Tips to upload large models/datasets

docs/source/guides/repository.md

@@ -71,7 +71,7 @@ Specify the `repo_id` of the repository you want to delete:
 >>> delete_repo(repo_id="lysandre/my-corrupted-dataset", repo_type="dataset")
 ```
 
-### Clone a repository (only for Spaces)
+### Duplicate a repository (only for Spaces)
 
 In some cases, you want to copy someone else's repo to adapt it to your use case.
 This is possible for Spaces using the [`duplicate_space`] method. It will duplicate the whole repository.

docs/source/guides/upload.md

@@ -106,7 +106,13 @@ but before that, all previous logs on the repo on deleted. All of this in a sing
 ... )
 ```
 
-## Non-blocking upload
+## Advanced features
+
+In most cases, you won't need more than [`upload_file`] and [`upload_folder`] to upload your files to the Hub.
+However, `huggingface_hub` has more advanced features to make things easier. Let's have a look at them!
+
+
+### Non-blocking uploads
 
 In some cases, you want to push data without blocking your main thread. This is particularly useful to upload logs and
 artifacts while continuing a training. To do so, you can use the `run_as_future` argument in both [`upload_file`] and
@@ -154,7 +160,7 @@ Future(...)
 Future(...)
 ```
 
-## Upload a folder by chunks
+### Upload a folder by chunks
 
 [`upload_folder`] makes it easy to upload an entire folder to the Hub. However, for large folders (thousands of files or
 hundreds of GB), it can still be challenging. If you have a folder with a lot of files, you might want to upload
@@ -193,7 +199,7 @@ notice.
 
 </Tip>
 
-## Scheduled uploads
+### Scheduled uploads
 
 The Hugging Face Hub makes it easy to save and version data. However, there are some limitations when updating the same file thousands of times. For instance, you might want to save logs of a training process or user
 feedback on a deployed Space. In these cases, uploading the data as a dataset on the Hub makes sense, but it can be hard to do properly. The main reason is that you don't want to version every update of your data because it'll make the git repository unusable. The [`CommitScheduler`] class offers a solution to this problem.
@@ -262,14 +268,14 @@ For more details about the [`CommitScheduler`], here is what you need to know:
     `scheduler.lock` lock to ensure thread-safety. The lock is blocked only when the scheduler scans the folder for
     changes, not when it uploads data. You can safely assume that it will not affect the user experience on your Space.
 
-### Space persistence demo
+#### Space persistence demo
 
 Persisting data from a Space to a Dataset on the Hub is the main use case for [`CommitScheduler`]. Depending on the use
 case, you might want to structure your data differently. The structure has to be robust to concurrent users and
 restarts which often implies generating UUIDs. Besides robustness, you should upload data in a format readable by the 🤗 Datasets library for later reuse. We created a [Space](https://huggingface.co/spaces/Wauplin/space_to_dataset_saver)
 that demonstrates how to save several different data formats (you may need to adapt it for your own specific needs).
 
-### Custom uploads
+#### Custom uploads
 
 [`CommitScheduler`] assumes your data is append-only and should be uploading "as is". However, you
 might want to customize the way data is uploaded. You can do that by creating a class inheriting from [`CommitScheduler`]
@@ -316,7 +322,7 @@ containing different implementations depending on your use cases.
 
 </Tip>
 
-## create_commit
+### create_commit
 
 The [`upload_file`] and [`upload_folder`] functions are high-level APIs that are generally convenient to use. We recommend
 trying these functions first if you don't need to work at a lower level. However, if you want to work at a commit-level,
@@ -371,10 +377,91 @@ In addition to [`upload_file`] and [`upload_folder`], the following functions al
 
 For more detailed information, take a look at the [`HfApi`] reference.
 
-## Push files with Git LFS
+## Tips and tricks for large uploads
+
+There are some limitations to be aware of when dealing with a large amount of data in your repo. Given the time it takes to stream the data,
+getting an upload/push to fail at the end of the process or encountering a degraded experience, be it on hf.co or when working locally, can be very annoying.
+We gathered a list of tips and recommendations for structuring your repo.
+
+
+| Characteristic     | Recommended        | Tips                                     |
+| ----------------   | ------------------ | ---------------------------------------- |
+| Repo size          | -                  | contact us for large repos (TBs of data) |
+| Files per repo     | <100k              | merge data into fewer files              |
+| Entries per folder | <10k               | use subdirectories in repo               |
+| File size          | <5GB               | split data into chunked files            |
+| Commit size        | <100 files*        | upload files in multiple commits         |
+| Commits per repo   | -                  | upload multiple files per commit         |
+
+_* Not relevant when using `git` CLI directly_
+
+Please read the next section to understand better those limits and how to deal with them.
+
+### Hub repository size limitations
+
+What are we talking about when we say "large uploads", and what are their associated limitations? Large uploads can be
+very diverse, from repositories with a few huge files (e.g. model weights) to repositories with thousands of small files
