Hacker Newsnew | past | comments | ask | show | jobs | submit | kpw94's commentslogin

Remember just 6 months ago that "Ggml.ai joins Hugging Face to ensure the long-term progress of Local AI" (https://news.ycombinator.com/item?id=47088037)

(Ggml.ai is llama.cpp.)

Curious if the “I consider HuggingFace more "Open AI" than OpenAI” sentiment in that top comment will still apply with NVIDIA as the boss now...


Nvidia is promoting "unmetered intelligence". Jensen wants everyone to have a spark dgx with an nvidia chip in their home.


Wasn’t Huang the guy who said your $500K employee should burn at least half his salary on tokens? I don’t think Nvidia is focused on cost-effective general computing for anybody.


It is possible to serve different markets with different agendas and have both of them serve nVidia’s business. Think of all the freemium software out there; widespread so everybody can use it, but the people who REALLY use it REALLY pay up.


Maybe but they make their real money in data centers. Certainly, they want to have fingers on every trend. Owning something the Government wants to control gives you more access to trade repression of open models for more Government favors and money.


NVidia has reasonable incentives to keep things open indefinitely, it wants people to use it's GPUs


Nvidia is halving the production of its higher end GPUs, doubling prices and heavily segmenting the market. They don't give a single shit about individuals. One wafer that makes 10 5090s that might sell at 3000 each, or one wafer that it already sold 6 months ago for 200k to one of the incestuous AI companies it works with?


Issue is that llama.cpp is the best way to run models on hardware that isn't nvidias.


A lot of llama.cpp contributions come from the community and ecosystem, like Unsloth. If something goes awry, I fully expect lots of forks.


There already are a lot of forks for things they decline to implement. TurboQuant, ROCmFPX, and more. I need to set up an agent that will loop on merging them.


Except when they have less than 16 gb of ram?


It'd be really nice if I could use my egpu 4090 on my macbook pro..


For LLM workloads this might help? https://docs.tinygrad.org/tinygpu/


Interesting, I've never seen this, thanks


Agree that go is the best due to its main design goal: A language that's simple for any programmer fitting that definition https://news.ycombinator.com/item?id=30688969.

> "They’re not capable of understanding a brilliant language but we want to use them to build good software. So, the language that we give them has to be easy for them to understand and easy to adopt."

This makes it a great language not just for young Googlers programmers, but also for LLM Agents!

IMO, the next big language will be similar philosophy, but without garbage collection. (is Zig the closest to filling that niche?)


> is Zig the closest to filling that niche?

Given what you said about Go, presumably that is Solod (https://solod.dev)


Interesting project.

2 gut feeling concerns:

- Strict subset of go might be confusing to an agent actually (trying to use unavailable go features)

- So -> c11 source to source compile loop might be confusing to agent: if So compiles to c11 does it guarantee c11 program compiles. If runtime exception (segfault etc), is it going to be easy for agent to map that back to original So code?


> OpenAI and Anthropic, which are gearing up for potentially massive IPOs, did not sign the letter.

Not anymore, OpenAI did sign it:

https://www.microsoft.com/en-us/corporate-responsibility/top...


In the context of local LLMs on limited hardware I've ran to the exact same conclusion: "tok/s" isn't the most useful metric when my personal North star metric, given my fixed hardware is: Model smart enough to execute my goals _in the minimum amount of time_.

Some models I tried (Mistral I think) had better tok/s, and roughly same billion parameters / scores on various benchmark... But they were _so_ verbose, that they generated many more tokens compared to a Qwen model of same caliber to answer the same thing.

So even though it had better generated tok/s, because so many more were generated, the clock time was longer.

And this compounds over mutli-turns: more generated token means more context used in the next turn (until some compaction or something runs)


Even more important in a local context is the difference between token generation and prompt processing speed. We tend to focus on the former, but for multi-turn/agentic workflows the latter can dominate.


Yeah definitely. I've recently commented on that: https://news.ycombinator.com/item?id=48557890


I feel like we need to see more proliferation of local LLMs to start seeing ones turned to be terse, rather than maxing the amount of tokens user pays for


> What it does:

>

> --jinja for tool calling support

Pretty sure this flag hasn't done anything for a while. It's enabled by default since ~November of last year


The huge spike of "lk-99" in science & frontier tech is amusing...

This is cool concept, would love a positive/negative sentiment computed for each comment that refers to a given word, so you can see trends of "cloudflare (positive)" vs "cloudflare (negative)" where first one counts comments only if sentiment confidence is greater than say 0.6 and the other one counts comments only if sentiment is less than 0.4 (assuming [0,1] sentiment score)


> About the generation speed: ~100-150 t/s on the RTX 5090 and ~40 t/s on the Mac

Curious if you can share the prefill speed too?

I run locally on a crappy desktop (some AMD iGPU with Vulkan llama.cpp, 32 GB DDR4 RAM) for experimentation. I get 15 tok/s on generation for the qwen & gemma4 MoE models. I get around 150 tok/s prefill speed.

Reason I'm asking about the prefill is looking at my stats at work, I use between 20M to peaks of 300M input tokens daily. Some of those token are cached but in general, I seem to have roughly 500x more input tokens than output. So interested in prefill tok/s stats.

