jyap

I run it with Llama.cpp on my RTX 3090. Also using the same Unsloth model.

My config is similar to: https://github.com/noonghunna/club-3090/blob/master/docs/eng...

I need to try out some of the other set ups mentioned in this repo for increased TPS.

The Usual Suspects

Per the title, this is a leaked benchmark. It’s not an Ars Technica full benchmark article.

Regarding market share and your friends and family recommendations, you’re thinking first world. Rest of the world wants and can only afford sub-$500 laptops.

It’s all relative. I’m at 24.4% but I have quite a few devices like Wemo light switches at the top of my DNS queries. Only have one Amazon Alexa device but that’s near the top as well.

IoT devices which constantly phone home will skew things.

There’s not many free as in open source alternatives out there. There’s traditional CAD such as Fusion 360 which gives you a limited amount of designs and is “free” for non-commercial use.

So then under the open source, own your designs umbrella you have stuff like FreeCAD which is similar to traditional CAD which means you have a learning curve.

OpenSCAD is programmatic which suits someone with a software engineering background. Plus being free in the sense of owning your designs.

His original sentence was life imprisonment without the possibility of parole.

So you can’t agree with the original sentence and then say he “absolutely deserved to be released.”

Without the chance of parole, a pardon from the president is one of the few ways he could get out of jail.

Quick correction: I currently use Wezterm on Linux and it has tabs. Alacritty does not for developer philosophical reasons.

Looking forward to checking out Ghostty.

I started using Devenv which utilizes Nix. You might want to check that out.

https://devenv.sh/

One example is that Microsoft owns GitHub.

A few that come to mind:

Jekyll - written in Ruby

Hugo - written in Go

Zola - written in Rust

This 236B model came out around September 6th.

DeepSeek-V2.5 is an upgraded version that combines DeepSeek-V2-Chat and DeepSeek-Coder-V2-Instruct.

From: https://huggingface.co/deepseek-ai/DeepSeek-V2.5

The article says:

Average response size also halved from 1506 bytes to 778 bytes, the compression algo in the Rust library must be better than the one in the JS library

If you want to watch a film similar to these themes, check out Playtime from 1967 by Jacques Tati. Something I watched recently.

Playtime works wonderfully as a cinematic satire of contemporary society and our roles.

The film's office scenes are particularly striking because they visualize the dehumanizing aspects of modern office design that cubicle farms would later become notorious for - the uniformity, the loss of privacy despite physical separation, and the way architecture can make people behave more like components in a machine than individuals.

The thing about Playtime is the constant onslaught of appropriate explorations on the theme of an increasingly technical life.

By the time the end of Playtime rolls by, what you get is an examination of what it is like to be human and hopefully you feel a sense of joy in being a part of the human race.

At a former workplace I wrote this tool which was put to use.

It utilizes GPG to store the secrets and Golang templates to support the files.

https://github.com/jyap808/jaeger

In the function signature, the return variable err is type error. Since it is named it is also defined and initialized as nil.

In the code, an error is found and it does not assign a value to err and just returns it as the error value.

So it returns nil as the error when it wants to return an error with a proper value.

The code should be something like:

    if ctx.Err() != nil {
      return 0, 0, ctx.Err()
    }

The human could discuss with the officer and say they were waiting for someone or be asked to move on out of the bike lane. The difference being the human interaction.

Don’t forget his Billionaire status came from fraudulent activities (taking other people’s money).

This isn’t punishing entrepreneurs.

Having done the ink jet song and dance for a while with multiple printers over the years I bought a color laser jet Brother HL-3170CDW printer 10 years ago for $170. Wireless and it just works. Economical as well with the cartridges.

For photo quality I just use the local pharmacy.

It makes “sense” based on the domain of the cloud provider being DevOps teams who are maintaining and using these CLI tools. Ie. What they use day to day.

For anything more advanced they offer language specific SDKs in Rust, Swift, Kolton, etc…

For example integrating storage in an iOS app.

Not necessarily. Depends on their definition and process.

“Voluntary regrettable” would imply the employee left voluntarily but regrettable because the company would have preferred they stayed.

“Non-regrettable” means they left for reasons that aren’t regrettable such as performance or natural attrition. Either they left on their own or were fired. Hard to say since they don’t specify.

Lol this kept me entertained for a bit. Found it a couple of weeks ago.

Hoping my number 1 ranking on Intermediate stays up a while.

I’ll need to see if I can crack that Expert high list.

Cheers

I’ve never used GH Copilot so can’t comment on that.

But having everything locally means no privacy or data leak issues.

You would want to test it out manually day to day. That’s always the best. Some models can out score but not actually be “better” when you use it.

But there is also the benchmarking: https://github.com/deepseek-ai/deepseek-coder

33B Instruct doesn’t beat 6.7B Instruct by much but maybe those % improvements mean more for your usage.

I run 6.7B since I have 16GB RAM.

Quantization of the model also makes a difference.

Golang grinding Leetcode coding buddy. And general coding buddy PR reviewer.

The results are on par or better than ChatGPT 3.5.

I often use it to delve deeper such as “is there an alternative way to write this?” Or “how does this code look?”

If you have an M-series Mac I recommend trying out LM Studio. Really eye opening and I’m excited to see how things progress.

Deepseek Coder Instruct 6.7b has been my local LLM (M1 series MBP) for a while now and that was my first thought… They selectively chose benchmark results to look impressive (which is typical).

I tested out StableLM Zephyr 3B when that came out and it was extremely underwhelming/unusable.

Based on this, Stable Code 3B doesn’t look to be worth trying out. Guessing if they could put out a 7B model which beat Deepseek Coder 6.7B they would have.

> But couldn't they have dozens of variations of the guy revealing his system?

Interesting thought. There is a big cost to this on the ad creator side. I also wonder about how they would make a customizable standardization for this?

In email we can use fields like your name and job title, for example with targeted LinkedIn recruiters.

I can foresee video ads which utilize AI calling your name.. I guess this is Minority Report..

But underlying this modern web ads already give us granular targeting though interests. So most of the benefit is already there.

Prime does give unlimited photo storage (not videos).

Probably the one perk keeping me locked in.

bsky-social-twh36-msidy

bsky-social-cd3fm-4j6hg

bsky-social-jgzjd-x7fn3

bsky-social-wfiqb-wdcej

bsky-social-y4x6q-55aiv

Merry Christmas (or happy holidays)