1FileLLM is a command-line tool designed to streamline the creation of information-dense prompts for large language models (LLMs). It aggregates and preprocesses data from a variety of sources, compiling them into a single text file that is automatically copied to your clipboard for quick use.
- Automatic source type detection based on provided path, URL, or identifier
- Support for local files and/or directories, GitHub repositories, GitHub pull requests, GitHub issues, academic papers from ArXiv, YouTube transcripts, web page documentation, Sci-Hub hosted papers via DOI or PMID
- Handling of multiple file formats, including Jupyter Notebooks (.ipynb), and PDFs
- Web crawling functionality to extract content from linked pages up to a specified depth
- Integration with Sci-Hub for automatic downloading of research papers using DOIs or PMIDs
- Text preprocessing, including compressed and uncompressed outputs, stopword removal, and lowercase conversion
- Automatic copying of uncompressed text to the clipboard for easy pasting into LLMs
- Token count reporting for both compressed and uncompressed outputs
- XML encapsulation of output for improved LLM performance
+--------------------------------+
| External Services |
|--------------------------------|
| GitHub API | YouTube API |
| Sci-Hub | ArXiv |
+--------------------------------+
|
|
v
+----------------------+ +---------------------+ +----------------------+
| | | | | |
| User | | Command Line Tool | | External Libraries |
|----------------------| |---------------------| |----------------------|
| - Provides input URL |--------->| - Handles user input| | - Requests |
| | | - Detects source |<--------| - BeautifulSoup |
| - Receives text | | type | | - PyPDF2 |
| in clipboard |<---------| - Calls appropriate | | - Tiktoken |
| | | processing modules| | - NLTK |
+----------------------+ | - Preprocesses text | | - Nbformat |
| - Generates output | | - Nbconvert |
| files | | - YouTube Transcript |
| - Copies text to | | API |
| clipboard | | - Pyperclip |
| - Reports token | | - Wget |
| count | | - Tqdm |
+---------------------+ | - Rich |
| +----------------------+
|
v
+---------------------+
| Source Type |
| Detection |
|---------------------|
| - Determines type |
| of source |
+---------------------+
|
v
+---------------------+
| Processing Modules |
|---------------------|
| - GitHub Repo Proc |
| - Local Dir Proc |
| - YouTube Transcript|
| Proc |
| - ArXiv PDF Proc |
| - Sci-Hub Paper Proc|
| - Webpage Crawling |
| Proc |
+---------------------+
|
v
+---------------------+
| Text Preprocessing |
|---------------------|
| - Stopword removal |
| - Lowercase |
| conversion |
| - Text cleaning |
+---------------------+
|
v
+---------------------+
| Output Generation |
|---------------------|
| - Compressed text |
| file output |
| - Uncompressed text |
| file output |
+---------------------+
|
v
+---------------------+
| Token Count |
| Reporting |
|---------------------|
| - Report token count|
| |
| - Copies text to |
| clipboard |
+---------------------+
- 2024-07-29:
- Updated output format to encapsulate content in XML tags. This change was implemented due to evaluations showing that LLMs perform better with prompts structured in XML.
- Added tests for GitHub issues and GitHub pull requests to improve robustness and reliability.
- Updated various processing functions to return formatted content instead of writing directly to files, improving consistency and testability.
- 2024-05-17: Added ability to pass path or URL as command line argument.
- 2024-05-16: Updated text colors.
- 2024-05-11:
- Updated requirements.txt.
- Added Rich library to
onefilellm.py.
- 2024-04-04:
- Added GitHub PR and issue tests.
- Added GitHub PR and issues.
- Added tests for GitHub PRs and issues.
- Added ability to concatenate specific GitHub issue and repo when GitHub issue URL is passed.
- Updated tests to include pull request changes.
- Added ability to concatenate pull request and repo when GitHub pull request URL is passed.
- 2024-04-03:
- Included the ability to pull a complete GitHub pull request given the GitHub pull request URL.
- Updated
onefilellm.pyto return an error when Sci-hub is inaccessible or no document is found.
Install the required dependencies:
pip install -U -r requirements.txtOptionally, create a virtual environment for isolation:
python -m venv .venv
source .venv/bin/activate
pip install -U -r requirements.txtTo access private GitHub repositories, generate a personal access token as described in the 'Obtaining a GitHub Personal Access Token' section.
Clone the repository or download the source code.