+(e.g. an image dataset).
+
+Under the hood, the Hub uses Git to version the data, which has structural implications on what you can do in your repo.
+If your repo is crossing some of the numbers mentioned in the previous section, **we strongly encourage you to check out [`git-sizer`](https://github.com/github/git-sizer)**,
+which has very detailed documentation about the different factors that will impact your experience. Here is a TL;DR of factors to consider:
+
+- **Repository size**: The total size of the data you're planning to upload. There is no hard limit on a Hub repository size. However, if you plan to upload hundreds of GBs or even TBs of data, we would appreciate it if you could let us know in advance so we can better help you if you have any questions during the process. You can contact us at datasets@huggingface.co or on [our Discord](http://hf.co/join/discord).
+- **Number of files**:
+    - For optimal experience, we recommend keeping the total number of files under 100k. Try merging the data into fewer files if you have more.
+      For example, json files can be merged into a single jsonl file, or large datasets can be exported as Parquet files.
+    - The maximum number of files per folder cannot exceed 10k files per folder. A simple solution is to
+      create a repository structure that uses subdirectories. For example, a repo with 1k folders from `000/` to `999/`, each containing at most 1000 files, is already enough.
+- **File size**: In the case of uploading large files (e.g. model weights), we strongly recommend splitting them **into chunks of around 5GB each**.
+There are a few reasons for this:
+    - Uploading and downloading smaller files is much easier both for you and the other users. Connection issues can always
+      happen when streaming data and smaller files avoid resuming from the beginning in case of errors.
+    - Files are served to the users using CloudFront. From our experience, huge files are not cached by this service
+      leading to a slower download speed.
+- **Number of commits**: There is no hard limit for the total number of commits on your repo history. However, from
+our experience, the user experience on the Hub starts to degrade after a few thousand commits. We are constantly working to
+improve the service, but one must always remember that a git repository is not meant to work as a database with a lot of
+writes. If your repo's history gets very large, it is always possible to squash all the commits to get a
+fresh start.
+- **Number of operations per commit**: Once again, there is no hard limit here. When a commit is uploaded on the Hub, each
+git operation (addition or delete) is checked by the server. When a hundred LFS files are committed at once,
+each file is checked individually to ensure it's been correctly uploaded. When pushing data through HTTP with `huggingface_hub`,
+a timeout of 60s is set on the request, meaning that if the process takes more time, an error is raised
+client-side. However, it can happen (in rare cases) that even if the timeout is raised client-side, the process is still
+completed server-side. This can be checked manually by browsing the repo on the Hub. To prevent this timeout, we recommend
+adding around 50-100 files per commit.
+
+### Practical tips
+
+Now that we've seen the technical aspects you must consider when structuring your repository, let's see some practical
+tips to make your upload process as smooth as possible.
+
+- **Start small**: We recommend starting with a small amount of data to test your upload script. It's easier to iterate
+on a script when failing takes only a little time.
+- **Expect failures**: Streaming large amounts of data is challenging. You don't know what can happen, but it's always
+best to consider that something will fail at least once -no matter if it's due to your machine, your connection, or our
+servers. For example, if you plan to upload a large number of files, it's best to keep track locally of which files you
+already uploaded before uploading the next batch. You are ensured that an LFS file that is already committed will never
+be re-uploaded twice but checking it client-side can still save some time.
+- **Use `hf_transfer`**: this is a Rust-based [library](https://github.com/huggingface/hf_transfer) meant to speed up
+uploads on machines with very high bandwidth. To use it, you must install it (`pip install hf_transfer`) and enable it
+by setting `HF_HUB_ENABLE_HF_TRANSFER=1` as an environment variable. You can then use `huggingface_hub` normally.
+Disclaimer: this is a power user tool. It is tested and production-ready but lacks user-friendly features like progress
+bars or advanced error handling.
+
+## (legacy) Upload files with Git LFS
 
 All the methods described above use the Hub's API to upload files. This is the recommended way to upload files to the Hub.
-However we also provide [`Repository`], a wrapper around the git tool to manage a local repository.
+However, we also provide [`Repository`], a wrapper around the git tool to manage a local repository.
+
+<Tip warning={true}>
+
+Although [`Repository`] is not formally deprecated, we recommend using the HTTP-based methods described above instead.
+For more details about this recommendation, please have a look at [this guide](../concepts/git_vs_http) explaining the
+core differences between HTTP-based and Git-based approaches.
+
+</Tip>
 
 Git LFS automatically handles files larger than 10MB. But for very large files (>5GB), you need to install a custom transfer agent for Git LFS: