Excited to share Prem-1B-SQL, a 1.3B parameter model built for Text-to-SQL! It hits 51.54% on the BirdBench private test—beating Claude-2 base and Qwen2.5-coder 7B, on par with GPT-4 baseline.
Fully fine-tuned with the PremSQL library on DeepSeek 1.3B. Check out our Huggingface
I just published my new article on how we can integrate custom
#LLMs
using
@LangChainAI
. I took the example of
@gpt4all
and shared my learnings on different ways to extend it. Check this out
The AI community is predicting that 2024 is going to be the year of the rise of synthetic data. And this is really true. Interested to know more about how those are generated (especially for tabular and text)?
checkout our latest blog:
#GenAI
#AI
#LLM
🍀 Over the last week, we have researched multiple serverless providers like
@modal_labs
,
@beam_cloud
, and
@runpod_io
to accelerate serverless deployment of LLMs like Gemma 2B,
@MistralAI
Mistral 7B v0.1/v0.2. This post reflects our experiences trying each of them:
⭐ Before
Gave a small lightning talk around PremSQL in
@pyconindia
24, thanks for the opportunity and will share the full video soon 🚀🚀
Ps: my high school extempore skills kinda helping me ahhaha
Excited to share Prem-1B-SQL, a 1.3B parameter model built for Text-to-SQL! It hits 51.54% on the BirdBench private test—beating Claude-2 base and Qwen2.5-coder 7B, on par with GPT-4 baseline.
Fully fine-tuned with the PremSQL library on DeepSeek 1.3B. Check out our Huggingface
Happy Monday. I know I am late to this game, but today, I published the very first blog of my written series on MakeMore.
For a while, I studied Andrej Karpathy's series and took some notes. I initially thought I would publish to publish all of it at
Hi there, Checkout my first hand's on session on Deploying ML apps using
@FastAPI
and
@Docker
. The session was hosted by
@CRforAi
's DL Symposium. Would love to get your feedback 😄
YouTube:
GitHub code:
#learning
with
#community
Well all you need is just some cool enthusiastic folks, a research paper to read and cup of chai thats it.. we read the paper GameGen by deepmind
Cheers for more to come 🙌
First ever B’Luru Research Enthusiasts Meetup (BREM) ✨
We sat and read a research paper individually and then came together to discuss it.
It’s been ages since I had such fun 😍😍
I ve always wanted to create a space for like minded nerds 🥰 to connect, BREM is Born today 🔥
#opensource
is the best way to learn about anything. Introduced Automatic Hyper-param search using
@OptunaAutoML
in Prompt2Model. Thanks to
@vijaytarian
and Prof.
@gneubig
,guiding me in the process the whole time. learned a ton. Check out Prompt2model 👉
Today we launch Benchmarks. A fully reproducible oss repo that benchmarks Llama 2 on popular engines like
@PyTorch
@NVIDIAAI
's TensorRT, etc on various precisions. Coming versions will analyze generation quality w.r.t change of engines. So stay tuned.
⭐
Weekend cook up ........
when
#recsys
meets
#Llama2
...
PS,
@huggingface
's TRL library is soo cool ... it just boils down to the dataset format to get started with baseline fine-tuning ...
blog coming soon 😁
Excited to announce our very first release of PremSQL. PremSQL is an open-source library designed to help developers create secure, fully local text-to-SQL solutions using small language models. It offers all the essential tools needed to build and deploy end-to-end text-to-SQL
Text-to-SQL has always been a very active research field. It has several use cases, most related to information extraction. Check out our latest blog post on the current state of text-to-SQL tasks with LLMs.
We have covered the progress on models,
2024 just got started and we can see quite a trend going on for SLMs like Phi, Model Merging, MoEs, Synthetic Data, and Attention-free
#LLMs
like Mamba.
In this blog, we are covering all of the concepts to understand Mamba, so do check this out 👇
#AI
Lately, I was getting so many resources around LLMs / general AI, and it got super overwhelming. So I just cleaned all of them up and piled them.
Now I gotta filter some of them out, so that I could cover some of the top-priority things by the end of
Saving this for later, coz this is special..., literally this blog post about the hiring process of
@genintelligent
just made my day...
Check this out...