Run the script using the following command:
python onefilellm.py
Or pass the URL or path in at the command line for the same behavior with less human interaction:
python onefilellm.py https://github.com/jimmc414/1filellmThe tool supports the following input options:
- Local file path (e.g., C:\documents\report.pdf)
- Local directory path (e.g., C:\projects\research) -> (files of selected filetypes segmented into one flat text file)
- GitHub repository URL (e.g., https://github.com/jimmc414/onefilellm) -> (Repo files of selected filetypes segmented into one flat text file)
- GitHub pull request URL (e.g., dear-github/dear-github#102) -> (Pull request diff detail and comments and entire repository content concatenated into one flat text file)
- GitHub issue URL (e.g., isaacs/github#1191) -> (Issue details, comments, and entire repository content concatenated into one flat text file)
- ArXiv paper URL (e.g., https://arxiv.org/abs/2401.14295) -> (Full paper PDF to text file)
- YouTube video URL (e.g., https://www.youtube.com/watch?v=KZ_NlnmPQYk) -> (Video transcript to text file)
- Webpage URL (e.g., https://llm.datasette.io/en/stable/) -> (To scrape pages to x depth in segmented text file)
- Sci-Hub Paper DOI (Digital Object Identifier of Sci-Hub hosted paper) (e.g., 10.1053/j.ajkd.2017.08.002) -> (Full Sci-Hub paper PDF to text file)
- Sci-Hub Paper PMID (PubMed Identifier of Sci-Hub hosted paper) (e.g., 29203127) -> (Full Sci-Hub paper PDF to text file)
The tool supports the following input options, with their corresponding output actions. Note that the input file extensions are selected based on the following section of code (Applicable to Repos only):
allowed_extensions = ['.xyz', '.pdq', '.example']The output for all options is encapsulated in LLM prompt-appropriate XML and automatically copied to the clipboard.
-
Local file path
- Example Input:
C:\documents\report.pdf - Output: The contents of the PDF file are extracted and saved into a single text file.
- Example Input:
-
Local directory path
- Example Input:
C:\projects\research - Output: Files of selected file types within the directory are segmented and saved into a single flat text file.
- Example Input:
-
GitHub repository URL
- Example Input:
https://github.com/jimmc414/onefilellm - Output: Repository files of selected file types are segmented and saved into a single flat text file.
- Example Input:
-
GitHub pull request URL
- Example Input:
https://github.com/dear-github/dear-github/pull/102 - Output: Pull request diff details, comments, and the entire repository content are concatenated into a single flat text file.
- Example Input:
-
GitHub issue URL
- Example Input:
https://github.com/isaacs/github/issues/1191 - Output: Issue details, comments, and the entire repository content are concatenated into a single flat text file.
- Example Input:
-
ArXiv paper URL
- Example Input:
https://arxiv.org/abs/2401.14295 - Output: The full paper PDF is converted into a text file.
- Example Input:
-
YouTube video URL
- Example Input:
https://www.youtube.com/watch?v=KZ_NlnmPQYk - Output: The video transcript is extracted and saved into a text file.
- Example Input:
-
Webpage URL
- Example Input:
https://llm.datasette.io/en/stable/ - Output: The webpage content and linked pages up to a specified depth are scraped and segmented into a text file.
- Example Input:
-
Sci-Hub Paper DOI
- Example Input:
10.1053/j.ajkd.2017.08.002 - Output: The full Sci-Hub paper PDF is converted into a text file.
- Example Input:
-
Sci-Hub Paper PMID
- Example Input:
29203127 - Output: The full Sci-Hub paper PDF is converted into a text file.
- Example Input:
The script generates the following output files:
uncompressed_output.txt: The full text output, automatically copied to the clipboard.compressed_output.txt: Cleaned and compressed text.processed_urls.txt: A list of all processed URLs during web crawling.
- To modify the allowed file types for repository processing, update the
allowed_extensionslist in the code. - To change the depth of web crawling, adjust the
max_depthvariable in the code.
To access private GitHub repositories, you need a personal access token. Follow these steps:
- Log in to your GitHub account and go to Settings.
- Navigate to Developer settings > Personal access tokens.
- Click on "Generate new token" and provide a name.
- Select the necessary scopes (at least
repofor private repositories). - Click "Generate token" and copy the token value.
In the onefilellm.py script, replace GITHUB_TOKEN with your actual token or set it as an environment variable:
-
For Windows:
setx GITHUB_TOKEN "YourGitHubToken" -
For Linux:
echo 'export GITHUB_TOKEN="YourGitHubToken"' >> ~/.bashrc source ~/.bashrc
All output is now encapsulated in XML tags. This change was implemented based on evaluations showing that LLMs perform better with prompts structured in XML. The general structure of the output is as follows:
<source type="[source_type]" [additional_attributes]>
<[content_type]>
[Extracted content]
</[content_type]>
</source>Where [source_type] could be one of: "github_repository", "github_pull_request", "github_issue", "arxiv_paper", "youtube_transcript", "web_documentation", "sci_hub_paper", or "local_directory".
This XML structure provides clear delineation of different content types and sources, potentially improving the LLM's understanding and processing of the input.
- For Repos, Modify this line of code to add or remove filetypes processed:
allowed_extensions = ['.py', '.txt', '.js', '.rst', '.sh', '.md', '.pyx', '.html', '.yaml','.json', '.jsonl', '.ipynb', '.h', '.c', '.sql', '.csv'] - For Web scraping, Modify this line of code to change how many links deep from the starting URL to include
max_depth = 2 - Token counts are displayed in the console for both output files.
