There is no support for ed25519 host keys (confirmed using ssh-audit). Would be nice to have though.

As an aside, you should use ssh-audit to get recommendations for what to disable as far as less than ideal options/configs go.

Here is the CO2 monitor https://github.com/oseiler2/co2monitor

I currently use Banktivity which is OK. Would love to hear from any others that have used Banktivity and migrated to something else. Ideally, there should be OFX support.

Missed a chance to call this swiftamp instead and avoid namespace collision.

Or swamp to be shorter

machamp

Isn't that a Pokeman?

Seems fishy

Agreed. I also wonder why they chose to test against a Mac Studio with only 64GB instead of 128GB.

Hi, author here. I crowd-sourced the devices for benchmarking from my friends. It just happened that one of my friend has this device.

FYI you should have used llama.cpp to do the benchmarks. It performs almost 20x faster than ollama for the gpt-oss-120b model. Here are some samples results on my spark:

  ggml_cuda_init: found 1 CUDA devices:
    Device 0: NVIDIA GB10, compute capability 12.1, VMM: yes
  | model                          |       size |     params | backend    | ngl | n_ubatch | fa |            test |                  t/s |
  | ------------------------------ | ---------: | ---------: | ---------- | --: | -------: | -: | --------------: | -------------------: |
  | gpt-oss 20B MXFP4 MoE          |  11.27 GiB |    20.91 B | CUDA       |  99 |     2048 |  1 |          pp4096 |       3564.31 ± 9.91 |
  | gpt-oss 20B MXFP4 MoE          |  11.27 GiB |    20.91 B | CUDA       |  99 |     2048 |  1 |            tg32 |         53.93 ± 1.71 |
  | gpt-oss 120B MXFP4 MoE         |  59.02 GiB |   116.83 B | CUDA       |  99 |     2048 |  1 |          pp4096 |      1792.32 ± 34.74 |
  | gpt-oss 120B MXFP4 MoE         |  59.02 GiB |   116.83 B | CUDA       |  99 |     2048 |  1 |            tg32 |         38.54 ± 3.10 |

Is this the full weight model or quantized version? The GGUFs distributed on Hugging Face labeled as MXFP4 quantization have layers that are quantized to int8 (q8_0) instead of bf16 as suggested by OpenAI.

Example looking at blk.0.attn_k.weight, it's q8_0 amongst other layers:

https://huggingface.co/ggml-org/gpt-oss-20b-GGUF/tree/main?s...

Example looking at the same weight on Ollama is BF16:

https://ollama.com/library/gpt-oss:20b/blobs/e7b273f96360

I see! Do you know what's causing the slowdown for ollama? They should be using the same backend..

Dude, ggerganov is the creator of llama.cpp. Kind of a legend. And of course he is right, you should've used llama.cpp.

Or you can just ask the ollama people about the ollama problems. Ollama is (or was) just a Go wrapper around llama.cpp.

Was. They've been diverging.

Now this looks much more interesting! Is the top one input tokens and the second one output tokens?

So 38.54 t/s on 120B? Have you tested filling the context too?

Yes, I provided detailed numbers here: https://github.com/ggml-org/llama.cpp/discussions/16578

Makes sense you have one of the boxes. What's your take on it? [Respecting any NDAs/etc/etc of course]

Curious to how this compares to running on a Mac.

TTFT on a Mac is terrible and only increases as the context increases, thats why many are selling their M3 Ultra 512GB

So so many… eBay search shows only 15 results, 6 of them being ads for new systems…

https://www.ebay.com/sch/i.html?_nkw=mac+studio+m3+ultra+512...

Does anyone know what was used to produce the graphs?

Do you mean charts? If so, it's Datawrapper: https://www.datawrapper.de/charts

One of the quite expensive paid plans, as the free one has to have "Created with Datawrapper" attribution at the bottom. I would guess they've vibe-coded their way to a premium version without paying, as the alternative is definitely outside individual people's budgets (>$500/month).

Inspecting the page, I can see some classes "dw-chart" so I looked it up and got to this: https://www.datawrapper.de/charts. Looks a bit different on the page, but I think that's it.

it is indeed Datawrapper like other posters have said. It works well for interactivity on a static blog like Hugo.

I saw a TikTok of someone saying that farmers are not stupid (due to the wide variety of skills to successfully farm) and were just betting on Trump not actually going through with tariffs.

It's hard to have any sympathy for such cynical behavior while simultaneously asking for handouts. Especially since the same people probably voted against others getting social services.

It also hurts when I drop the iPad mini on my face. In fact, I was considering getting a Pro Max to replace both a iPhone Pro and iPad mini combo but figured it might too big of a compromise.

I wonder if anyone has successfully gone down this path.

Do you have a link to a GitHub repo for this? Also, will it be mosh compatible?

Not yet, but hope to have something up in September! It’s unfortunately not most compatible - I thought about that but didn’t see a lot of value and there were some downsides like re-implementing an encryption layer that doesn’t make sense if you use WebRTC. Just curious, what’s your use case for mosh compatibility?

Does Whispering support semantic correction? I was unable to find confirmation while doing a quick search.

Hmm, we support prompts at both 1. the model level (the Whisper supports a "prompt" parameter that sometimes works) and 2. transformations level (inject the transcribed text into a prompt and get the output from an LLM model of your choice). Unsure how else semantic correction can be implemented, but always open expand the feature set greatly over the next few weeks!

They might not now how whisper works. I suspect that the answer to their question is 'yes' and the reason they can't find a straightforward answer through your project is that the answer is so obvious to you that it's hardly worth documenting.

Whisper for transcription tries to transform audio data into LLM output. The transcripts generally have proper casing, punctuation and can usually stick to a specific domain based on the surrounding context.