Soami Kapadia Profile
Soami Kapadia

@KapadiaSoami

687
Followers
160
Following
19
Media
79
Statuses

Following curiosity | SWE intern @GroqInc

Joined April 2023
Don't wanna be here? Send us removal request.
@KapadiaSoami
Soami Kapadia
3 months
Mixture of Agents on Groq Introducing a fully configurable, Mixture-of-Agents framework powered by @GroqInc using @LangChainAI You can configure your own MoA version using the @streamlit UI through the framework. details + links below๐Ÿ‘‡๐Ÿงต
9
131
786
@KapadiaSoami
Soami Kapadia
3 months
thanks for signing my cube @karpathy
Tweet media one
Tweet media two
14
7
362
@KapadiaSoami
Soami Kapadia
2 months
Llama 3.1 is all you need Super simple terminal assistant using @AIatMeta Llama 3.1 70B Powered by @GroqInc . Gave it a shell, write file and web search tool. It looked at my files, added a README and made modifications instantly. No agentic workflow, nothing special,
9
26
138
@KapadiaSoami
Soami Kapadia
2 months
Thanks for everything @GroqInc โค๏ธ
Tweet media one
Tweet media two
Tweet media three
Tweet media four
9
1
71
@KapadiaSoami
Soami Kapadia
5 months
Stole @JonathanRoss321 โ€™s room at @GroqInc HQ on my first day in office๐Ÿคฉ
Tweet media one
2
1
35
@KapadiaSoami
Soami Kapadia
3 months
we reached a 100 stars in 2 days wthhh thanks for the support everyone!
Tweet media one
@KapadiaSoami
Soami Kapadia
3 months
Mixture of Agents on Groq Introducing a fully configurable, Mixture-of-Agents framework powered by @GroqInc using @LangChainAI You can configure your own MoA version using the @streamlit UI through the framework. details + links below๐Ÿ‘‡๐Ÿงต
9
131
786
2
2
29
@KapadiaSoami
Soami Kapadia
5 months
Open Source + Lightning Fastโšก >>>
@GroqInc
Groq Inc
5 months
How fun is this? Our new intern made a demo showcasing Gemma 7B vs ChatGPT 4 Turbo controlling two Ken characters in a retro game. You can see how many more moves Gemma produces in the text screen (red text). What a creative fun side-by-side. Groq vs Groq next?
19
44
269
0
1
19
@KapadiaSoami
Soami Kapadia
3 months
v late buildspace update cause i procrastinated. @_buildspace @_nightsweekends
Tweet media one
3
0
20
@KapadiaSoami
Soami Kapadia
3 months
i spend way too much time doing this
Tweet media one
5
0
18
@KapadiaSoami
Soami Kapadia
3 months
Why @GroqInc ? it runs really really fast. A 3 x 3 layer MoA agent as in the video (10 LLM calls, 4 sequential) runs in ~3 seconds.
2
0
16
@KapadiaSoami
Soami Kapadia
3 months
What is MoA? Mixture of Agents combines multiple LLMs to work together in cycles to answer a prompt. In each layer LLMs run in parallel and their outputs are then combined and fed into the next layer. The final layer's responses are given to the main LLM for the end response.
2
0
16
@KapadiaSoami
Soami Kapadia
4 months
50+ registered for @GroqInc dev meetup and AMA w/ @JonathanRoss321 & @sundeep ๐Ÿ™Œ Super stoked to see some cool demos from fellow builders (been cooking something myself). (i can finally ask them to increase my rate limits)
2
2
13
@KapadiaSoami
Soami Kapadia
3 months
Why MoA/agentic workflows? Accuracyโฌ†๏ธ: Allows LLMs to "think", refine, breakdown complex tasks into smaller sub-tasks. Costsโฌ‡๏ธ: Using a combination of smaller, open-source models can significantly reduce costs even if you are calling an LLM multiple times.
1
0
12
@KapadiaSoami
Soami Kapadia
2 months
Introducing: Potato 10P๐Ÿ• @lordspline and I fine-tuned Potato the Dog using a custom RL pipeline in less than ~20 mins and with only 30 examples. should we open source the deets on @huggingface ? Training Statsโฌ‡๏ธ
1
1
11
@KapadiaSoami
Soami Kapadia
2 months
whoever named this case kebab-case is a genius
Tweet media one
0
0
10
@KapadiaSoami
Soami Kapadia
3 months
