simrat hanspal retweeted
1. Diffusion Language Models
Language models do not always have to generate text one token at a time. Diffusion language models start with a masked sentence, predict many words in parallel, and refine them over multiple rounds. Models like LLaDA and Mercury are exploring a new path for text generation.
courses.vizuara.ai
simrat hanspal retweeted
First, @simsimsandy walked us through GPU architecture, optimizations, CUDA, and the challenges of running large ML models on GPUs, with a special look at the attention mechanism, KV-Cache optimizations, and PagedAttention!
simrat hanspal retweeted
🚀Join us in Chennai next week for our hands-on workshop: "Building AI Agents with RAG and Functions" 🤖✨
Limited seats available, so hurry and secure your spot! 🏃♂️💨
🔗 Register now: lu.ma/6lrnyo1b
#AIWorkshop #ChennaiEvents #llm #genai
simrat hanspal retweeted
Surprised how Twitter Influencers have got more insider informations and conclusions than anyone else!
Also, because Apple uses OpenAI means it couldn't make it on its own 🤦🏽
simrat hanspal retweeted
If you’re a developer and you’re not using @cursor_ai, I have one simple question: why?
simrat hanspal retweeted
Join us for the Meta Llama 3 Hackathon in San Francisco from May 11-12!
Excited to share that we're hosting the first ever Llama 3 Hackathon with @cerebral_valley + @SHACK15sf in San Francisco. Prizes include a total of $10K+ in cash + partner credits to kickstart projects.
Register to join us May 11-12: partiful.com/e/p5bNF0WkDd1n7…
simrat hanspal retweeted
NEW VIDEO - Rabbit R1: Barely Reviewable
youtu.be/ddTV12hErTc
This is the pinnacle of a trend that's been annoying for years: Delivering barely finished products to win a "race" and then continuing to build them after charging full price. Games, phones, cars, now AI in a box
simrat hanspal retweeted
meanwhile I’m still on free @GoogleColab
First @NVIDIA DGX H200 in the world, hand-delivered to OpenAI and dedicated by Jensen "to advance AI, computing, and humanity":
simrat hanspal retweeted
Books, papers and meet-ups - it's all happening @fifthel
If you want to discuss AI, ML or DL research papers, or want to share an interesting AI sci-fi book, head to The Fifth Elephant community at has.gy/GtfX and has.gy/cPIn
@jackerhack @gautambhatia88
simrat hanspal retweeted
Now that dad jokes are underway, let's get geeky abt NLP and LLMs. On 9 Feb, @fifthel is hosting @simsimsandy for discussing the paper LLMs are long shot learners - has.gy/TO81
Interested? Intimidated? FOMO creeps? Come off to @HasuraHQ at 6 pm and hang out with peers
simrat hanspal retweeted
@FlutterDevelopers Gear up to join @invideoOfficial to build an innovative mobile app using Flutter, paired with a high-performance video rendering engine in Rust. Dive into multifaceted development, linking Dart, Swift/Kotlin, and Rust to work together smoothly. You'll enhance user experiences, work with tools like GraphQL, go_router, and craft custom widgets. This role is a unique chance to be at the forefront, combining cutting-edge programming languages and frameworks in one state-of-the-art application. Your code will make a real difference here!
About us :
@invideoOfficial, we pride ourselves on creating the world's easiest AI-based video creation software. Our software enables users to transform simple text prompts into ready-to-publish videos with engaging scripts, relevant footage, and human-like voiceovers.
invideo ai quick demo:
youtube.com/watch?v=x4VJ7DSk…
Discussing creativity with @shreyas part 3/3:
xitter.gq/sanket_invideo/status/…
Here's our MSN (Must Have - Should Have - Nice to Have) list to guide you:
docs.google.com/document/d/1…
@safierinx @shivamsupr @jawadsh123_ @_rajrajhans @TheRohanSanap @_anshulk @adilmas13
#flutterdev
I had in one of my post mentioned that a lot of early stage companies follow late stage rules (in principle, I had said that) also mostly because operators in the US speak a lot about what they do (when they are 100M) and we want to follow. Rules of every stages are different. The problem is we want to only know the rules of largest companies/success stories. Unfortunately, late stage rules are sexy and recent and thus they get told the most. Creativity 3/3
simrat hanspal retweeted
On 6 December, Mahathi B. will discuss the paper which explains all you need to know about transformer architecture and #NLP with @simsimsandy and @zaph0id
RSVP to join has.gy/yVVS and enrich your understanding of sequence-to-sequence tasks. @fifthel @hasgeek
simrat hanspal retweeted
Introducing @simsimsandy, a trailblazing AI Engineer at @HasuraHQ, poised to captivate audiences at MLDS 2024. With an illustrious career spanning over a decade in the AI realm, Simrat specializes in Natural Language Processing and currently spearheads the definition of AI product strategy at Hasura. Her expertise has been honed through impactful roles at renowned organizations such as Amazon, VMware, FI Money, and Nirvana Insurance.
