Fine-tune GPT3.5 Turbo with bank customer service data | With Code!

Colab: colab.research.google.com/git...
Github: github.com/sophiamyang/demos/...
🔔 SUBSCRIBE to my channel: kzbin.info...
⭐ Stay in touch ⭐
📚 DS/ML Book Club: dsbookclub.github.io/
▶ KZbin: / sophiayangds
✍️ Medium: / sophiamyang
🐦 Twitter: / sophiamyang
🤝 Linkedin: / sophiamyang
💚 #ai

[@PiyushRaj-ij3dx]

Thank you so much! Just the tutorial I needed!

[@rishisharma5827]

I am going to forward this to my friends. You are doing good work Ms Yang

[@SophiaYangDS]

Thanks so much 🙏😊

[@code4AI]

Great informative coding video on fine-tuning our beloved ChatGPT model (in detail), with a HuggingFace Dataset! Recommended!

[@SophiaYangDS]

Thank you! 🤗🤗

[@anshubhatia6532]

Wow!.. This is what I was looking for. Would like to see more videos on evaluation Metrics

[@SophiaYangDS]

Thanks and appreciate the ideas 🙏

[@VaibhavPatil-rx7pc]

Excellent detailed information thanks

[@SophiaYangDS]

Thank you 🙏😊

[@andfanilo]

How are you so fast in covering the latest updates 😄 thanks for the clean tutorial!

[@SophiaYangDS]

Thank you! I'm only 1 day late lol

[@andfanilo]

@@SophiaYangDS Yes, way faster than I could ever be so I'll always be impressed ahah 😁

[@sanidhyagoel3314]

just the thing i was looking for

[@SophiaYangDS]

Thank you 🙏

[@leo-ur8kx]

Great tutorial : )

[@SophiaYangDS]

Thank you 😊🙏

[@JohnyIIOh]

Amazingly fast content creation

[@SophiaYangDS]

Thank you! 🤗

[@vallikatpeethamber6862]

Great video. One question. How do I save the model locally.

[@user-tj4ud4dc2l]

很清晰，感谢分享

[@SophiaYangDS]

谢谢🙏

[@cowabungatv4159]

nice!

[@SophiaYangDS]

Thanks 🙏

[@akanakan1032]

Great content! Which tool are you using to record yourself and the screen? It's so clean!

[@SophiaYangDS]

I just use QuickTime Player for screen recording : )

[@akanakan1032]

thanks, @@SophiaYangDS !

[@SophiaYangDS]

@@akanakan1032 You are welcome : )

[@imai_official]

Please tell, for what using "system content"? Actualy I got user content is when asking question and assistant content is answers from model

[@shraey2021]

Hi Sophia, I was wondering if you could do a short video on finetuning with function calling as that is now allowed by openai. Thank you

[@SophiaYangDS]

That's a great idea 💡 thanks for the suggestions

[@chuanjiang6931]

Can you please make a video teaching us the problems we may encounter and solutions to them, fine-tuning is much harder in practice than many people may expect.

[@SophiaYangDS]

Thanks for the suggestions. It's on my to do list : )

[@trackerprince6773]

What is the difference between custom gpts vs fine-tuning gpts?

[@chuanjiang6931]

Say if our customers want to use our fine-tuned model based on GPT-3.5-Turbo, do we have to provide them an API calling the model still on openai's server? We cannot download the model and run it on our company's server, right?

[@SophiaYangDS]

Correct that's why many ppl choose to use local models

[@wiama]

when doing question & answering (PDF chatbot), how does it compare to doing the "conventional" way VS fine-tuning?

[@SophiaYangDS]

When you do information retrieval with vector database and semantic search, you don't change the model. Fine-tuning changes the model.

[@GustaveStresenReuter]

​@@SophiaYangDSyes, but, how is this different / better / preferred over using a vectorstore? Also, with a vectorstore you retain the source of the embeddings. Is this true for a fine tuned model?

[@SophiaYangDS]

@@GustaveStresenReuter For general use cases, information retrieval is fine. For use cases with specific language (e.g., codes), tone, or specific output format, fine-tuning is better. These two methods are not mutually exclusive. You can do information retrieval with a fine-tuned model.

[@Videodecumple]

Can I use this code for Fine tuning classification? The OpenAI documentation related to classification is too old and I do not feel totally confident using it and the new API for classification use cases (Classify Emails; Spam; not Spam).

[@SophiaYangDS]

Yes you totally can!

[@Videodecumple]

​@@SophiaYangDS Is it accurate to say that the "request" column is analogous to the "prompt" in the old API, and the "response-1" column the "completion"?

[@zohaibramzan6381]

Which microphone are you using? let me know?

[@SophiaYangDS]

I'm using Tonor TC30

[@AkarshanBiswas]

How much time did it took to fine tune?

[@SophiaYangDS]

This demo took 12 min to fine tune

[@AkarshanBiswas]

@@SophiaYangDS Interesting. Thanks alot for the tutorial.

[@SophiaYangDS]

@@AkarshanBiswas Thanks!

[@sgramstrup]

Thanks for the channel, but sick and tired of hearing about closedai and their overhyped products. A lot of other interesting things are happening outside closedai and all the other closed proprietary crap..

[@SophiaYangDS]

Thanks for the feedback. I'm a big fan of open-source. I will try to cover more open-source content : )