benreesman

12 ms·

91.
by benreesman 1y ago
GP's solution is the one that I (and I think most people) ultimately wind up using. But it's not a nice time in the usual "oh we'll just compile it" sense of a typical package. flash-attn in particular has its build
92.
by benreesman 1y ago
It's really difficult to do Python projects in a sound, reproducible, reasonably portable way. uv sync is in general able to build you only a package set that it can promise to build again. But it can't in general build torch-tens
93.
by benreesman 1y ago
I think jujitsu is interesting it it's own right! It seems non-obvious that you would have to prohibit git commands in general, they're already "buyer beware" with the current tool (and arcanist for that matter). Certain
94.
by benreesman 1y ago
I'm a pre-mercurial arcanist refugee who tends to promote Graphite in teams that are still struggling with mega-PRs and merge commits and other own goal GitHub-isms. Big fan in general even with the somewhat rocky scaling road we'
95.
by benreesman 1y ago
In my defense its a remix of several copypastas.
96.
by benreesman 1y ago
I also consume all the heavily sibsidized LLM tokens I can find a use for. Great deal for us, the people who funded it? Not so fun.
97.
by benreesman 1y ago
Technically what people often refer to as "Linux" should more properly be called "GNU slash Linux", or as I've taken to calling it, "GNU plus Linux", GNU being a holistic operating system with multiple ker
98.
by benreesman 1y ago
There's no weakness in identifying an adversarial situation and acting accordingly. Your adversary is playing to win. Why should you forfeit the contest?
99.
by benreesman 1y ago
I think OP just means that in the sentiment analysis parlance, not in the critical of the post sense. Though it does sort of show the Overton window that a pretty bland argument against always believing some rich dudes buckets as negative e
100.
by benreesman 1y ago
Anything adversarial is going to be the best people using the best assist: military drones, trading, anything performance hot pathed (AI inference itself), thats all too competitive to be anything but hybrid. But by a similar argument, most
101.
by benreesman 1y ago
It is unfortunate that pretty basic "save/load" functionality is still spotty and underdocumented, seems pretty critical. I use gptel and a folder full of markdown with some light automation to get an adequate approximation o
102.
by benreesman 1y ago
This is a very common statement but doesn't match my experience at all, unless you expand "business rules" to mean "not code already". There's plenty of that work, and it goes by many names ("enterprise&qu
103.
by benreesman 1y ago
People will go to extremely great lengths to debate the appropriate analogy for how these things work, which is fun I guess but in a "get high with a buddy" sense at least to my taste. Some of how they work is well understood (a l
104.
by benreesman 1y ago
I tend to agree in this instance. But this is why you don't build a public brand of doing shit very much like this constantly. Innocent until proven guilty is the right default, but at some point when you've been accused of miscon
105.
by benreesman 1y ago
Try doing CUDA stuff. It's a chemical fire. And the money would make solving it would fund arbitrary largesse towards OSS in perpetuity.
106.
by benreesman 1y ago
I'm guessing it's the right PyTorch and FlashAttention and TransformerEngine and xformers and all that for the machine you're on without a bunch of ninja-built CUDA capability pain. They explicitly mention PyTorch in the blog
107.
by benreesman 1y ago
It's always been possible to vibe code, it's just really fast now! I've done slipshod work full of bugs and security problems and thrown it over the fence hoping it will stand up long enough to be someone else's problems
108.
by benreesman 1y ago
Really appreciate a fairminded Rust person chiming in. It's a disservice to the language (which is cool!), and it makes the community look bad. Rust might gain some adoption among people who are new to high performance software and see
109.
by benreesman 1y ago
The RESF is not a conspiracy theory, its a known thing. Especially on HN: https://news.ycombinator.com/item?id=14178950
110.
by benreesman 1y ago
I think we have a winner. This sort of personal toolsmithing is fun, and you can try out some new programming language or whatever. We all love a good excuse to build something small-to-medium sized for our own perfect "tailor fit&quot
111.
by benreesman 1y ago
Strix Halo is one of the tier-1 out-of-the-box perfect targets for for // hypermodern // nixos so I'm tuning it for the EVO-X2 which was the first desktop one available (and we'll support omarchy's hyprlan
112.
by benreesman 1y ago
Minimum wages are an economically imperfect (as you've pointed out) but politically possible way to put some downward pressure on much, much bigger failures of our species and society to have attitudes and policies around acceptable
113.
by benreesman 1y ago
Eh, as a network administrator you want the netlogs on by default and you very clearly onboard everyone to the network with a memorable warning to do their personal browsing over some other interface. You've usually got at least some m
114.
by benreesman 1y ago
We as a community have decided to absolutely drench the front page with low-effort hot takes by non-practitioners about one of many areas of modern neural network ... progress. This low-effort hot take is every bit as "valid" as
115.
by benreesman 1y ago
Google will win the LLM game if the LLM game is about compute , which is the common wisdom and maybe true, but not foreordained by God. There's an argument that if compute was the dominant term that Google would never have been anyt
116.
by benreesman 1y ago
Doesn't the fact that a modern FPGA-centric (probably ASICs in the mix too at this point) hybrid NIC/order-parser/state-machine thing is rumored to be able to hit glass-to-glass of ~20-40ns mean that the speed game is hotter
117.
by benreesman 1y ago
I don't think it's either useful or particularly accurate to characterize modern disagg racks of inference gear, well-understood RDMA and other low-overhead networking techniques, aggressive MLA and related cache optimizations tha
118.
by benreesman 1y ago
`ggerganov` is one of the most under-rated and under-appreciated hackers maybe ever. His name belongs next to like Carmack and other people who made a new thing happen on PCs. And don't forget the shout out to `TheBloke` who like singl
119.
by benreesman 1y ago
I won't use `ollama` on principle. I use `llama-cli` and `llama-server` if I'm not linking `ggml`/`gguf` directly. It's like, two extra commands to use the one by the genius that wrote it and not the one that the guys ju
120.
by benreesman 1y ago
I'm torn on this, I was a fan of the project from the very beginning and never sent any of my stuff upstream, so I'm less than a contributor but more than don't care, and it's still non-obvious how the split happened. Bu

More ›