TGViewer
Channel Public Channel
Linux - Reddit

Linux - Reddit

@r_linux

Stay up-to-date with everything Linux!
Content directly fetched from the subreddit just for you.

Powered by : @r_channels
Open in Telegram
Subscribers
1.06K
Photos
4.6K
Videos
207
Links
41.9K

Showing posts older than #42244 · Back to latest

Older Posts 20 shown
Post #42238 56
Compromised Fedora account linked to suspicious AI-generated contributions

>In May, a Fedora developer discovered that an allegedly rogue agent had been pestering the project in a number of ways: reassigning bugs, fabricating unhelpful replies to bugs, and even persuading maintainers to merge questionable code into the Anaconda installer. It also submitted a number of pull requests (PRs), some accepted, to several upstream projects. The Fedora account associated with the agent has had its group privileges revoked and the messes have been mopped up, but the motive behind the agent's actions is still a mystery.

>Later on May 27, Williamson said that Giovannini had replied to him privately to say that his credentials had been compromised and that he was not the one behind the AI system. "Obviously we should therefore treat any actions it has taken with suspicion", Williamson said.

>Martin Kolman, a member of the Anaconda team, said the events were "really problematic" even if not malicious. The team had spent a lot of time reviewing PRs from what seemed to be an eager contributor: "while it started to look off after a while, all the replies were still like this - a bit weird, but still *plausible*. He also theorized that it could be an attacker working their way up to malicious activity, much like the XZ backdoor

>

https://lwn.net/SubscriberLink/1077035/c7e7c14fbd60fae9/

https://redd.it/1u3194t
@r_linux
Post #42237 54
GNOME Foundation Announces its First Fellows
https://discourse.gnome.org/t/gnome-foundation-announces-its-first-fellows/35446?u=bragefuglseth

https://redd.it/1u2wueq
@r_linux
GNOME Discourse GNOME Foundation Announces its First Fellows The GNOME Foundation has selected the first recipients who will receive funding through its new Fellowship program, and is delighted to announce that Peter Eisenmann and Sophie Herold will begin work as our first Fellows in July. Sophie and Peter are both…
Post #42234 45
AMD says XDNA1 Linux LLM support isn't available. I used AI-assisted development to get a full transformer layer running on a Ryzen 7 8845HS NPU.

I've been experimenting with the NPU in my Ryzen 7 8845HS laptop (Hawk Point / XDNA1) on Fedora Linux.

A lot of the current AMD Ryzen AI documentation focuses on XDNA2 hardware, and community projects like FastFlowLM don't currently support XDNA1. I was curious whether the hardware was actually incapable of running LLM workloads on Linux, or if the software support had simply moved on.

Full disclosure: I used ChatGPT and Codex heavily throughout this project. I am not an AI compiler engineer and I couldn't have done this without AI-assisted code investigation, debugging and porting work.

The rough progression was:

* Verified the NPU was working through `amdxdna`
* Investigated older RyzenAI-SW releases
* Found public Phoenix/XDNA1 artifacts (`1x4.xclbin`, qlinear\_2, transaction binaries)
* Built the modern Linux XRT/XDNA userspace stack
* Got AMD's old Phoenix GEMM transactions executing on Linux
* Validated 24 transformer-relevant GEMM shapes
* Validated real quantized int4/BF16 inference paths
* Built a reusable Linux `XDNA1QLinear` wrapper
* Executed a complete synthetic Llama-2-style transformer layer

Current status:

* Quantized int4 weights
* BF16 activations
* Q/K/V/O projections on the NPU
* MLP projections on the NPU
* RMSNorm, RoPE, attention and activation functions on CPU
* Deterministic repeatable results
* No Windows involved

Some interesting numbers:

* \~6 ms per transformer layer (warm)
* \~116 MiB resident memory per layer
* 8-layer test stack completed successfully
* Memory scaling appears linear
* No kernel/XRT leaks observed so far

What I have not done:

* No real model weights yet
* No llama.cpp integration
* No token generation
* No end-to-end LLM inference

At this point it looks less like a hardware limitation and more like an engineering project. The old XDNA1 path appears to still be functional under Linux when paired with the modern amdxdna stack.

I'm mostly posting because I couldn't find many examples of people doing anything substantial with XDNA1 NPUs on Linux, and I thought others might find it interesting.

If there's interest, I'm happy to clean up the code and publish the project on GitHub.

https://redd.it/1u2tld0
@r_linux
Reddit From the linux community on Reddit Explore this post and more from the linux community
Post #42225 47
AMDGPU Linux Driver Preps For HDMI 2.1 Compliance Testing
https://www.phoronix.com/news/AMDGPU-DC-Preps-HDMI-Comp-Test

https://redd.it/1u23qrd
@r_linux
Phoronix AMDGPU Linux Driver Preps For HDMI 2.1 Compliance Testing While not as exciting as features like HDMI 2.1 FRL and Display Stream Compression itself, as part of AMD's efforts to provide a fully open-source HDMI 2.1 driver implementation for AMDGPU, new code is being prepped for their kernel driver to support the…
Older posts →
Threads Profile ViewerView any public Threads profile without an account.Open ThreadLook →Writing with AI? Make it sound human.Metric37 rewrites AI drafts so they read naturally. Free AI detector, 1,500 words free.Try Metric37 →