vforno

6 ms·

31.
by vforno 3mo ago
if you like, colibrì always needs to improve so if you have ideas or anything else you are welcome for pull request issues and also benchmarks!
32.
by vforno 3mo ago
It’s a good question. In theory MPI could distribute experts across nodes. In practice, for small clusters the added network latency usually hurts more than it helps. Better suited for big clusters with fast interconnects. For now we'
33.
by vforno 3mo ago
Really thanks!
34.
by vforno 3mo ago
No because I have only 32gb of ram too low
35.
by vforno 3mo ago
Maybe we can see some integration!
36.
by vforno 3mo ago
Really thanks!!
37.
by vforno 3mo ago
In the readme you can see benchmark which everyone with different hardware is running Colibrì, and I have to say I've seen great times! I'm always doing more to improve!
38.
by vforno 3mo ago
I'd love to! Right now I only have a very consumer-grade computer that I've had fun with! We'll see!
39.
by vforno 3mo ago
What causes problems is the rewriting in this case are only read while writing is the cache! However, I'm working to improve more and more and make some parts lighter!
40.
by vforno 3mo ago
Antirez is the number one!thanks really thanks!
41.
by vforno 3mo ago
Thank you so much, it's true! It all started with this spirit!
42.
Show HN: Getting GLM 5.2 running on my slow computer(github.com)
937 points by vforno 3mo ago|240 comments
43.
Show HN: GONK – Lightweight Edge-Native API Gateway in Go(github.com)
3 points by vforno 4mo ago|0 comments
44.
by vforno 4mo ago
I'm trying it, really interesting!
45.
by vforno 4mo ago
Thanks to you! I will update the model later to make it more and more optimized but you will see everything you need in the readme.
46.
by vforno 4mo ago
Absolutely yes! With nanoeuler I learned so much by testing every little detail of the project. Every little part you see has been tested and proven several times so that it could be understood and worked.
47.
by vforno 4mo ago
Yes, because it has many separate kernels instead of aggressive merges like PyTorch (with Torch Compile). Each pass (norm, matmul, residual, RoPE, etc.) launches its own kernel, which increases launch overhead and memory traffic. CuBLAS hel
48.
by vforno 4mo ago
Hi, in nanoeuler I use cuBLAS (NVIDIA's super-optimized library) for all matrix multiplications, with the tensor cores in TF32 mode. It's the same thing PyTorch uses underneath, so it's very fast. What I've optimized (an
49.
by vforno 4mo ago
Really thanks If you need any help or have any questions I'm here.
50.
by vforno 4mo ago
Hi, thanks for the comment. Nanoeuler is starting as a study and research project that will obviously improve over time. I'll do my best to make the readme and other things more readable. Thank you very much.
51.
by vforno 4mo ago
Most part of trasformer and sft!
52.
by vforno 4mo ago
Hi, the uploads are one after the other because it was a long, step-by-step research project where I tested the code on another machine. I admit that I'm slowly making up for the commits on all the projects. For Gutenberg and Shakespea
53.
by vforno 4mo ago
Hi, a couple of hours, not too much! Including sft!
54.
by vforno 4mo ago
yes yes tested on a 4070 ti 16gb everything worked without problems!
55.
Show HN: NanoEuler – GPT-2 scale model in pure C/CUDA from scratch(github.com)
55 points by vforno 4mo ago|26 comments
56.
Show HN: Loomabase – Column-level CRDT sync for SQLite + Postgres(github.com)
2 points by vforno 4mo ago|0 comments
57.
by vforno 4mo ago
E-mail sent!
58.
by vforno 4mo ago
Of course I'll write to you right away! Thank you so much!
59.
Show HN: NanoEuler – GPT-2 scale model in pure C/CUDA from scratch(github.com)
6 points by vforno 4mo ago|3 comments
60.
Show HN: Loomabase – Offline-first column-level CRDT sync for SQLite + Postgres(github.com)
2 points by vforno 4mo ago|0 comments

More ›