Hacker News new | past | comments | ask | show | jobs | submit | quibono's comments login

Is Ollama just the porcelain around llama.cpp? Or is there more to it than that?

They also decided to rehost the model files in their own (closed) library/repository + store the files split into layers on disk, so you cannot easily reuse model-files between applications. I think the point is that models can share layers, I'm not sure how much space you actually save, I just know that if you use both LM Studio + Ollama you cannot share models but if you use LM Studio + llama.cpp you can share the same files between them, no need to download duplicate model weights.

yes, it's a convenience wrapper around llama.cpp

The main feature IMO is the model library. llama.cpp on its own does not come with any built in way to download and manage models.

Is there a way to actually write anything more complex using React hooks without shooting oneself in the foot and ending up having to debug useEffect dependency chains?

I think I like the idea of react hooks and have got better at using them... Maybe it's just me but it really feels like an abstraction that's (too?) easy to abuse in the context of SPAs.


The author says they went about learning CL the wrong way. I wonder if there is a standard "community approved" way of learning the language?


Not standard, but hopefully worth mentioning: the thing that's clicked best for me is the docs on https://ciel-lang.org/ ("batteries included" Common Lisp image). The examples for how to use it's curated libraries matches how I try to integrate a new language into my toolbox.

It hit the front page a while ago too: https://news.ycombinator.com/item?id=41401415


Gentle Introduction to Symbolic Computation is a great book, I learned a lot. The 1990 version available for free below has aged well but you need to look elsewhere for getting setup with emacs and slime or whatever environment you want.

https://www.cs.cmu.edu/~dst/LispBook/


I began learning Common Lisp (CL) from the Common Lisp HyperSpec (CLHS): <https://www.lispworks.com/documentation/HyperSpec/Front/Cont...>. When I began learning CL about two decades ago, I did know of any other easily available source, so CLHS was my only source back then and I think it has served me well.

A popular recommendation these days is Practical Common Lisp (by Peter Seibel): <https://gigamonkeys.com/book/>.


Correction: s/I did know/I did not know/


There's this great post by Steve Losh:

https://stevelosh.com/blog/2018/08/a-road-to-common-lisp/


If you were to start over would you still go for Koa, or something else instead?


I did start several small projects and always used koa without reconsidering it.


How mature is the ecosystem built around this? Is this a production-ready framework or more of a beta/WIP thing?


Hey, any chance you could link it please?



Thank you!



Yes, GP noticed that the image is AI generated and they made an assumption that probably the text is AI generated too, since the image is.

The image is AI generated.

The text is not AI generated.


Take the hint: your slop hero image is damaging the rest of your page by association. If you don't want people to assume your page is garbage, don't lead with garbage.


Feedback received.


I personally don't actually mind and don't infer the rest of the page to be garbage. Maybe because I don't consider the image garbage either.


I think the video linked in the post does this, except for the 2-dimensional case.


What are some use cases for lambdas? I've not got around to doing anything with them yet and I wonder what they are mostly used for.


I think lambdas work really well for doing things as a reaction.

So for example, uploading an image to s3 could trigger a job, or having them run in parallel on a job queue since they theoretically scale way better then spinning up lots of 'job processors' to deal with influx.

One of the most fringe but interesting usecases I'm aware of is it being suggested as part of co2 (and cost) reduction. The main idea is that since lambdas only run the exact amount of time needed, we don't continue running/reserving costly machine time. This is still something where research is being done and does mostly pass the responsibility to the infrastructure / PaaS solution though


Here's a strong opinion semi-loosely held:

The fundamental difference between functions-as-a-service and just deploying to a damn server is that you get the ability to scale very quickly.

Our app has a feature which is very demanding but used infrequently. We can split the workload and spin up dozens of lambdas instantly to provide acceptable response times. Wouldn't want to have a server over-provisioned to handle that amount of scale at all times.

All the other purported benefits (e.g. you can plug Lambda into S3 events) are just ways to do vendor lockin to yourself. And the ops-less-ness is cool but also you could deploy to a container platform for that.


I've heard they are good for prototyping things where you need to change APIs and spin up new stuff easily. It might be better than using a service where you're basically required to have devops.

That said, you can also just spin up a VPS and use systemd to manage a service at small scale. There are also other PaaS solutions in addition to "serverless/lambda".

I've heard some horror stories about "lambda" pricing as well, so I'd make sure to read into that before becoming too dependent.


Do you think having a normalization layer between serverless and server-based approaches could actually help with the issue of becoming too dependent?

I find that the tight integration with providers like Azure, AWS, or CloudFront can make switching between them a very costly upfront decision.

Another disadvantage of serverless is that it’s quite challenging to produce an on-premises or enterprise-ready codebase if the need suddenly arises.


Being able to run things locally (like this post describes) is the only way anything should be accepted.


Or you can upload a cgi script to a web host like hetzner, then you have “serverless” too.


Could someone give me a quick run down on what Astro is good for please? Looking at the website it seems to cater heavily towards "marketing sites, blogs, e-commerce websites". I'm wondering if this is reflected in the design and dev experience or if it could be used for "generic" websites too.


You can use it for anything especially with its new server islands.

https://astro.build/blog/astro-5/#server-islands

So you can mix static rendered along with server side rendering pieces and get the benefits of both.

Also it's frontend framework agnostic and you can mix and match.

So you can have a dynamic react component, with a static Astro component, svelte component all in the same page on theory.

https://docs.astro.build/en/concepts/islands/

Then for backend stuff it kinda depends on how you deploy it. It could just be a single monolithic server or serverless per API route and probably everything in between.


Consider applying for YC's Spring batch! Applications are open till Feb 11.

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

Search: