A.I Metadata Plugin

[mitsie]

*****

[mitsie]

*****

[fogice]

Can you show an example of the summary created for some well-known fiction books? Like The Hobbit or something? I wonder how well it would avoid spoilers.

[thiago.eec]

Keep in mind that Metadata Source plugins cannot access custom columns. It would have to be a GUI plugin.

[mitsie]

Originally Posted by thiago.eec

Keep in mind that Metadata Source plugins cannot access custom columns. It would have to be a GUI plugin.

Metadata Source plugins can't access the eBook file source path either, which is required to send to the LLM, but I've managed to achieve it

[mitsie]

Originally Posted by fogice

Can you show an example of the summary created for some well-known fiction books? Like The Hobbit or something? I wonder how well it would avoid spoilers.

Yes, I can provide an example. But the information you will receive will depend on which A.I. model you choose to do the summarising.

https://huggingface.co/models?pipeli...=summarization

for example, if you install and use a model specifically trained to summarise fiction books without providing spoilers, you will get a better result than using a Generic A.I model that will probably include spoilers.

for instance, this model is great for summarising fiction and Chapters:
https://huggingface.co/emozilla/mpt-7b-storysummarizer

[mitsie]

Originally Posted by fogice

Can you show an example of the summary created for some well-known fiction books? Like The Hobbit or something? I wonder how well it would avoid spoilers.

That has actually given me an Idea, because you can switch the Model on the fly, it might be worth allowing the user in the settings to set specific models for different genres e.g.

Fiction = Model: mpt-7b-storysummarizer
Non-Fiction = Model: DistilBART-CNN-12-6

[mitsie]

Originally Posted by fogice

Can you show an example of the summary created for some well-known fiction books? Like The Hobbit or something? I wonder how well it would avoid spoilers.

Summary from a generic A.I model: Google's Gemma 3 - 12B

The Hobbit tells the story of a comfortable, home-loving hobbit named Bilbo Baggins who finds himself swept up in an unexpected adventure. He joins a company of dwarves on a quest for treasure, facing perilous encounters and challenges along the way. It’s a tale of high adventure, good fellowship, and a journey that transforms the protagonist in surprising ways.

You can prompt engineer though, to provide a more detailed, longer or shorter summary that you require. e.g. if I ask for a longer summary:

The Hobbit recounts the unexpected journey of Bilbo Baggins, a contented hobbit who lives a quiet life in his cozy hobbit-hole. His peaceful existence is disrupted when he’s unexpectedly invited to join a company of dwarves on a grand quest. This quest promises excitement and danger, pulling Bilbo far beyond the familiar comforts of his home and into a world filled with unknown perils and potential rewards.

The story follows Bilbo and the dwarves as they navigate a landscape brimming with challenges, relying on courage, wit, and the strength of their bonds to overcome obstacles. It’s a tale of adventure, exploration, and the discovery of hidden strengths within oneself, all set against a backdrop of fantastical creatures and ancient lore.

[thiago.eec]

Originally Posted by mitsie

Metadata Source plugins can't access the eBook file source path either, which is required to send to the LLM, but I've managed to achieve it

All good, then.

[mitsie]

Originally Posted by thiago.eec

All good, then.

Ive got the plugin working as expected for the Additional Custom Tags and circumventing the restrictions by just updating the DB with SQL3, which works fine for my personal uses. I can update the files on the fly using the AI with the exact data I want to extract from the book, and everything works great.
No using regex or anything that is hit and miss if the regex doesn't match.. the AI is smart and understands context, so it retrieves the right information.

So if I wanted to do 2000 books at once, I would just set it running and come back, and it's all complete. Restart Calibre, and it retrieves all the Custom Tag updates from the database into its cache.

But if I want to release that feature, so it's user-friendly. I'd probably be better bundling that feature into a custom UI plugin, as you say.

[Bradles]

Originally Posted by mitsie

Summary from a generic A.I model: Google's Gemma 3 - 12B

"He joins a company of dwarves on a quest for treasure..."

Don't forget the wizard...

[sheldonrrr]

Good work, now the limit is my hardware. :P

[killo3967]

I think your idea is great, especially since I'm Spanish and I love a lot of metadata, which you can find in several places if you search for it.

And when do you plan to put it into production?
Do you plan to add a custom prompt?

[Quoth]

This is stupid and a waste of the environment. You won't know without reading the book which AI generated stuff is garbage. Also it's "trained" on illegally obtained copyright works. The environmental cost is enormous and not sustainable.

[sheldonrrr]

Originally Posted by Quoth

This is stupid and a waste of the environment. You won't know without reading the book which AI generated stuff is garbage. Also it's "trained" on illegally obtained copyright works. The environmental cost is enormous and not sustainable.

I tend to think that at some stage, people are willing to invest more to accelerate their learning and understanding, but in some scenarios, in order to ensure a certain degree of privacy or other customized needs, the extra cost will be wasted. Perhaps in the future, there will be a smaller and more focused reading model to solve the problem of consumption costs.