Rahul A R @ar_rahular - Twitter Profile

over 2 years ago

Embrace the romance of Valentine's day with Breadcrumbs. From intimate cooking dates in your kitchen to exploring global flavors and personalized nutrition, Breadcrumbs is here to elevate your culinary journey. #ValentinesDay #Cooking

0

1

0

31

ar_rahular retweeted

BreadCrumbsAI @BreadCrumbsAI

over 2 years ago

Discover the Joy of Cooking with BreadCrumbs! Our AI assistant transforms your kitchen into a playground of flavors. Whether you're a novice or a seasoned cook, BreadCrumbs is here to guide, inspire, and simplify your cooking journey.

0

1

0

23

ar_rahular retweeted

Philipp Schmid

@_philschmid

over 2 years ago

Improve LLMs for Retrieval Augement Generation (RAG) with Instruction Tuning for Search 🔍 A new paper defines how to create RAG-friendly datasets to improve LLMs. This approach was used to create INTERS (INstruction Tuning datasEt foR Search). 🚀 INTER Implementation 1️⃣ Collected 43 datasets covering 20 distinct search-related tasks 2️⃣ Manually crafted for each task a description and 12 unique templates. 3️⃣ Templates were used to generate synthetic data using GPT-4 4️⃣ Fine-tuned & evaluated open LLMs on the dataset Insights 🧠 - Includes query understanding, expansion, reformulation, summarization, ranking, and more. 🌟 - Few-shot generally improve performance, especially in complex tasks. 🏆 - Fine-tuned Llama 7B Chat was better than fine-tuned Llama 7B base. 📈 - INTERS improvement TriviaQA by 77.5%, indicates strong RAG skills. 💡 - Could be used to train a smaller query rewrite model. 🥇 - INTER Llama outperforms Llama Chat on all IR related benchmarks.

_philschmid's tweet photo. Improve LLMs for Retrieval Augement Generation (RAG) with Instruction Tuning for Search 🔍
A new paper defines how to create RAG-friendly datasets to improve LLMs. This approach was used to create INTERS (INstruction Tuning datasEt foR Search). 🚀

INTER Implementation
1️⃣ Collected 43 datasets covering 20 distinct search-related tasks
2️⃣ Manually crafted for each task a description and 12 unique templates.
3️⃣ Templates were used to generate synthetic data using GPT-4
4️⃣ Fine-tuned & evaluated open LLMs on the dataset

Insights
🧠 - Includes query understanding, expansion, reformulation, summarization, ranking, and more.
🌟 - Few-shot generally improve performance, especially in complex tasks.
🏆 - Fine-tuned Llama 7B Chat was better than fine-tuned Llama 7B base.
📈 - INTERS improvement TriviaQA by 77.5%, indicates strong RAG skills.
💡 - Could be used to train a smaller query rewrite model.
🥇 - INTER Llama outperforms Llama Chat on all IR related benchmarks.

3

213

36

202

32K

ar_rahular retweeted

Shashi Tharoor

@ShashiTharoor

over 2 years ago

My speech on “Preserving Historical Memory” at the Quadrennial Congress of the International Conference on Archived, Abu Dhabi, October 13, 2023: https://t.co/xur8Xg47iL

6

115

23

1

35K

Who to follow

happy Home pets

@oluwanishorlah

pet adoption,dogs, cats, birds, lover of Christ, music enthusiast. https://t.co/5Rnuyg6ZMg

Clark Kent

@clark_kent_2023

Free thinker, wannabe writer.

Raathzzz🎧✨

@MeRatheesh

Everyone thinks of changing the world, but no one thinks of changing himself. Leo Tolstoy We are Indian's Firstly and Lastly..! Ambedkar

ar_rahular retweeted

Yann LeCun

@ylecun

almost 3 years ago

This is huge: Llama-v2 is open source, with a license that authorizes commercial use! This is going to change the landscape of the LLM market. Llama-v2 is available on Microsoft Azure and will be available on AWS, Hugging Face and other providers Pretrained and fine-tuned models are available with 7B, 13B and 70B parameters. Llama-2 website: https://t.co/PKrrXgHdem Llama-2 paper: https://t.co/aINNrXNhMb A number of personalities from industry and academia have endorsed our open source approach: https://t.co/N7HwgW9Suh

382

15K

3K

4K

4M

Rahul A R

@ar_rahular

Who to follow

Last Seen Users on Sotwe

Trends for you

Most Popular Users