Simple Summarization on DeepSeek-R1

RL is key
↳ but hard to make it helpful alone.
↳ 4 stage pipeline (good start + reasoning RL + SFT + safety RL) = o1 level performance.
↳ Distilling R1-Zero outputs = o1-mini level.

Model: huggingface.co/deepseek-ai
Paper: github.com/deepseek-ai/...

and you can also select and use Gemini, Mistral, and LLaMA as a generative model.

Out-of-the-box data sources include Local, Google Cloud Storage, Google Drive, Slack, Jira, making it easy to create PoCs for a wide range of use cases.

For example, you can select and use GCP's { RagManagedDb, Vector Search, Feature Store } or third-party { Weaviate, pinecone } as underlying DB. In addition, you can select GCP's { text-embedding, gecko } or the open source model { e5-base | large | small } as an embedding model,

You can configure the desired RAG pipeline with various combinations, and you can also use the backend service developed and provided by Google.

Vertex AI RAG Engine: A developers tool Build robust and grounded generative AI applications with Vertex AI RAG Engine, reducing hallucinations and enhancing accuracy.

Google's Vertex AI RAG Engine

Google launched a RAG-specific service called "Vertex AI RAG Engine." It can be understood as providing infrastructure for RAG on the Google Cloud Platform and supporting libraries that can be easily utilized.

developers.googleblog.com/en/vertex-ai...

blog on Hugging Face Daily Papers that is updated on a daily basis
: deep-diver.github.io/ai-paper-rev...

core project (paper-reviewer)
: github.com/deep-diver/p...

Please give ⭐️ to reach 700 on GitHub!!

updates on ai-paper-reviewer!

core
✦ supporting open source Layout Parsing model from
@OpenDataLab_AI

✦ scrapping papers from
@openreviewnet

blog
✦ display papers by the dates added in
@huggingface
Daily Papers. Up to 3 latest days are managed, then archived

link 👇

I share these kinda contents that I actually build myself with collaborators.

If you are curious and want to know what's coming, please follow me!

Cheers 🍻

GitHub - deep-diver/paper-reviewer: Generate a comprehensive review from an arXiv paper, then turn it into a blog post. This project powers the website below for the HuggingFace's Daily Papers (https:... Generate a comprehensive review from an arXiv paper, then turn it into a blog post. This project powers the website below for the HuggingFace's Daily Papers (https://huggingface.co/papers). - d...

And this project got 550 @github.com 🌟 in a month. Notably, it comes with audio podcast for every papers whose quality is quite comparable to NotebookLM.

github.com/deep-diver/p...

LlamaDuo: LLMOps Pipeline for Seamless Migration from Service LLMs to Small-Scale Local LLMs The widespread adoption of cloud-based proprietary large language models (LLMs) has introduced significant challenges, including operational dependencies, privacy concerns, and the necessity of contin...

My first ever full paper in the field of AI. This is quite unique exp since I am not ML background at all.

arxiv.org/abs/2408.13467

I am Chansung. I love collab with others for building cool AI project and writing paper

Recent ones 👇
1. Paper on @arxiv

LlamaDuo: LLMOps Pipeline for Seamless Migration from Service LLMs to Small-Scale Local LLMs

2. OSS
AI Paper Reviewer: gen text and poscast of papers

Find links below 🔗