Huge Thank you for llama.cpp btw!!


Here are the prefill speeds:

    Device 0: NVIDIA GeForce RTX 5090, compute capability 12.0, VMM: yes, VRAM: 32109 MiB
  | model                          |       size |     params | backend  |  fa |            test |                  t/s |
  | ------------------------------ | ---------: | ---------: | -------- | --: | --------------: | -------------------: |
  | qwen35 27B Q4_K - Medium       |  15.92 GiB |    27.32 B | CUDA     |   1 |   pp2048 @ d512 |      3714.02 ± 10.85 |
  | qwen35 27B Q4_K - Medium       |  15.92 GiB |    27.32 B | CUDA     |   1 |  pp2048 @ d1024 |      3684.86 ± 15.21 |
  | qwen35 27B Q4_K - Medium       |  15.92 GiB |    27.32 B | CUDA     |   1 |  pp2048 @ d2048 |       3650.80 ± 8.53 |
  | qwen35 27B Q4_K - Medium       |  15.92 GiB |    27.32 B | CUDA     |   1 |  pp2048 @ d8192 |       3473.88 ± 0.97 |
  | qwen35 27B Q4_K - Medium       |  15.92 GiB |    27.32 B | CUDA     |   1 | pp2048 @ d32768 |       2754.69 ± 4.07 |

  ggml_metal_device_init: GPU name:   MTL0 (Apple M2 Ultra)
  | model                          |       size |     params | backend  | fa |            test |                  t/s |
  | ------------------------------ | ---------: | ---------: | -------- | -: | --------------: | -------------------: |
  | qwen35 27B Q8_0                |  26.62 GiB |    26.90 B | MTL      |  1 |   pp2048 @ d512 |        379.75 ± 0.21 |
  | qwen35 27B Q8_0                |  26.62 GiB |    26.90 B | MTL      |  1 |  pp2048 @ d1024 |        377.15 ± 0.35 |
  | qwen35 27B Q8_0                |  26.62 GiB |    26.90 B | MTL      |  1 |  pp2048 @ d2048 |        371.46 ± 0.91 |
  | qwen35 27B Q8_0                |  26.62 GiB |    26.90 B | MTL      |  1 |  pp2048 @ d8192 |        344.84 ± 0.41 |
  | qwen35 27B Q8_0                |  26.62 GiB |    26.90 B | MTL      |  1 | pp2048 @ d32768 |        222.42 ± 5.29 |

Btw, based on your numbers, I think our use cases are quite different. I use the agent for very targeted sessions - basically things that are clear to me how to do, just want to automate them. My workflow is usually: new session -> read this, this and this -> do that. I.e. I don't let it wander at all in the codebase, so I rarely exceed the context window.

Also, I get a lot of mileage from the ngram-based speculative decoding functionality [0] as it allows me to iterate on the implementation much faster.

[0] https://github.com/ggml-org/llama.cpp/pull/19164


Thanks! Super helpful.

I do use it the same way as you're describing on personal projects at home, in a very crude manner (pasting code snippets in llama server web UI prompt. Next will attempt OpenCode)

At work I use it in similar manner with more mature tools, but the vast majority of token spend comes from a totally different workflow: "pretend the AI is a fleet of junior/intern engineer you're delegating work to", where the agent will on its own do the implementation, commit the changes etc.

It does indeed spend a lot of tokens wandering the codebase, talking to MCPs, loading skills etc.


> gemma (unsloth/gemma-4-26B-A4B-it-GGUF) models

Since you're running quantized (at UD-Q4_K_XL) , check out the "qat" models (unsloth/gemma-4-26B-A4B-it-qat-GGUF) !

- https://huggingface.co/unsloth/gemma-4-26B-A4B-it-qat-GGUF (With "Jun 9 Update: Added MTP support.")

- https://blog.google/innovation-and-ai/technology/developers-...


TIL:

> Quantization-Aware Training (QAT) [...] allows preserving similar quality to bfloat16 while dramatically reducing the memory requirements to load the model


How is the the QAT models at coding? I looked for opinions since the release and haven't found much.


I did the opposite switch:

In ~2015 got an Xbox one, as a media center it was an awesome experience:

Kinect voice control to play/pause and other things way before Google home/Amazon echo ecosystem were mature.

Free OTA channels via TV tuner and well designed OneGuide (with ability to pause and rewind).

And of course all the Netflix and other apps, Plex server etc.

But strategically it seems Microsoft decided they wanted to look more like Playstation, focused on gaming (at that time paid Xbox live subscription vs free Playstation)

And as gaws points out, they seem to recently announce to double down on the gaming stuff.

So when they discontinued OneGuide. I picked Roku since they seem to be focused on the media experience primarily... but unsure how I feel about this acquisition news.


> What's hard to figure out here?

Negative externalities are hard to figure out.

Since parent mentions "toxic byproduct": Say you're the company that invented Teflon pans. you made billions. You saved billions in time for all the users of the pans... A true entrepreneurial success.

But, by how many billions did you fuck up the environment, people's health etc with the spread of PFAS everywhere?


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: