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

Those are multiple-choice questions. If some of them are "trick questions", where obvious answer (e.g. the value taken directly from question's text) is wrong, bad model might perform worse than a dice.

On the other hand, not sure where from 25% baseline for random answers come from. Since this is multiple-choice-out-of-4 test, random guessing should be correct in 1 in 15 cases, not 1 in 4.


Especially when the page source actually contains all the content (so no need to fetch it in JS), part of which is even presented in title.


It has the content three times (not including the title). Once in json (as escaped html) and twice in meta tags (as plain text). They used to have it in plain html too but that got hidden then removed.

You see read the unescaped html here:

https://beige.party/users/intransitivelie/statuses/117057396...

No stinking js or app required!


Selecting parts of the code (in F7 mode or with SHIFT-LEFT/RIGHT) highlights which individual pixels corresponds to the actual letters.


Some other examples:

- ACPI configuration for power management and platform stuff [1]

- Bitcoin transactions [2]

- TrueType fonts [3]

[1] https://wiki.osdev.org/AML

[2] https://en.bitcoin.it/wiki/Script

[3] https://learn.microsoft.com/en-us/typography/opentype/spec/t...


Since ACPI was mentioned, let's not forget about EFI!

https://uefi.org/specs/UEFI/2.10/22_EFI_Byte_Code_Virtual_Ma...


Since that page is a little dense, the higher-level version: PCI supports Option ROMs (OpRoms) - plug in device like a NIC or a GPU, your BIOS actually loads compiled code from it and executes it on the CPU. In many systems for example PXE booting (net booting) is actually a function of the NIC, executing code on the CPU to load an operating system. We're talking actual x86/x86_64 machine code here running in the privileged pre-boot environment. Not portable or secure in any way. OpRoms _may_ now be checked for SecureBoot signatures on systems where that's set up properly at least.

EFI ByteCode (EBC) is meant to help at least the portability side. I'm not sure if anybody is actually delivering devices with EBC OpRoms yet though. I'm also not sure if anybody is looking at using the EBC VM to sandbox untrusted OpRoms.


"Yet"? The only card anyone's ever found that shipped with an EBC option ROM was from about 20 years ago, nobody's migrating to EBC and the general approach is to just emulate the x86 instructions instead. And secure boot has been verifying option ROMs since 2012.


Does this imply that plugging in a NIC into an ARM or PowerPC machine might fail to pxe boot if the manufacturer hasn’t prepped code for those platforms?


Not "might" - will.

That's why there were separate "Mac editions" of certain cards (like GPUs) - the Option ROMs were different to support the Mac's frankensteined PPC OpenFirmware-like setup, and later to provide early EFI option roms when most x86-targeting cards were shipping with classic VBIOS.

EDIT: And while there was x86 emulator on many firmwares, it was often not enough to run everything, and x86 NIC firmware won't work for netbooting a PPC machine


The network is the computer


Is there a way to use this (or similar) model in Visual Studio? Extensions on Visual Studio Marketplace are clunky and sluggish at best, if they even work at all.


If you mean VSCode (or any other editor):

> We’re open sourcing the model weights so the community can build fast, privacy-preserving autocomplete for every IDE - VSCode, Neovim, Emacs, and beyond.

https://blog.sweep.dev/posts/oss-next-edit


No, I mean Visual Studio (the IDE), not Visual Studio Code (the editor).


Of course they are different products, but is there really a meaningful distinction between VS Code and an IDE? For all i care VS Code is a complete IDE.


You need to add (official) extensions for that though. Which makes VSCode more flexible


I thought there was already a generic plugin for this :(. Let's wait for one then ha, or I may just make one.


Also:

* Reverse polish notation

* Chinese remainder theorem

* Byzantine generals problem


All names given by people who were NOT of that nationality.


Did you, by any chance, play Outer Wilds recently?


> LoRa can reach a range of up to 3 miles (4.8 km) in urban areas, and up to 10 miles (16 km) in rural areas

In mountainous area LoRa on 868MHz band reaches over 100km. Last month we had a stratospheric balloon with a Meshtastic node attached. It established direct (albeit intermittent) connection between Warsaw and Berlin.


I've tested LoRa from Hill<->Hill and Flat Land<->Flat Land and 30-60km worked with plenty of SNR left.

On 868MHz with standard 5dBi omnis.

Thesis: as long you are using one of the more robust LoRa settings it always will work as long you have LoS or at least only lightly obstructed LoS.


If you have line of sight, a simple flashlight would work for communication

A robust communication needs to work in valleys


Blinking a flashlight 1200 times per second is a bit exhausting.


What's the data transmission rate with the balloon, put of curiosity?


Hard limit is 3.5kbps with 148dB link budget (Medium Fast preset in Meshtastic [1]). In practice - a packet worth of hundred bytes every half a second or so.

[1] https://meshtastic.org/docs/overview/radio-settings/


Is that the cap of the whole system or can multiple users use it with individual caps of 3.5kbps? That's somehow both faster and slower than I expected.


If one node kept transmitting non-stop, it could push data out at 3.5kbps and no one else in that range can transmit at the same time.

However, in EU there is a legal limit of 1% duty cycle on 868MHz band and collision avoidance mechanism, meaning on average you can send a packet (up to 255 bytes) once a minute.


Thanks for the info, that's really interesting!


> I'm a c# dev with near 20 years experience

Which IDE do you use? JetBrains Rider?


Not the person you asked, but I'm in a similar boat (15 years, polyglot but a lot of C#).

I mostly use VS Code to be honest. I use VSCode for other languages and for a long time it was the only graphical editor to have good remote development (over SSH) support.

Rider has that feature now though and is pretty nice too. I typically jump over to it when I need to profile something as it integrates with dotTrace. If you're coming from full-fat Visual Studio you'll probably prefer Rider.


Rider; however that's on a Windows work machine. We are a solid way to getting a linux/mac dev env going; maybe 30% is netstandard2.0, 10% is net9, the remainder net472 (including an old school non-sdk web app on IIS). Maybe ~ million LOC in its 14 year lifespan.

My personal dev is shifting to Rust.


> if you run it at the full 262144 tokens of context youll need ~65gb of ram

What is the relationship between context size and RAM required? Isn't the size of RAM related only to number of parameters and quantization?


The context cache (or KV cache) is where intermediate results are stored. One for each output token. Its size depends on the model architecture and dimensions.

KV cache size = 2 * batch_size * context_len * num_key_value_heads * head_dim * num_layers * element_size. The "2" is for the two parts, key and value. Element size is the precision in bytes. This model uses grouped query attention, which reduces num_key_value_heads compared to a multi head attention (MHA) model.

With batch size 1 (for low-latency single-user inference), 32k context (recommended in the model card), fp16 precision:

2 * 1 * 32768 * 8 * 128 * 36 * 2 = 4.5GiB.

I think, anyway. It's hard to keep up with this stuff. :)


Yes but you can quantise the KV cache too just like you can the weights.


A 24GB GPU can run a ~30b parameter model at 4bit quantization at about 8k-12k context length before every GB of VRAM is occupied.


Not quite true. Depends on number of KV heads. GLM4 32b at IQ4 quant and Q8 context can run full context with only 20GiB VRAM.


No. Your KV cache is kept in memory also.


Whats the space complexity for context size? And who is trying to drop it into linear complexity?


I mean...where do you think context is stored?


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

Search: