<!--Copyright 2024 The HuggingFace Team. All rights reserved.
Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
the License. You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
specific language governing permissions and limitations under the License.
-->
# Overview
Generating high-quality outputs is computationally intensive, especially during each iterative step where you go from a noisy output to a less noisy output. One of 🤗 Diffuser's goals is to make this technology widely accessible to everyone, which includes enabling fast inference on consumer and specialized hardware.
This section will cover tips and tricks - like half-precision weights and sliced attention - for optimizing inference speed and reducing memory-consumption. You'll also learn how to speed up your PyTorch code with [`torch.compile`](https://pytorch.org/tutorials/intermediate/torch_compile_tutorial.html) or [ONNX Runtime](https://onnxruntime.ai/docs/), and enable memory-efficient attention with [xFormers](https://facebookresearch.github.io/xformers/). There are also guides for running inference on specific hardware like Apple Silicon, and Intel or Habana processors.
[Molecule conformation](https://www.nature.com/subjects/molecular-conformation#:~:text=Definition,to%20changes%20in%20their%20environment.) generation | [](https://colab.research.google.com/github/huggingface/notebooks/blob/main/diffusers/geodiff_molecule_conformation.ipynb) | ❌
More coming soon!
\ No newline at end of file
* Pipeline functionality: these techniques modify the pipeline or extend it for other applications. For example, pipeline callbacks add new features to a pipeline and a pipeline can also be extended for distributed inference.
* Improve inference quality: these techniques increase the visual quality of the generated images. For example, you can enhance your prompts with GPT2 to create better images with lower effort.
<!--Copyright 2024 The HuggingFace Team. All rights reserved.
Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
the License. You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
specific language governing permissions and limitations under the License.
-->
# Overview
A pipeline is an end-to-end class that provides a quick and easy way to use a diffusion system for inference by bundling independently trained models and schedulers together. Certain combinations of models and schedulers define specific pipeline types, like [`StableDiffusionXLPipeline`] or [`StableDiffusionControlNetPipeline`], with specific capabilities. All pipeline types inherit from the base [`DiffusionPipeline`] class; pass it any checkpoint, and it'll automatically detect the pipeline type and load the necessary components.
This section demonstrates how to use specific pipelines such as Stable Diffusion XL, ControlNet, and DiffEdit. You'll also learn how to use a distilled version of the Stable Diffusion model to speed up inference, how to create reproducible pipelines, and how to use and contribute community pipelines.
> This notebook is not actively maintained by the Diffusers team. For any questions or comments, please contact [natolambert](https://twitter.com/natolambert).
This is an experimental research notebook demonstrating how to generate stable 3D structures of molecules with [GeoDiff](https://github.com/MinkaiXu/GeoDiff) and Diffusers.
"This colab is design to run the pretrained models from [GeoDiff](https://github.com/MinkaiXu/GeoDiff).\n",
"The visualization code is inspired by this PyMol [colab](https://colab.research.google.com/gist/iwatobipen/2ec7faeafe5974501e69fcc98c122922/pymol.ipynb#scrollTo=Hm4kY7CaZSlw).\n",
"\n",
"The goal is to generate physically accurate molecules. Given the input of a molecule graph (atom and bond structures with their connectivity -- in the form of a 2d graph). What we want to generate is a stable 3d structure of the molecule.\n",
"\n",
"This colab uses GEOM datasets that have multiple 3d targets per configuration, which provide more compelling targets for generative methods.\n",
"\n",
"> Colab made by [natolambert](https://twitter.com/natolambert).\n",
"\u001b[33mWARNING: Running pip as the 'root' user can result in broken permissions and conflicting behaviour with the system package manager. It is recommended to use a virtual environment instead: https://pip.pypa.io/warnings/venv\u001b[0m\u001b[33m\n",
"\u001b[2K \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m757.0/757.0 kB\u001b[0m \u001b[31m52.8 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
"\u001b[2K \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m163.5/163.5 kB\u001b[0m \u001b[31m21.9 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
"\u001b[2K \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m40.8/40.8 kB\u001b[0m \u001b[31m5.5 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
"\u001b[2K \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m596.3/596.3 kB\u001b[0m \u001b[31m51.7 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
"\u001b[?25h Building wheel for diffusers (pyproject.toml) ... \u001b[?25l\u001b[?25hdone\n",
"\u001b[33mWARNING: Running pip as the 'root' user can result in broken permissions and conflicting behaviour with the system package manager. It is recommended to use a virtual environment instead: https://pip.pypa.io/warnings/venv\u001b[0m\u001b[33m\n",
"\u001b[2K \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m432.7/432.7 kB\u001b[0m \u001b[31m36.8 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
"\u001b[2K \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m5.3/5.3 MB\u001b[0m \u001b[31m90.9 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
"\u001b[2K \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m35.3/35.3 MB\u001b[0m \u001b[31m39.7 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
"\u001b[2K \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m115.1/115.1 kB\u001b[0m \u001b[31m16.3 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
"\u001b[2K \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m948.0/948.0 kB\u001b[0m \u001b[31m63.6 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