Gonna give my first ever lightning ⚡ talk on the current scenario on LLM evaluation. Do join us on Nov 7th with
@LightningAI
. We also have other cool talks on Distributed training, understanding attention etc.. Check out more here 👇
I used to spent a lot of time around GraphML before LLMs... I started those once again for recsys and referred to my notes for a quick brush up, written a yr ago... Thought it would be good to share
#learning
#in
#public
#opensource
#ml
All nighter Saturday hack… hybrid conversational search … ft Text to SQL (and embedding based)
In this picture, the prompt was: show me some sneakers and pants which matches with the top (uploaded image)
Compared
@MistralAI
's mistral-medium
@OpenAI
's gpt-3.5 and
@togethercompute
's Mistral-7B instruct to code the Attention Module from scratch and also to validate. And it seems like gpt-3.5's implementation is fully functional. Although Together is super fast, the quality 🤕
Hey everyone! 👋Just wrote a blog about my journey exploring ML engineering and MLOps. Sharing some of my experiences and learning along the way. Check it out!
#MLengineering
#MLOps
#dataengineering
Ok, finally a personal website where I can write about my past works and current progress and also some detailed blogposts ...
ps: did't know I had a Japanese name embedded in my actual name lol
how to start a startup?
> step 1: choose an emoji
> lemme start: 🧬🔧
Congratulations to me, I started a new company, its name is BuilDNA ... now watch me how I raise 1B $ seed in my dreams ....
Tool calling is now available in
@LangChainAI
and
@premai_io
. You can now easily do function calling across your favourite models like Claude, Llama, GPT.
Try it out now.
🚀 We are open-sourcing our first version of Prem text2sql. Code generation LLMs and LLM-assisted coding have almost reached convergence. However, LLM-assisted data analysis is a much more complex problem. Our previous blog discussed various datasets, models, evaluation methods,
@elonmusk
Master pieces like this gives the depth (or the gravity) of significance of a technology ....
Intertwined with one who makes those technology and one who spreads the significance of that through these musics / movies so that the coming gen can be influenced ....
hats off to
🌄 We launched our very first Small Language Model Prem 1B and Prem 1B Chat. At our main focus is to make smaller models which can understand english construct properly to follow instruction and retrieve information when proper context given.
✨ When we
🚀 We are excited to introduce our first series of open-source Small Language Foundation Models, Prem 1B and Prem 1B chat. These models are available on
@huggingface
under APACHE LICENSE 2.0. Read our release blog here:
🎯 Our goal is to create models
Finally, wrapped our first Gen AI meetup by
@LightningAI
. It was truly an amazing experience sharing my learnings on LLM evaluation along with
@aniketmaurya
@ishandutta0098
and
@ravithejads
and their insightful talks. Can't wait to have more meetup like this again😄. resources 👇
All nighter Saturday hack… hybrid conversational search … ft Text to SQL (and embedding based)
In this picture, the prompt was: show me some sneakers and pants which matches with the top (uploaded image)
Good Sunday everyone, Today we release the very first version of Sanskriti Bench app. Last week, we had great launch from
@guneetsk99
and awesome round of shoutouts from
@Analyticsindiam
, and supporters like
@huggingface
's initiative of Data-is-Better-Together
Well well, all my devops x LLM friends out there, I heard it is a pain to manage GPU resources, installations, dependencies issues while deploying your LLMs on K8s.
With Prem Operator, all of those can be done with just few commands, check this out 👇
You should take a notebook along when you read
@chipro
's blog post. It is full of valuable information. I mean in just two days, I have learned some gigantic stuff (still need to ascimilate a lot) about the essence of resume and ML going real time. More about those in a 🧵(1/7)
@doesdatmaksense
I used to maintain mine... and now I get feel more guilty that I am not... ughhhh
also for iPad somehow I landed to "FreeNotes" which is surprisingly free .. lol
Just going through
@nishparadox
's profile... And i must say... The work he has done at
@docsumoai
is incredible and something I day dream about (like literally doing things single handedly .. Robust ML +Software + design and leaving a legacy)... Real inspiration 😃
It was awesome contributing a chapter for this course. Community NLP Course when?
ps. I contributed the chapter: Finetuning ViT for Object Detection. Do give it a read:
Also shout out to amazing folks who have contributed the chapters.
@AssemblyAI
's YouTube channel is so amazing. Super crisp and informative videos. Do check those out if you guys are into ML. They have numerous series related to Deep learning and
#ML
and ML deployments. Huge shoutout to
@python_engineer
and
@misraturp
for such amazing content.
Less go ⚡️
Exited to talk about Evaluation scenario of LLMs and how to easily get started with lit-gpt. Its goanna be fun meeting all the AI enthusiasts in
#Bangalore
Join us for a series of lightning talks on LLMs at our first-ever Bangalore meetup! We'll deep dive into evaluation and connecting to custom data sources with LlamaIndex on November 7.
