diff --git a/README.md b/README.md
index b68314f297..ec6ab30aad 100644
--- a/README.md
+++ b/README.md
@@ -40,9 +40,14 @@ The Guiding Principles when building `torchtitan`
 * Minimal changes to the model code when applying multi-dimensional parallelism.
 * Bias towards a clean, minimal codebase while providing basic reusable / swappable components.
 
-`torchtitan` has been showcasing PyTorch's latest distributed training features, via pretraining Llama 3.1 LLMs of various sizes.
-To accelerate contributions to and innovations around torchtitan, we host an [`experiments`](torchtitan/experiments) folder. We look forward to your contributions!
+`torchtitan` has been showcasing PyTorch's latest distributed training features, via support for pretraining Llama 3.1 LLMs of various sizes.
 
+## Contributing
+
+We look forward to your contributions!
+
+* To accelerate contributions to and innovations around torchtitan, we host an [`experiments`](torchtitan/experiments) folder. New ideas should start there. To contribute, follow the [`experiments guidelines`](torchtitan/experiments/README.md).
+* For fixes and contributions to core, follow these [`guidelines`](CONTRIBUTING.md).
 
 ## Llama 3.1 training
 
diff --git a/torchtitan/experiments/README.md b/torchtitan/experiments/README.md
index 6442ea71f8..10b90ac1d4 100644
--- a/torchtitan/experiments/README.md
+++ b/torchtitan/experiments/README.md
@@ -10,7 +10,7 @@ We provide this `experiments/` folder to host experiments that add significant v
 3. An experiment should reuse existing `torchtitan` code as much as possible, such as modules in [`components/`](../components/) (via a new [`TrainSpec`](../protocols/train_spec.py)) and [`train.py`](../train.py). For a list of extension points we provide, please refer to [docs/extension.md](../../docs/extension.md).
     - The extension points are subject to change. We kindly request that contributors provide feedback if they encounter issues reusing any components, rather than simply using a copy-and-paste approach.
     - The degree to which existing components are reused and whether duplications are legit will also be a criteria of whether an experiment would be accepted.
-4. Each experiment is independent from other experiments, and can have its own dependencies (on top of [core dependencies](../../requirements.txt)), and its own tests.
+4. Each experiment is independent from other experiments, and can have its own dependencies (on top of [core dependencies](../../requirements.txt)), and its own tests. An experiment should not contain vendor-specific code, such as kernels written in a proprietary language. Those can be hosted outside as dependency.
 5. The dependency from `experiments` to `core` is one-way. Anything in `experiments` is optional for `core` to run successfully. In particular, development in `core` is not blocked by breakage in `experiments`. We will utilize GitHub's [CI mechanism](https://docs.github.com/en/actions/writing-workflows/workflow-syntax-for-github-actions#onpushpull_requestpull_request_targetpathspaths-ignore) to help test an experiment periodically and only if the experiment itself is affected by a PR.
 6. Each experiment needs to have an owner. The owner is responsible to work with `torchtitan` team to maintain the quality and healthiness of an experiment, which includes
     - adapting an experiment to changes in `core` and fix broken tests, no later than the next official `torchtitan` release;