Hugging Face's logo

LLHF
community
Activity Feed

AI & ML interests

None defined yet.
danieldk 
posted an update 28 days ago
view post
Post
175
We have recently added Torch Stable ABI support to kernels and kernel-builder. This allows kernel developers to target a particular Torch version and the kernel will be supported on that Torch version and later Torch versions (up to ~2 years).

This makes it much easier to write kernels with long-term support and not just the last two Torch releases.

We have also started rolling out Stable ABI support to kernels in kernels-community, starting with Flash Attention 3, supporting Torch 2.9 and later as well as CUDA versions starting at 12.6:

https://huggingface.co/kernels/kernels-community/flash-attn3/tree/v1/build
  • 1 reply
·
victor 
posted an update 2 months ago
view post
Post
2950
Sharing how I built the LongCat-Video-Avatar 1.5 Space (+500k views on X) in one agent session. Gave a coding agent its own AI lab on ZeroGPU, framed the goal, walked away. It designed, deployed, tested against the live API, fixed, shipped.

Full recipe with the copy-paste prompt: https://huggingface.co/blog/victor/building-zerogpu-spaces-autonomously
  • 1 reply
·
danieldk 
posted an update 2 months ago
view post
Post
269
Two large changes in kernel-builder this week:

kernel-builder now links libstdc++ dynamically. To support a wide range of systems, we build against libstdc++ from manylinux_2_28 (EL 8 and later).

Following our Torch support policy that the current and previous Torch versions are supported, Torch 2.10 support was removed. We will soon also support the Torch stable ABI, so that it is possible to write kernels that support a large number of Torch versions.
victor 
posted an update 4 months ago
view post
Post
6226
Want to share my enthusiasm for zai-org/GLM-5.1 here too 🔥

I think we have it: our open source Claude Code = GLM-5.1 + Pi (https://pi.dev/) - Built a Three.js racing game to eval and it's extremely impressive. Thoughts:

- One-shot car physics with real drift mechanics (this is hard)

- My fav part: Awesome at self iterating (with no vision!) created 20+ Bun.WebView debugging tools to drive the car programmatically and read game state. Proved a winding bug with vector math without ever seeing the screen

- 531-line racing AI in a single write: 4 personalities, curvature map, racing lines, tactical drifting. Built telemetry tools to compare player vs AI speed curves and data-tuned parameters

- All assets from scratch: 3D models, procedural textures, sky shader, engine sounds, spatial AI audio!

- Can do hard math: proved road normals pointed DOWN via vector cross products, computed track curvature normalized by arc length to tune AI cornering speed

You are going to hear about this model a lot in the next months - open source let's go - and thanks z-ai🚀🚀
  • 5 replies
·
victor 
submitted a paper to Daily Papers 5 months ago
mryab 
submitted a paper to Daily Papers 5 months ago
victor 
posted an update 6 months ago
view post
Post
2851
Interesting article: use Claude Code to help open models write CUDA kernels (for eg) by turning CC traces into Skills. They made a library out of it 👀

https://huggingface.co/blog/upskill
danieldk 
posted an update 6 months ago
view post
Post
2843
kernels 0.12 is out! 🎉

Changes:

* Support for kernel version branches to gracefully roll out kernel API changes.
* Support for PyTorch 2.10.
* kernel-builder is now merged into the kernels repo.
* Initial support for standardized kernel benchmarks.

https://github.com/huggingface/kernels/releases/tag/v0.12.0
eliebak 
submitted a paper to Daily Papers 7 months ago
victor 
posted an update 7 months ago
danieldk 
posted an update 9 months ago
hfkwr 
authored a paper 10 months ago
lysandre 
posted an update 11 months ago
view post
Post
9404
We're kick-starting the process of Transformers v5, with @ArthurZ and @cyrilvallez !

v5 should be significant: we're using it as a milestone for performance optimizations, saner defaults, and a much cleaner code base worthy of 2025.

Fun fact: v4.0.0-rc-1 came out on Nov 19, 2020, nearly five years ago!
  • 6 replies
·
eliebak 
posted an update 11 months ago
view post
Post
4612
Super excited to announce that our research team at Hugging Face will be doing an AMA on reddit r/LocalLLaMA.

