iAkashPaul

Those slanted underlines on hyperlinks was really cool though

OnePlus 6T with Android 14 is still working like a charm, this even beat out Nothing Phone 1 in getting the latest Android update

haha, it needs some finesse to start swiping in from outside the trackpad

Anyone interested in a flyby of a Pokemon themed forest HEIC/wallpaper for macOS?

That calculator widget in the sidebar was my favourite. One swipe & it's exactly where it's supposed to be, was hoping it would be back since widgets came back now, but no.....

That recent push by Roku for accepting updated EULA around arbitration makes quite a lot more sense

If only Zed got more traction & in-turn extensions, that's more performance than what most need

Was expecting better battery life with 5nm->3nm, but it doesn't appear to be the case sadly. Logic board in the lid & more room in bottom case for battery could boost to 24hrs hopefully in the future.

Exactly

True, Grammar has been around with Llama.cpp since the early days IIRC. Microsoft had Guidance as well & now TGI supports it as well. End game for Langchain/Llamaindex if just guarantee or structure adherence was the only reason to use & beg the LLM for usable output.

TGI just integrated Guidance in 1.4.3, that by itself can support both grammar/JSON/Pydantic & tool invocation/function calling.

Langchain & Llamaindex plus Fructose really need to skip the structure adherence work & move to chunking/KG generation since that's the next pain point to tackle.

I think PyTorch recently added FP8 & even Intel's Neural-Compressor has experimental support. But yeah most HF LLM models are currently just INT8 or FP4/NF4 loaded thanks to BitsandBytes by Tim Dettmers.

Basically augmenting users parsing PDFs & looking to prefill values into Excel instead of typing it all out. Ex. Liabilities, time-period/frequencies mentioned, owners of clauses etc.

Running Mistral-Instruct-0.1 for call/email summarization, Mixtral for contract mining & OpenChat to augment agentic chatbot equipped with RAG tools(Instruct again).

Experience has been great, INT8 tradeoffs are acceptable until hardware FP8(FP4 anyone?) becomes more widely & cheaply available. On-prem costs have been absorbed already for few boxes of A100s & legacy V100s running millions of such interactions.

This is something for the Apple Vision Ultra to integrate along with electrochromic glass or some physical cover to go from full pass through to fully immersed VR. Would look like a clear version of those wrap around Star Trek glasses.

Indirectly via TGI

I've quantised 2B & 7B Instruct models here https://huggingface.co/iAkashPaul/gemma-2b-it-gguf & https://huggingface.co/iAkashPaul/gemma-7b-it-gguf

Unable to get proper string reversal even with Mixtral

https://hf.co/chat/r/H8Jrn_f

Based off Karpathy's BPE video highlighting ChatGPT being unable to reverse this string-> .DefaultCellStyle

I tend to have a preflight script which is run prior to the final docker command. This let's the layers have the cached weights & avoids dealing with downloading or making changes to the codebase for loading downloaded weights. Would shave off 10s from both providers.

The lack of kitted out editors for it is a bigger problem, plenty WYSIWYG editors exist but primarily for CMS apps only & not for standalone usage.

Markdown editors with JS macros & embedded CSVs should have been pushed harder by public institutions instead of proprietary formats & licensing

Taildrop, WebRTC based Snapdrop or plain old 'python -m http.server 2077'

Use this whitespace reducing config with HiddenBar for a compact menu bar icon set

HiddenBar

HiddenBar is a FOSS alternative to Bartender, combined with reduced whitespace it'll be quite nice!

Who knows we might just get Microsoft Mesh through Apple Vision Pro after all instead of HoloLens updates

The amount of obfuscation & head-in the-ground was nauseating to read

Wonder how difficult it would be to write something akin to PyO3 but with python syntax for writing rust programs, just like a subset

This and PixelExperience have kept my devices ahead of even new OEM devices.

Summarisation for calls, emails. Lots of extraction tasks & closed domain chatbots.

Deployment is usually FastAPI for business logic, Langchain or MS/Guidance library, LLM hosted via. HF-TGI server