for relaxing dependency constraints for better usability

INSTALL.md

@@ -9,7 +9,21 @@ cd cosmos-predict1
 Cosmos runs only on Linux systems. We have tested the installation with Ubuntu 24.04, 22.04, and 20.04.
 Cosmos requires the Python version to be `3.10.x`. Please also make sure you have `conda` installed ([instructions](https://docs.conda.io/projects/conda/en/latest/user-guide/install/index.html)).
 
-### Inference
+### Dependency Management
+
+This project defines its dependencies in two files to support different use cases:
+
+- **`requirements.txt`**: This file contains pinned dependencies with exact versions. Use this to create a reproducible environment that mirrors the setup used for the original research. This is the recommended approach for running the provided examples or for post-training to ensure compatibility.
+
+- **`pyproject.toml`**: This file specifies flexible dependency version ranges. This is for users who want to integrate `cosmos-predict1` as a library into a larger Python project. This approach minimizes the chance of conflicts with other packages in your environment.
+
+Choose the installation method below that best suits your needs.
+
+### For Reproducing the Research Environment
+
+These instructions are for setting up a development environment that matches the one used for the research paper.
+
+#### Inference
 
 The below commands creates the `cosmos-predict1` conda environment and installs the dependencies for inference:
 ```bash
@@ -44,7 +58,7 @@ You can test the environment setup for inference with
 CUDA_HOME=$CONDA_PREFIX PYTHONPATH=$(pwd) python scripts/test_environment.py
 ```
 
-### Post-training
+#### Post-training
 
 The below commands creates the `cosmos-predict1` conda environment and installs the dependencies for post-training. This is the same as required for inference but with an additional package `apex` for training with bfloat16.
 ```bash
@@ -68,3 +82,14 @@ You can test the environment setup for post-training with
 ```bash
 CUDA_HOME=$CONDA_PREFIX PYTHONPATH=$(pwd) python scripts/test_environment.py --training
 ```
+
+### For Using as a Library
+
+If you want to use `cosmos-predict1` in your own project, you can install it from the cloned repository. This will use the flexible dependencies from `pyproject.toml`.
+
+```bash
+# From the root of the cosmos-predict1 repository
+pip install .
+```
+
+This will install `cosmos-predict1` and its dependencies. Note that this method may pull different dependency versions than what is in `requirements.txt`, which may be necessary for compatibility with your other project dependencies.

README.md

@@ -40,7 +40,7 @@ We provide a comphrehensive set of examples to illustrate how to perform inferen
 
 ### Installation
 
-Please refer to [INSTALL.md](INSTALL.md) for general instructions on environment setup.
+Please refer to [INSTALL.md](INSTALL.md) for detailed instructions on environment setup and dependency management. This document explains how to install for both reproducibility and for use as a library.
 
 ### Inference with pre-trained Cosmos-Predict1 models
 * [Inference with diffusion-based Text2World models](/examples/inference_diffusion_text2world.md) **[with multi-GPU support]**

pyproject.toml

@@ -33,15 +33,47 @@ description = "Cosmos World Foundation Model"
 readme = "README.md"
 requires-python = ">=3.10"
 license = {text = "Apache-2.0"}
-dynamic = ["dependencies"]
+dependencies = [
+    "attrs>=25.1.0",
+    "better-profanity>=0.7.0",
+    "boto3>=1.35.99,<2.0",
+    "decord>=0.6.0",
+    "diffusers>=0.32.2",
+    "einops>=0.8.1",
+    "huggingface-hub>=0.29.2",
+    "hydra-core>=1.3.2,<2.0",
+    "imageio[pyav,ffmpeg]>=2.37.0",
+    "iopath>=0.1.10",
+    "ipdb>=0.13.13",
+    "loguru>=0.7.2",
+    "mediapy>=1.2.2,<2.0",
+    "megatron-core>=0.10.0",
+    "nltk>=3.9.1,<4.0",
+    "numpy>=1.26.4,<2.0",
+    "nvidia-ml-py>=12.535.133,<13.0",
+    "omegaconf>=2.3.0",
+    "opencv-python>=4.10.0.84,<5.0",
+    "pandas>=2.2.3,<3.0",
+    "peft>=0.14.0",
+    "pillow>=11.1.0",
+    "protobuf>=4.25.3,<5.0",
+    "pynvml>=12.0.0",
+    "pyyaml>=6.0.2,<7.0",
+    "retinaface-py>=0.0.2",
+    "safetensors==0.5.3",
+    "scikit-image>=0.25.2",
+    "sentencepiece>=0.2.0",
+    "termcolor>=2.5.0,<3.0",
+    "torch>=2.6.0,<3.0",
+    "torchvision>=0.21.0",
+    "tqdm>=4.66.5,<5.0",
+    "transformers>=4.49.0,<5.0",
+]
 
 [project.urls]
 homepage = "https://github.com/nvidia-cosmos/cosmos-predict1/"
 issues = "https://github.com/nvidia-cosmos/cosmos-predict1/issues"
 
-[tool.setuptools.dynamic]
-dependencies = {file = ["requirements.txt"]}
-
 [tool.setuptools]
 packages = ["cosmos_predict1"]