=x, edge_index=edge_index.t().contiguous())
-```
-
-## Q&A
-### My default Python is not 3.11, how I can instruct `poetry` to use Python 3.11
-Use `poetry env use` to select Python version, more details are at more details https://stackoverflow.com/questions/60580113/change-python-version-to-3-x
-### I don't want to use Python 3.11, how to change configs and make it reproducible 
-If you aim to use `poetry`, the steps are following
-- edit file `pyproject.toml`
-- select a Python version, then `poetry shell`
-- generate new `poetry.lock` by run `poetry lock`
-
-If you aim to follow `conda`
-- edit file `environment.yml`
-- create a new environment that has the version you want
-- switch to that environment
+print(data)
 
-## Current limitation
-- `torch_geometric` with CPU version in `poetry` has a problem. I created a discussion at https://github.com/pyg-team/pytorch_geometric/discussions/7788
+classifier = pipeline("sentiment-analysis")
+result = classifier("We are very happy to show you the 🤗 Transformers library.")
+print(result)
+```

environment.yml

@@ -1,29 +1,41 @@
 name: ml-venv
+channel_priority: disabled
 channels:
+  - pytorch
+  - nvidia
+  - pyg
   - conda-forge
-  - defaults
 dependencies:
-  - python=3.11
+  - python=3.12
+  - pip
+  - blas
+  - networkx
   - numpy
+  - scipy
   - pandas
+  - scikit-learn
   - matplotlib
-  - pip
   - jupyterlab
   - ipywidgets
-  - scipy
-  - scikit-learn
   - panel
   - pytest
+  - pytorch::pytorch=2.4.1
+  - pytorch::pytorch-cuda=12.4
+  - pytorch::torchvision=0.19.1
+  - pytorch::torchaudio=2.4.1
+  - pyg=2.6.1
+  - transformers=4.46.0
+  - sentencepiece
+  - lightning
+  - click
+  - wandb
+  - dvc
+  - torchmetrics
+  - evaluate
   - pip:
-    - click
-    - torch
-    - torchvision 
-    - torchaudio
-    - transformers
-    - pytorch-lightning
-    - wandb
-    - torch_geometric
-    - torchmetrics
-    - accelerate
-    - datasets
-    - dvc
+    - pyg_lib==0.4.0+pt24cu124
+    - torch_scatter==2.1.2+pt24cu124
+    - torch_sparse==0.6.18+pt24cu124
+    - torch_cluster==1.6.3+pt24cu124
+    - torch_spline_conv==1.2.2+pt24cu124
+    - -f https://data.pyg.org/whl/torch-2.4.0+cu124.html