smhx

7 ms·

1.
Interaction Models(thinkingmachines.ai)
334 points by smhx 4mo ago|53 comments
2.
by smhx 2y ago
If you're the author, unfortunately I have to say that the blog is not well-written -- misinformed about some of the claims and has a repugnant click-baity title. you're getting the attention and clicks, but probably losing a lot
3.
by smhx 2y ago
the author got a couple of things wrong, that are worth pointing out: 1. PyTorch is going all-in on torch.compile -- Dynamo is the frontend, Inductor is the backend -- with a strong default Inductor codegen powered by OpenAI Triton (which n
4.
by smhx 2y ago
it's not a new PyTorch feature. It's just a showcase of existing PyTorch features (including libtorch) as an end-to-end example. On the server-side it uses libtorch, and on mobile, it uses PyTorch's executorch runtime (that&#
5.
by smhx 2y ago
that's a direct implication that they're waiting for a liquidity event before they speak
6.
by smhx 2y ago
most of the GenAI players use both PyTorch and JAX, depending on the hardware they are running on. Character, Anthro, Midjourney, etc. are dual shops (they use both). xAI only uses JAX afaik.
7.
by smhx 3y ago
You've created a superior llama/mistral-derivative model -- like https://old.reddit.com/r/LocalLLaMA/comments/17vcr9d/llm_com... How can you convince the world to use it (and pay you)? Step 1:
8.
by smhx 3y ago
>What's a fair benchmark? the absolute golden benchmarks are https://github.com/pytorch/benchmark They are a diverse set of userland code taken from github as-is and made into benchmarks.
9.
by smhx 4y ago
thanks for the heads-up, looks like we didn't yank the CPU wheels on those dates. will get to them in the next set of working hours, as its an unlikely scenario (not only do you have to install the wheel of a specific date, you also ha
10.
by smhx 4y ago
> So technically, if you are pulling the older version of pytorch-nightly (specifically 2.0.0.dev20221230), it will still pull that compromised dependency (because torch have explicit version lock to it). All PyTorch nightlies with this
11.
by smhx 4y ago
This is the last one. It was also the first one.
12.
by smhx 4y ago
this only affects the nightly pytorch. the stable pytorch build doesn't depend on `torchtriton`. the nightly pytorch moved to depend on our own secured `pytorch-triton` now, secured on PyPI and our nightly channel.
13.
by smhx 4y ago
the version numbers are different, even though the package names are the same. A stable version will have a version number such as `1.13.0`, where as a nightly version will have the date in the version number, such as `2.0.0.dev20221230`. Y
14.
by smhx 4y ago
ShardedTensor was merged / subsumed into DTensor: https://dev-discuss.pytorch.org/t/rfc-pytorch-distributedten... Lots of development and traffic happening here: https://github.com/pytorch/tau
15.
Welcoming PyTorch to the Linux Foundation(linuxfoundation.org)
47 points by smhx 4y ago|4 comments
16.
by smhx 4y ago
It can run much faster. For example, using the PyTorch nvFuser JIT gives a 50% speedup: https://old.reddit.com/r/MachineLearning/comments/xa75km/p_p...
17.
by smhx 4y ago
counter-example: CPython, NumPy, SciPy and many critical well-used open-source projects that have tiny volunteer teams and operational budgets that are minuscule compared to something like React / PyTorch
18.
Unionizing for Profit: How to Exploit the Power of Unions in C++(dev-discuss.pytorch.org)
2 points by smhx 5y ago|0 comments
19.
by smhx 5y ago
Whether using Lua or Python, for GPU-based scientific computing, the need to drop into a C call is the same. The overhead of Python vs Lua never really mattered. While we were based on top of LuaJIT, we couldn't use the JIT for anythin
20.
by smhx 5y ago
We ship everything needed for userland -- including parts of CUDA/CuBLAS and CuDNN that we need (which is why our binaries are so fat). GPU drivers would be kernel-land and I don't think we actually can install GPU drivers as part
21.
by smhx 5y ago
Author here. Being conscious about speed and performance is different from making that your competitive advantage or USP. Our main focus is usability, and one of our secondary focuses is to not look like clowns in the performance department
22.
Growing open-source: from Torch to PyTorch(soumith.ch)
3 points by smhx 5y ago|0 comments
23.
by smhx 5y ago
another good resource that's more specific to machine learning is https://paperswithcode.com/datasets
24.
by smhx 6y ago
I've been helping on the project, it's lead by Chris Cummins and Hugh Leather. Just a heads-up for folks, we haven't fully cleaned up and gotten ready for public attention yet, we are 90% there. Once we are golden, we're
25.
by smhx 6y ago
@m0zg if you could comment on the issue you quoted with any details, it would be really helpful to us. Unlike what is reported in the github issue, `permute` isn't the regression. For reference, one of the core devs added more details
26.
by smhx 6y ago
PyTorch maintainer here: we're looking into that and if needed will issue 1.5.1 asap. It didn't show up in release testing, which among other things does end-to-end imagenet runs with ResNet50 and a few other models (i.e. time and
27.
by smhx 6y ago
notable co-releases along with PyTorch 1.5: - TorchServe: model serving infrastructure for scalable model deployment - TorchElastic w/Kubernetes: fault-tolerant "elastic" neural network training, allowing nodes to join and le
28.
by smhx 7y ago
Hi I am a PyTorch maintainer. NVIDIA has dropped CUDA support for macOS: http://www.cgchannel.com/2019/11/nvidia-drops-macos-support-... This was pretty evident for a few years, and it's one of the top reason
29.
by smhx 7y ago
There's also fastapi, which is well-respected in Python. Someone wrapped fastapi / rabbitRPC to serve PyTorch models (with auto-batching to increase serving efficiency) in https://github.com/catalyst-team/reac
30.
Preferred Networks Migrates Its Deep Learning Research Platform to PyTorch(preferred.jp)
5 points by smhx 7y ago|0 comments

More ›