Register now ⚡
#MachineLearning
#DeepLearning
#LLMs
Research paper code base -> robust and maintainable code ...
Sharing some personal learnings: suppose you are using a research paper code repository and want to restructure it to extend it well.
Well, well, well, it will never be in the first place because most of the
🚀 Presenting Benchmarks v2. We believe in giving back to the open-source community. Benchmarks v2 acts as a single source of reproducible and transparent truth to understand nuances associated with different inference engines and choose the right one for your use case and
To all the web developers and engineers feeling overwhelmed by the constant stream of new developments in Generative AI, check out Prem AI. It saves a lot of time, and things just works, without much hassle.
Whoou
@perplexity_ai
's pplx-70b-online does give awesome results when questions asked beyond standard knowledge cutoffs
ps. what does this mean? seems interesting 🤠
- "ChatGPT (Perplexity.js Foundation)"
- "ChatGPT (Perplexity.js Foundation)"
is it some sort of leak?
#Mamba
, a potential replacement for transformer-based
#LLMs
. But the paper is a mathematical jugglery and hard to understand. This overview delves into essential tools for understanding the paper's complexity. Dive in for a concise grasp!
#Mathematics
#AI
🐍 You might probably have heard by now about the alternative of transformers-based LLMs: Mamba by
@tri_dao
,
@_albertgu
, and the team.
For a beginner, it would be hard to understand in the first place. But don't worry, we got you:
is now integrated / compatible with
@stanfordnlp
DsPY,
@LangChainAI
@llama_index
and
@qdrant_engine
If your projects are made on top of these libraries but you are falling short of quality choices of LLMs, or confused with observability or implementing
Well well, you have a groundbreaking idea around Generative AI, but struggling to have paid API versions, compute, or even seek mentorship, we got you all covered 😄
checkout
@premai_io
's latest Startup Grant Program
#AI
#LLM
#Startup
#GenAI
👋 Calling all AI enthusiasts, tech wizards, and forward-thinkers! We are announcing our Startup Grants Program: Unlock Your AI Startup's Potential With Prem!
Learn more about the Grant Program here:
🧵 Thread
Well, I guess the best
#2022WrapUp
is here... It's really inspiring not for the aspiring
#MachineLearning
community but for any aspiring techy out there...
Today is my first day at
@NVIDIAAI
! 🥳
-From learning to code at 29
-through learning ML
@fastdotai
-winning a
@kaggle
competition
-jobs at 🔥 startups
-moving continents thx to AI
-to joining the illustrious Merlin team ❤️
I am beyond grateful 🙏
Will make this one count!
Pov: you get some results, you achieved what you expected but now you are happy but not happy because you know you can achieve higher ... lesson learned: aim for unreasonably high ... then you will get the result which you can be happy (for a longer time)
🚀 We are open-sourcing our first version of Prem text2sql. Code generation LLMs and LLM-assisted coding have almost reached convergence. However, LLM-assisted data analysis is a much more complex problem. Our previous blog discussed various datasets, models, evaluation methods,
I just published how we can build a Knowledge Base for our Custom LLMs using
@LangChainAI
,
@trychroma
, and
@nomic_ai
's GPT4All.
This is the 4th blog of my mini-series, building end-to-end LLM-powered apps without Open AI. Do Check out 👉
Domain specific LLM fine tuning always confused me. After several iterations I along with
@MojoAnalytics
published our first blog on best practices and insights on fine tuning LLMs using
@huggingface
and managing experiments using
@weights_biases
👉
#LLM
Guys, please tell me if I am missing something. Have we reached the saturation level when it comes to deep learning? Sure, transformers (or the multi-headed attention mechanism) have been the pioneering point for what we see now, but is that it? Fundamentally, it is a very simple
@darkyboy_
mine one is temporary hacku deployement mostly, like let's say I want to show someone for the next 24 hrs, I use
@ngrokHQ
from my local, colab etc etc .. works great and for long time solution use
@modal_labs
they have super nice and intuitive way to deploy models serverless
We have seen amazing oss models like
#Mistral
and
#Llama2
, etc. However, technically those are not fully open-sourced.
@llm360
takes things to the next level where they fully open-sourced their whole journey. LLM research now can happen with less crazy expenses. Curios? 👇
🌐 LLM360 by
@llm360
unlocks the true significance of open-source LLM research. Transparency, collaboration, innovation, and sharing learning in the community take center stage. Want to learn more?
Check out our blog post here:
@doesdatmaksense
I am literally missing my class 12 books right now for analytical geometry and revising some probability stuffs
I used to do sn dey if you heard about it