Welcome to the reviews repository for HF Daily Papers! This project gathers insights and reviews of machine learning and AI research papers, and presents them in a creative and informative manner via Claude, GPT-4o, FLUX and others.
👉 HFday.ru
- 📚 Classification by topics
- 📅 Sorting by publication date and HF addition date
- 🔄 Syncing every 2 hours
- 💻 Hosted on GitHub
- 🌏 English, Russian, and Chinese
- 📈 Top by week/month (in progress)
Claude and GPT requests, including structured output and embeddings, use OpenRouter through the OpenAI SDK. Set OPENROUTER_API_KEY in your local environment and add a repository Actions secret with the same name in GitHub. Both review workflows use this secret; CLAUDE_KEY and OPENAI_KEY are no longer required.
The daily reviews and topic classification use anthropic/claude-haiku-5.5; user-requested reviews use anthropic/claude-sonnet-5.5 with low reasoning effort. GPT-4o mini workloads use openai/gpt-6-luna, and the GPT-4o fallback uses openai/gpt-6.1-terra, both with reasoning disabled. Embeddings use openai/text-embedding-3-small to preserve compatibility with existing vectors.
Run the manual Check OpenRouter integration workflow to verify all four chat models, structured output, and embeddings with small live requests. Locally, run python -m scripts.check_openrouter with OPENROUTER_API_KEY set.
The project is deployed automatically using GitHub Actions with the following workflow:
- Workflow Name: Issue Doomgrad Paper Review
- Trigger: The workflow runs every 2 hours (
cron: "0 */2 * * *") and can also be triggered manually. - Jobs:
- Checkout Repository Content: The repository content is checked out using
actions/checkout@v2. - Set Up Python: Python 3.9 is set up using
actions/setup-python@v2. - Install Dependencies: Required dependencies are installed using
pip install -r requirements.txt. - Run Python Script: The
main.pyscript is executed to generate the reviews. - Commit Changes: The changes are committed with a timestamped message.
- Push Changes: The changes are pushed back to the GitHub repository.
- Checkout Repository Content: The repository content is checked out using
This automation ensures that the reviews are always up-to-date and available for the audience without manual intervention.
The generated HTML page is deployed at HFday.ru, allowing users to browse the reviews easily.
This project is licensed under the MIT License. Feel free to use and modify it as needed.
Enjoy exploring the world of machine learning papers with our unique, creatively generated reviews! Stay tuned and keep learning! 💾🛠️🧠