surya
@suryasure05
Followers
2K
Following
2K
Media
49
Statuses
372
inference @groqinc // ece @westernu
Toronto, Ontario
Joined October 2023
I spent my summer building TinyTPU : An open source ML inference and training chip. it can do end to end inference + training ENTIRELY on chip. here's how I did it👇:
71
341
3K
probably the single best resource for understanding transformers from first principles
I spent months illustrating how Transformers actually work. Not just what they do, but why they’re built this way. The history, design choices, and intuition behind every layer. From RNNs → Attention → Multi-Head → FFNs → Positional Encoding. Here's everything I wish I
1
0
5
I spent months illustrating how Transformers actually work. Not just what they do, but why they’re built this way. The history, design choices, and intuition behind every layer. From RNNs → Attention → Multi-Head → FFNs → Positional Encoding. Here's everything I wish I
8
9
23
Just built Curserve with @nathanbarrydev @Alexkranias and @PranavTadepalli . It’s a server-side coding agent framework that’s 30x faster than spawning subprocesses and eliminates network latency entirely. We also placed in the Crater sponsor prize! Heres how we did it.
7
6
12
noise cancelling headphones might be the best investment I’ve ever made after my macbook
1
0
13
There is probably no better way to really understand CUDA than writing RTL. Honestly writing RTL just makes parallelism really easy to think about no matter the language.
Please, for the sake of your career don't do this. FPGAs are no good beyond prototyping and emulation and once you have two generations of hardware deployed only incremental blocks of hardware are ever taken to the FPGA. They don't scale, they are slow, and most importantly
8
15
371
after being in the industry for a bit, I’m finally starting to understand exactly how much LLMs are changing the role of a software engineer. the value in being a great swe now is in being able to understand systems from first principles and ask the right questions, since you
1
2
49
don’t be surprised when the saksham stock skyrockets in a few months
currently building a transformer from scratch in jax to understand the architecture, and how ML compilers work. finished the file that processes the embeddings, and implemented RoPE. also took a look at the JAXPR and StableHLO IR’s and drew a computational graph :)
2
0
16
I built a neural net from scratch in NumPy to solve the XOR problem. It sounds simple, but the process was way more confusing (and interesting) than I expected. Here’s how it went: 👇 https://t.co/18LgMMWW1k
github.com
Contribute to michaeltrbo/mlp-xor-problem development by creating an account on GitHub.
3
5
19
It’s official: McLaren F1 x Groq Bringing inference speed at a winning cost to the grid and beyond. See you in Singapore. 🧡🏁
23
37
328
one of the coolest projects I’ve ever seen
I spent the past month reimplementing DeepMind’s Genie 3 world model from scratch Ended up making TinyWorlds, a 3M parameter world model capable of generating playable game environments demo below + everything I learned in thread (full repo at the end)👇🏼
0
0
14
incredibly grateful for this opportunity
Stoked to welcome the TinyTPU team to Groq @evanliin, @XanderChin, @suryasure05, @kennykgguo How it started How it's going
6
1
73
Stoked to welcome the TinyTPU team to Groq @evanliin, @XanderChin, @suryasure05, @kennykgguo How it started How it's going
26
20
346
Introducing our co-hosts for this season 🫶 (@krupaad @_richapandya @kennykgguo @achettimada @ambiguousNull)
2
2
12
western university has talented builders flying under the radar. now they have a spotlight. introducing the mustangs network 🐎
13
4
72
I wrote a short article about LFM-2's (by @LiquidAI_ ) hybrid architecture w/ illustration + simple pytorch impl.
13
18
201
the air canada app is next level — i got a notification exactly when my suitcase entered the baggage carousel not sure if all airlines have this feature, but it’s such a quality if life improvement
2
0
11
you can just do things for fun and get life changing opportunities
0
3
36
wrote a piece on how we do tiling and gradient descent on TinyTPU included some animations of the weight and bias gradient descent
9
63
654