Come ask any questions to the team behind SmolLM, FineWeb and more! And who knows, maybe there’ll be a shiny new release to talk about?

Thursday 4th September, 8AM-11AM PST 🤗

science
eliebak 
posted an update 11 months ago
view post
Post
789
Motif 2.6B tech report is pretty insane, first time i see a model with differential attention and polynorm trained at scale!

> It's trained on 2.5T of token, with a "data mixture schedule" to continuously adjust the mixture over training.
> They use WSD with a "Simple moving average" averaging the last 6 ckpt every 8B token.
> They trained on Finemath, Fineweb2, DCLM, TxT360.
> Lot of details in the finetuning data they used, for instance they used EvolKit and did some "dataset fusion" to have more compressed knowledge into the data.
> They mention they also tried Normalized GPT, QK-Norm and Cross Layer Attention.

Motif-Technologies/Motif-2.6B
Wauplin 
posted an update about 1 year ago
view post
Post
3760
Say hello to hf: a faster, friendlier Hugging Face CLI ✨

We are glad to announce a long-awaited quality-of-life improvement: the Hugging Face CLI has been officially renamed from huggingface-cli to hf!

So... why this change?

Typing huggingface-cli constantly gets old fast. More importantly, the CLI’s command structure became messy as new features were added over time (upload, download, cache management, repo management, etc.). Renaming the CLI is a chance to reorganize commands into a clearer, more consistent format.

We decided not to reinvent the wheel and instead follow a well-known CLI pattern: hf <resource> <action>. Isn't hf auth login easier to type and remember?

The full rationale, implementation details, and migration notes are in the blog post: https://huggingface.co/blog/hf-cli

  • 7 replies
·
eliebak 
posted an update about 1 year ago
view post
Post
4860
Kimi K2 tech report is full of gems as always. Here are my notes on it:

> MuonClip: Pretty crazy how after 70k the training stabilizes and the QK-clip is basically inactive. There is also no loss in perf with QK-clip which is not trivial at all (at small scale but with aggressive threshold). Also a cool explanation of why muon makes the logit explode in appendix E (tl;dr is that muon makes the singular value of the update matrix higher)
> Sparsity scaling laws to justify their ratio, they have a very solid training infra that allows the model to be trained at this sparsity level, they could have increased even more but as sparsity increases the training becomes less efficient.
> They diminish the number of attention heads to make it more efficient for long context since attention heads are a big bottleneck for long context. They also remove 2 of the 3 "first dense" layers in the dsv3 arch.

With the sparsity and attention heads (divided by 2) they achieve 83% increased flops compared to deepseek v3 arch at 128k.

> Data: Rephrasing is KEY. They do a lot more synthetic data generation and rephrase their corpus to have different styles, for longer documents they do it by chunk. I'm (half) surprised by the fact that ONLY 1 epoch (assuming same number of training tokens I think?) of data rephrased 10 times has better accuracy than 10 epochs of the same data rephrased once.
> They do rewriting for Math and Knowledge, for Math they apply the ShallowMath recipe and instruct the model to rephrase in a "learning note" style
> They talk about diversity and probably have some internal stuff/eval to test that, as always still a bit unclear for me how to properly measure that.

The infra is also very nice, quick summary:
> PP=16 (1F1B schedule, a bit custom), EP=16, zero1
> No FP8 computation but for storage of specific layers, selective recomputation for inexpensive block, activation offloading to CPU
danieldk 
posted an update about 1 year ago
view post
Post
2074
kernels 0.8.0 is out: https://github.com/huggingface/kernels/releases/tag/v0.8.0

This release refines kernel selection in the kernelize function:

• You can now register kernels for certain CUDA capability ranges.
• Rather than doing exact mating of modes, fall back to other compatible modes. If you are kernelizing for inference, but you only registered a training + torch.compile kernel, it will use that kernel since it is compatible with inference as well.
  • 1 reply
·
danieldk 
posted an update about 1 year ago
danieldk 
posted an update about 1 year ago
view post
Post
390
Kernels 0.7.0 is out: https://github.com/huggingface/kernels/releases/tag/v0.7.0 🚀

This release makes it possible to register multiple kernels for a layer. Do you have a super-fast kernel for inference and another kernel for training? Register them both and kernelize will pick the kernel depending on whether you are going to do training or inference.