"\u001b[2K \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m212.2/212.2 kB\u001b[0m \u001b[31m21.5 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
"\u001b[2K \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m95.8/95.8 kB\u001b[0m \u001b[31m12.8 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
"\u001b[2K \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m140.8/140.8 kB\u001b[0m \u001b[31m18.8 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
"\u001b[2K \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m7.6/7.6 MB\u001b[0m \u001b[31m104.3 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
"\u001b[2K \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m148.0/148.0 kB\u001b[0m \u001b[31m20.8 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
"\u001b[2K \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m231.3/231.3 kB\u001b[0m \u001b[31m30.0 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
"\u001b[2K \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m94.8/94.8 kB\u001b[0m \u001b[31m14.0 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
"\u001b[2K \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m58.8/58.8 kB\u001b[0m \u001b[31m8.0 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
"\u001b[?25h\u001b[33mWARNING: Running pip as the 'root' user can result in broken permissions and conflicting behaviour with the system package manager. It is recommended to use a virtual environment instead: https://pip.pypa.io/warnings/venv\u001b[0m\u001b[33m\n",
"\u001b[0m"
]
}
],
"source": [
"%cd /content\n",
"\n",
"# install latest HF diffusers (will update to the release once added)\n",
"\u001b[2K \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m36.8/36.8 MB\u001b[0m \u001b[31m34.6 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
"\u001b[?25hRequirement already satisfied: Pillow in /usr/local/lib/python3.7/site-packages (from rdkit) (9.2.0)\n",
"Requirement already satisfied: numpy in /usr/local/lib/python3.7/site-packages (from rdkit) (1.21.6)\n",
"Installing collected packages: rdkit\n",
"Successfully installed rdkit-2022.3.5\n",
"\u001b[33mWARNING: Running pip as the 'root' user can result in broken permissions and conflicting behaviour with the system package manager. It is recommended to use a virtual environment instead: https://pip.pypa.io/warnings/venv\u001b[0m\u001b[33m\n",
"\u001b[0m"
]
}
],
"source": [
"!pip install rdkit"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "88GaDbDPxJ5I"
},
"source": [
"### Get viewer from nglview\n",
"\n",
"The model you will use outputs a position matrix tensor. This pytorch geometric data object will have many features (positions, known features, edge features -- all tensors).\n",
"The data we give to the model will also have a rdmol object (which can extract features to geometric if needed).\n",
"The rdmol in this object is a source of ground truth for the generated molecules.\n",
"\n",
"You will use one rendering function from nglviewer later!\n",
"Requirement already satisfied: pyparsing!=3.0.5,>=2.0.2 in /usr/local/lib/python3.7/site-packages (from packaging->ipykernel>=4.5.1->ipywidgets>=7->nglview) (3.0.9)\n",
"Requirement already satisfied: six>=1.5 in /usr/local/lib/python3.7/site-packages (from python-dateutil>=2.8.2->jupyter-client>=6.1.12->ipykernel>=4.5.1->ipywidgets>=7->nglview) (1.16.0)\n",
"Building wheels for collected packages: nglview\n",
" Building wheel for nglview (pyproject.toml) ... \u001b[?25l\u001b[?25hdone\n",
" Created wheel for nglview: filename=nglview-3.0.3-py3-none-any.whl size=8057538 sha256=b7e1071bb91822e48515bf27f4e6b197c6e85e06b90912b3439edc8be1e29514\n",
" Stored in directory: /root/.cache/pip/wheels/01/0c/49/c6f79d8edba8fe89752bf20de2d99040bfa57db0548975c5d5\n",
"\u001b[33mWARNING: Running pip as the 'root' user can result in broken permissions and conflicting behaviour with the system package manager. It is recommended to use a virtual environment instead: https://pip.pypa.io/warnings/venv\u001b[0m\u001b[33m\n",
"\u001b[0m"
]
},
{
"output_type": "display_data",
"data": {
"application/vnd.colab-display-data+json": {
"pip_warning": {
"packages": [
"pexpect",
"pickleshare",
"wcwidth"
]
}
}
},
"metadata": {}
}
],
"source": [
"!pip install nglview"
]
},
{
"cell_type": "markdown",
"source": [
"## Create a diffusion model"
],
"metadata": {
"id": "8t8_e_uVLdKB"
}
},
{
"cell_type": "markdown",
"source": [
"### Model class(es)"
],
"metadata": {
"id": "G0rMncVtNSqU"
}
},
{
"cell_type": "markdown",
"source": [
"Imports"
],
"metadata": {
"id": "L5FEXz5oXkzt"
}
},
{
"cell_type": "code",
"source": [
"# Model adapted from GeoDiff https://github.com/MinkaiXu/GeoDiff\n",
"# Model inspired by https://github.com/DeepGraphLearning/torchdrug/tree/master/torchdrug/models\n",
"The config attributes {'type': 'diffusion', 'network': 'dualenc', 'beta_schedule': 'sigmoid', 'beta_start': 1e-07, 'beta_end': 0.002, 'num_diffusion_timesteps': 5000} were passed to MoleculeGNN, but are not expected and will be ignored. Please verify your config.json configuration file.\n",
"Some weights of the model checkpoint at fusing/gfn-molecule-gen-drugs were not used when initializing MoleculeGNN: ['betas', 'alphas']\n",
"- This IS expected if you are initializing MoleculeGNN from the checkpoint of a model trained on another task or with another architecture (e.g. initializing a BertForSequenceClassification model from a BertForPreTraining model).\n",
"- This IS NOT expected if you are initializing MoleculeGNN from the checkpoint of a model that you expect to be exactly identical (initializing a BertForSequenceClassification model from a BertForSequenceClassification model).\n"
"Grab a google tool so we can upload our data directly. Note you need to download the data from ***this [file](https://huggingface.co/datasets/fusing/geodiff-example-data/blob/main/data/molecules.pkl)***\n",
"\n",
"(direct downloading from the hub does not yet work for this datatype)"
"/usr/local/lib/python3.7/dist-packages/ipykernel_launcher.py:4: UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.clone().detach() or sourceTensor.clone().detach().requires_grad_(True), rather than torch.tensor(sourceTensor).\n",