For further details and to stay informed visit : lnkd.in/g3tK5UZs
#MLDS #MLDS2024 #MachineLearning #Developers #AIInnovation
simrat hanspal retweeted
While everyone's still buzzing from #OpenAIDevDay, OpenAI has quietly planted a revolutionary feature in the garden of AI: 🌱 'Seeding'.
This allows for (mostly) deterministic outputs from their models, promising a new era of consistency in AI results
And they didn't even make a noise about it! 🤯 #OpenAI
simrat hanspal retweeted
Easy Prompt Engineering for RAG 📝 ⚙️
Problem: A lot of users want to customize prompts for complex LLM apps (RAG/agents), but most frameworks (like ours) obfuscate this w/ layers of abstractions.
Solution: Let’s get back to the basics - we’re launching a big new @llama_index feature to let you view/set all prompts for any module, no matter how complex, in 2 lines of code ⚡️
This includes the following
✅ All query engines
✅ Response synthesis
✅ Evaluators
✅ Retrievers/rerankers (which use LLMs)
Importantly, this works no matter how nested your module is; we’ll recursively fetch prompts and collapse it into a flat list. Easily view and set prompts with the given key.
This allows you to get/set prompts on any RAG pipeline no matter how complex it is 💪
Take a look at some of the screenshots below 🖼️. All you have to do is `get_prompts` and `set_prompts`!
Full guide/cookbook here 🧑🍳: github.com/run-llama/llama_i…
simrat hanspal retweeted
State of AI Engineering - @barrnanas
~850 AI engineers, surveyed:
- OpenAI models most popular
- 50/50 split between self-hosted and 3rd party vector DBs
- No best practices yet for prompt management
- SWEs are getting into AI; many SWEs with 10y+ experience have <1y exp with AI
simrat hanspal retweeted
I gave a talk at Seoul National University.
I titled the talk “Large Language Models (in 2023)”. This was an ambitious attempt to summarize our exploding field.
Video: youtu.be/dbo3kNKPaUA
Slides: docs.google.com/presentation…
Trying to summarize the field forced me to think about what really matters in the field. While scaling undeniably stands out, its far-reaching implications are more nuanced. I share my thoughts on scaling from three angles:
1) Change in perspective is necessary because some abilities only emerge at a certain scale. Even if some abilities don’t work with the current generation LLMs, we should not claim that it doesn’t work. Rather, we should think it doesn’t work yet. Once larger models are available many conclusions change.
This also means that some conclusions from the past are invalidated and we need to constantly unlearn intuitions built on top of such ideas.
2) From first-principles, scaling up the Transformer amounts to efficiently doing matrix multiplications with many, many machines. I see many researchers in the field of LLM who are not familiar with how scaling is actually done. This section is targeted for technical audiences who want to understand what it means to train large models.
3) I talk about what we should think about for further scaling (think 10000x GPT-4 scale). To me scaling isn’t just doing the same thing with more machines. It entails finding the inductive bias that is the bottleneck in further scaling.
I believe that the maximum likelihood objective function is the bottleneck in achieving the scale of 10000x GPT-4 level. Learning the objective function with an expressive neural net is the next paradigm that is a lot more scalable. With the compute cost going down exponentially, scalable methods eventually win. Don’t compete with that.
In all of these sections, I strive to describe everything from first-principles. In an extremely fast moving field like LLM, no one can keep up. I believe that understanding the core ideas by deriving from first-principles is the only scalable approach.
simrat hanspal retweeted
We found a public AI repo on GitHub, exposing over 38TB of private files – including personal computer backups of @Microsoft employees 👨💻
How did it happen? 👀
A single misconfigured token in @Azure Storage is all it takes 🧵⬇️
simrat hanspal retweeted
You can now finetune any sbert embeddings model in @llama_index - no labels needed 💫
Given any corpus, use the LLM to generate a synthetic query dataset to train your embedding model.
Best of all, it’s an extra 5 lines to plugin/improve your RAG app 👇
gpt-index.readthedocs.io/en/…




































