Making ML Make Sense @ https://xitter.cf/t.co/mD4mMifuVg 👾
Philadelphia, PA
Joined November 2021
- Tweets62
- Following266
- Followers151
- Likes123
Anish Shah retweeted
Looking to go beyond basic LLM experiments and build real, production-grade applications?
We have two free virtual courses designed for AI engineers of all levels who want to master RAG and LLM evaluations.
Here’s what you’ll learn 🧵
Anish Shah retweeted
🏆🏆🏆🏆 Four papers at COLM were awarded outstanding paper awards. Congratulations to the authors!
Anish Shah retweeted
The new @wandb RAG++ course uses the v2 API and the new command-r models. In addition to learning about nuanced RAG techniques, you also get to work with the new API in the course colabs thanks to the credits from @cohere.
Register here: wandb.me/rag
We’ve released updated versions of our APIs! This upgrade aims to improve the developer experience, making it easier and faster for developers to build with Cohere.
cohere.com/blog/new-api-v2
Anish Shah retweeted
📣 I am happy to announce that @wandb Weave is now integrated with Instructor (@jxnlco)!
🧶 Weave will automatically capture traces for Instructor. To start tracking, call `weave.init()` and use the library as normal.
👉 Learn more at weave-docs.wandb.ai/guides/i…
Anish Shah retweeted
Announcing our new RAG++ course, now available in collaboration with @CohereAI and @Weaviate_io. Created for engineers looking to build production-ready RAG systems.
The course covers everything from evaluation strategies and data preprocessing to advanced retrieval techniques and prompt optimization and includes hands-on exercises with code notebooks and Cohere credits. Register here: wandb.courses/courses/rag-in…
Anish Shah retweeted
Defining your grading criteria for LLM outputs is an organic process that evolves the more time you spend on it
In “Who Validates the Validators” @sh_reya et al highlight this
This weekend in our SF office a good chunk % of hackers will learn this 😃
lu.ma/judge
Anish Shah retweeted
Today I gave @llama_index's workflows a go to tackle the NeurIPS AI hackercup competition.
I created a Workflow to iterate from an initial solution, run the generated solution, and check against the expected output.
It is effortless to define steps with the expected inputs and outputs. Define your Events and their attributes, and the workflow will be triggered in order when you hit run.
I love how the @wandb Weave + step decorators play nice together.
Try on colab: colab.research.google.com/gi…
Thanks to @LoganMarkewich for all the help!
#hackercupai
Anish Shah retweeted
I've been evaluating chunking mechanisms while prepping our RAG course's data ingestion lesson.
Structured, semantic, and syntactic chunking each uniquely impact RAG performance.
Learn to choose the right approach for your use case: wandb.me/rag-course
Anish Shah retweeted
⚡️ AI Hacker Cup Lightning Comp
Today we're kicking off a ⚡️ 7-day competition to solve all 5 of the 2023 practice Hacker Cup challenges with @MistralAI models
Our current baseline is 2/5 with the starter RAG agent (with reflection)
@MistralAI api access provided
Details👇
Anish Shah retweeted
Join us on September 10 to learn how to build production-ready RAG systems. Learn from @ash0ts about optimizing pipelines, enhancing queries, and scaling solutions for real-world applications. Ideal for tech leads and product managers driving AI innovation. Register now: wandb.ai/site/resources/even…
Anish Shah retweeted
Join me tomorrow. I'm presentimg a codegen agent featuring RAG-based episodic memory and reflexion. Whether you're aiming to enhance your LLM toolkit or just curious about the latest techniques, this talk will deliver essential insights.
🔗events.bizzabo.com/NeurIPSha…
Anish Shah retweeted
Stoked to partner with the team to host 7 lectures, a lot of agents creativity needed
📺 wandb.me/hackercup
NeurIPS Meta Hacker Cup AI track gonna be 🔥
Had a super insightful webinar with @GroqInc on debugging AI-generated code! Their inference speed with Llama 3.1 is mind-blowing, making trial and error so much smoother.
youtube.com/watch?v=B70jJYPV…
Anish Shah retweeted
I will be talking tomorrow on-
- LLM landscape
- Need for structured output - function calling, json, constrained decoding, more
- RAG
- LLM system evaluation
- @wandb Weave for building LLM applications correctly
If it excites you, consider showing up. 💫⭐🌟
Will be speaking on 9th, Aug at the @AnalyticsVidhya's Data Hack Summit 24.
More details here: analyticsvidhya.com/datahack…
Anish Shah retweeted
Hallucinations in AI-assisted coding are a significant challenge for developers. Join Daniel Loman from @GroqInc and @ash0ts from Weights & Biases as they share strategies to overcome this issue in our technical webinar on August 6. Register now: streamyard.com/watch/hikcAge…
Anish Shah retweeted
Evaluated Llama 3.1 70B (@FireworksAI_HQ) as well and is performant at the same level as Mistral Large 2 128B.
- Llama is outperforming on GSM8k (math) while Mistral is outperforming on MATH benchmark.😅
- Llama > Mistral (reasoning tasks)
- Mistral > Llama (Q&A tasks)
I am excited to share some evaluation results done on the MixEval hard benchmark. It was done to see the performance of the two newly launched models - @AIatMeta's Llama 3 405b and @MistralAI's Mistral Large 2.
Check out the results below! 👇
Ran Chain of Density over the full 90 page Llama 3 paper. The user instruction was probably asking for too many things given the available output length. Any other questions I should ask to the paper?
Check out the Chain of Density process breakdown at: wandb.ai/wandb_fc/arxiv-read…
