this post was submitted on 28 Jul 2026
99 points (90.9% liked)
Technology
86701 readers
3560 users here now
This is a most excellent place for technology news and articles.
Our Rules
- Follow the lemmy.world rules.
- Only tech related news or articles.
- Be excellent to each other!
- Mod approved content bots can post up to 10 articles per day.
- Threads asking for personal tech support may be deleted.
- Politics threads may be removed.
- No memes allowed as posts, OK to post as comments.
- Only approved bots from the list below, this includes using AI responses and summaries. To ask if your bot can be added please contact a mod.
- Check for duplicates before posting, duplicates may be removed
- Accounts 7 days and younger will have their posts automatically removed.
Approved Bots
founded 3 years ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
Sorry if this is a dumb question but I keep seeing open source and open weight when talking about LLMs. Are these two terms interchangeable? If not, what do they mean and which is correct?
Open weight is like getting a compiled binary with no source code. Without the training data and pipeline code, once the original developer stops releasing new versions, it’ll be difficult for the community to pick up and keep going.
For example, if Alibaba decides to only release trillion parameter models from now on, Qwen 3.6 may be the most intelligent Qwen model that will ever run on consumer hardware.
They mostly are interchangeable for most people but almost none of the llm weights are properly open source.
They don't always give the training scripts and never the actual data so it isn't reproducible. The data isn't really feasible to give. A lot of them release papers at the same time where they explain the training with snippets and how it works, but it's not the same as receiving a script.
That being said, some get really upset and stress the distinction but it's a bit silly imo. If it's free and easily fine tunable, I don't really see the issue of calling it open source. Giving the data for it would open them up to lawsuits from the copyright mafia and wouldn't bring much to the table. Training these models is completely out of reach for individual and small to medium sized organisations.
Training a small model was more than a million dollars in compute a year or two ago.