@danielbigham actually, using open source models (that cost ~10x less than proprietary models) in an agentic workflow can reduce costs while improving/retaining accuracy. That means you can call a model like @GroqInc llama3-70b ~10 times for it to be the same cost as 1 gpt-4o call.
0
1
8
@KapadiaSoami
Soami Kapadia
6 months
Imagine a place where everything you upload becomes part of a vast, interconnected web, linking all to all. @achyut_benz and I spent yesterday building just that.
4
1
8
@KapadiaSoami
Soami Kapadia
1 month
im a teapot
Tweet media one
2
0
4
@KapadiaSoami
Soami Kapadia
8 months
Good progress. Good team. Good vibes. Good hack. @_buildspace sweatshirt What more can u ask for
Tweet media one
0
2
4
@KapadiaSoami
Soami Kapadia
2 months
@MatthewBerman Thanks for the mention!!
0
0
4
@KapadiaSoami
Soami Kapadia
2 months
agi is here
Tweet media one
0
0
3
@KapadiaSoami
Soami Kapadia
1 year
Tweet media one
0
1
3
@KapadiaSoami
Soami Kapadia
1 month
prof has no chill
Tweet media one
3
0
3
@KapadiaSoami
Soami Kapadia
1 month
@AarushSah_ accurate
1
0
2
@KapadiaSoami
Soami Kapadia
2 months
He even generalized to use both paws even though we trained him on only the left paw!
1
0
2
@KapadiaSoami
Soami Kapadia
3 months
@AarushSah_ Appreciate it ๐Ÿซก
0
0
2
@KapadiaSoami
Soami Kapadia
3 months
@lifebypixels @GroqInc @LangChainAI @streamlit Thank you for all your support! Couldnโ€™t have asked for a better team to work with :)
0
0
2
@KapadiaSoami
Soami Kapadia
6 months
Introducing Athena. To contribute: 1. add text or a youtube video link. thats it. the community graph will be updated with whatever you added. Start contributing here:
0
0
2
@KapadiaSoami
Soami Kapadia
5 months
wait that was @sama we were ft with?
@advaitpaliwal
Advait Paliwal
5 months
thereโ€™s a line to see @sama ?
Tweet media one
2
1
139
0
0
1
@KapadiaSoami
Soami Kapadia
6 months
@kiwicopple @achyut_benz @nextjs @vercel @supabase @LangChainAI @shadcn @threejs @bunjavascript We are adding that! Right now we just extract the relations, do some very simple post processing and add it to the graph. Lotta cool stuff planned for this stay tuned ๐Ÿ‘€
1
0
1
@KapadiaSoami
Soami Kapadia
3 months
@BenjaminKlieger Thanks for the shoutout Ben!
0
0
1
@KapadiaSoami
Soami Kapadia
2 months
Stats Base Model: Potato Time Taken: 15 mins Examples: 30 multi-modal examples Final Loss: 0 (thats right) Epochs: 30 Batch Size: 1 Params Fine Tuned: 2 (according to our calculations)
1
0
1
@KapadiaSoami
Soami Kapadia
7 months
@grok + @GroqInc combo would go hard๐Ÿ”ฅ๐Ÿ”ฅ๐Ÿ”ฅ
@ClementDelangue
clem ๐Ÿค—
7 months
๐Ÿ‘€๐Ÿ‘€๐Ÿ‘€
Tweet media one
13
35
342
0
0
1
@KapadiaSoami
Soami Kapadia
8 months
Learned the fundamentals of Speedcubing from him when I was 9, I now can solve a Rubik's cube in under 9 seconds. Learned the fundamentals of Deep Learning from him, I can now build and train an LLM from scratch. Thank you for guiding me throughout my journey. @karpathy ๐Ÿ
@karpathy
Andrej Karpathy
8 months
@NairAanish oh that ship has sailed, sorry :D actually one of my favorite meets at OpenAI was a cubing session with two very fast cubers, one of them a former world's record holder. I can't cube anywhere near my prior level anymore so it was a bit embarassing alongside but really fun.
24
14
831
0
0
1
@KapadiaSoami
Soami Kapadia
3 months
@1davidyu1 @_nightsweekends @_buildspace the hardest part for me is to start self reflecting. cant wait to see what ur cookin up ๐Ÿ”ฅ
0
0
1
@KapadiaSoami
Soami Kapadia
3 months
@AarushSah_ @FarzaTV @_buildspace @_nightsweekends very much needed. looking forward to seeing you build this!
0
0
1
@KapadiaSoami
Soami Kapadia
3 months
@yacineMTB @karpathy Heโ€™s the best
0
0
1
@KapadiaSoami
Soami Kapadia
6 months
thought i was a bot lol. does that mean im that cracked?
Tweet media one
0
0
1