There's a difference between iterating on your code patterns until you get something working and iterating on the solution as requirements evolve. You're only two months in so you still have loads to learn, but eventually you'll come to know all the "best practices" so that when you're presented with a problem, you can immediately know what the ideal solution should look like. The caveat there is that the ideal solution is subject to change as the situation changes.
This is why it's important to have API access available (for a reasonable price, let alone free). The people who want Twitter data are now loading tons of completely unnecessary and heavy multimedia content just to parse it manually for the comparatively light text data they care about. It would literally save Twitter money to give the data away.
As an experienced programmer, I can say there's nothing inherently wrong with hardcoding data. If you're not presented with clear requirements up front (i.e. all the different use cases your code should handle) then it makes more sense to get something working and refactor as you go. In your case you said the data was subject to change, which doesn't necessarily mean you could abstract it out from the start (unless you were told exactly how it would change). In general, however, software development is all about iteratively refining your assumptions about how things should work while simultaneously juggling the changing demands of stakeholders. One of the most useful rules of thumb is the DRY approach (don't repeat yourself). This is complemented by WET (write exactly twice). Together this means that if you find yourself repeating logic three times, that's enough for you to refactor into a single, generalized function. Repeating twice is generally fine because your third use case might be sufficiently different that a premature refactor is a waste of effort. As you gain more experience programming, you'll find it's less about technical proficiency and more about working efficiently, creating fewer headaches, etc.
Any way to prune LLMs?
Hey, I'm working on some local LLM applications and my goal is to run the smallest model possible without crippling performance. I'm already using 4 bit GPTQ but I want something smaller. These models have been trained on such a massive amount of data but my specific use case only touches a very very small fraction of that, so I would imagine it's possible to cut away large chunks of the model that I don't care about. I'm wondering if there has been any work on runtime pruning of LLMs (not just static pruning based on model weights) based on "real world" data. Something like: you run the model a bunch of times with your actual data and monitor the neuron activations to inform some kind of pruning process. Does anyone here know about something like that?
Lemmy Guide For Reddit Refugees
Lemmy Background
Hey there, fam! Welcome to Lemmy, the lit and zestful fediverse alternative to Reddit. Named after the legendary Lemmy Kilmister, we're here to vibe with the online community experience. Here's the 411 to help you get the lowdown on our citrus-infused world:
What's the Deal with Lemmy, Though?
Lemmy is like the chillest spot on the interwebs, where we can come together, share dope content, and connect with fam. It's all about embracing the spirit of Lemmy Kilmister, the OG frontman of Motörhead, and spreading those good vibes.
FAQ: Frequently Asked Questions
- What are communities? Here at Lemmy, we call our subreddits "communities." They're vibrant spaces where like-minded folks come together to discuss, share, and connect on specific topics. Think of them as hubs of shared interests and discussions.
- Who are the lemons? Lemons are us, the lit and active users of Lemmy.
- What's the deal with zest points? Instead of "karma," we use "zest points" to keep track of the love and clout our posts and comments get.
- How do I show love with an upvote? In Lemmy, we keep it real by giving a "squeeze" to posts and comments we're feeling, on god.
- And downvotes? If something's whack or doesn't hit right, we "pulp" the post or comment, no cap.
- Can I tweak my posts? Fo sho! You can edit your posts in Lemmy, but make sure to flex and mention your edits to keep it all honest and transparent, fr fr.
Reddit/Lemmy Lingo Decoder
To help you fit right in, here's a quick decoder to translate some Reddit lingo into our Lemmy swag:
- "The narwhal bacons at midnight" becomes "When the thunder roars, the lemons gather", on god
- Karma becomes "zest points", no cap.
- Upvoting becomes squeezing, cuz we feelin' it.
- Downvoting becomes pulping, when something's just not hittin' right.
- "Edit: Thanks for the Gold Kind Stranger!" becomes "Edit: Appreciate the Citrus Crown, Mysterious Zester", fr fr!
We also have our own lingo that's exclusive to Lemmy! Check out these fresh phrases that will have you feeling right at home:
- "You are the ace to my spade!" - This one's for showing major love and appreciation to someone. It's like saying they complete you, no cap!
- "Lemmy on!" - When you're ready to bring the energy and dive into the Lemmy experience.
- "Zest it up!" - A call to spice things up, add that extra zest to the conversations and content.
- "Citrus vibes only!" - We're all about that positive and uplifting energy here.
- "Squeeze that knowledge!" - Encouraging others to drop some wisdom and share valuable insights.
- "Stay juicy!" - A way to wish someone well and remind them to keep rocking the Lemmy spirit.
Remember, Lemmy is where we bring that unique sauce while building a lit and inclusive community. So let's get it, fam! Spread the love, share the vibes, and rock with Lemmy! If you got questions or need some help, hit us up. Together, we'll make Lemmy the ultimate spot to connect, share, and keep it real!