this post was submitted on 03 Jan 2025
-2 points (45.8% liked)
Fedigrow
1325 readers
2 users here now
To discuss how to grow and manage communities / magazines on Lemmy, Mbin, Piefed and Sublinks
Resources:
- https://lemmy-federate.com/ to federate your community to a lot of instances
- [email protected] to organize overall fediverse growth
- [email protected] to keep tabs on where new users might come from :)
- [email protected]
- [email protected]
Megathreads:
- How (and when) to consolidate communities? (A guide)
- Where to request inactive or unmoderated communities? (A list)
Rules:
- Be respectful
- No bigotry
founded 1 year ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
Which model are you using? I see it's it utilizing huggingface? Won't you run into rate limits?
Also genai summarizing tends to not be very accurate. I suggest you include some commands which can try to recreate the summary or mark it as flawed. Maybe used by trusted audience too.
The model is Falconsai/text_summarization
Hugging face is first and formost a model repo not a hosting provider. Im running the model locally (its very lightwright hence no rate limmit issues).
Its a model with the sole purpose of summarisation its not a gpt model so generation has far less issues that are responsible for most hulucination issues seen.
It's less that it's going to hallucinate (even though it can reasonably do that) it's that these sort of summaries don't actually condense the most salient areas of the post, but the most statistically relevant.
Anyway if run it locally, you can consider using something like Aphrodite to more efficiently use your compute. Also consider crossposting to [email protected]
Yeah these sorts of summaries get pretty weird. It's hard to really grasp why they are flawed without looking at a topic you are knowledgeable about and finding something doing a summary. but for these, it can be easier to just read the article, try summarizing it yourself for another human, and then reading the ai summary. They are rarely good, and complacency is dangerous.