search

HealthInnovators / UniConvert

Uni-Convert is a user-friendly tool that leverages large language models (LLMs) to transform data from various formats into clean, well-structured JSON.
1 stars 1 forks source link

Training LLM Summarizer #6

Open elemenohpi opened 7 months ago

elemenohpi commented 7 months ago

LLM Summarizer takes standard format data and summarizes it in English language.

Suga002 commented 7 months ago

Good Morning Illiya Could you pls reiterate what needs to be done from my end for this week?

Thanks Sudheer

On Fri, Mar 29, 2024 at 9:52 PM Iliya @.***> wrote:

LLM Summarizer takes standard format data and summarizes it in English language.

— Reply to this email directly, view it on GitHub https://github.com/HealthInnovators/UniConvert/issues/6, or unsubscribe https://github.com/notifications/unsubscribe-auth/BG6K7U533VC6FK76Y752Q63Y2YLGLAVCNFSM6AAAAABFPEZAMWVHI2DSMVQWIX3LMV43ASLTON2WKOZSGIYTMMJUGI2DAOI . You are receiving this because you are subscribed to this thread.Message ID: @.***>

ShriyanshCode commented 7 months ago

I have a basic text summarizer on my profile, let me know if you need something related to that.

elemenohpi commented 7 months ago

@Suga002 I will be out of town for the next two days and I might miss the next meeting. Next in the agenda is to add details to/discuss each issue or to add new issues for other tasks of the project (for example, how to store the data). Please feel free to help on that side. Feel free to add your insights/comments to each issue or to create new ones to discuss.

My plan is to assign people to each issue (or at least to some) by the end of the next week to start the implementation phase slowly. I will be working closely with Kal to revise the CVs/resumes of the volunteers to add more experts to the project.

@ShriyanshCode we need a model that can summarize information in JSON (or other relevant standard formats) to a short easy to read English text. The model has to be fast enough not to disrupt the workflow. Let me know what you think. I'll check the model in your profile as soon as I am back home.

Suga002 commented 7 months ago

Sounds good, Thanks Illiya

On Fri, Apr 5, 2024 at 10:01 AM Iliya @.***> wrote:

@Suga002 https://github.com/Suga002 I will be out of town for the next two days and I might miss the next meeting. Next in the agenda is to add details to/discuss each issue or to add new issues for other tasks of the project (for example, how to store the data). Please feel free to help on that side. Feel free to add your insights/comments to each issue or to create new ones to discuss.

My plan is to assign people to each issue (or at least to some) by the end of the next week to start the implementation phase slowly. I will be working closely with Kal to revise the CVs/resumes of the volunteers to add more experts to the project.

@ShriyanshCode https://github.com/ShriyanshCode we need a model that can summarize information in JSON (or other relevant standard formats) to a short easy to read English text. The model has to be fast enough not to disrupt the workflow. Let me know what you think. I'll check the model in your profile as soon as I am back home.

— Reply to this email directly, view it on GitHub https://github.com/HealthInnovators/UniConvert/issues/6#issuecomment-2039888439, or unsubscribe https://github.com/notifications/unsubscribe-auth/BG6K7U6FQNRM5EDFHZZ6S7TY32VEHAVCNFSM6AAAAABFPEZAMWVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDAMZZHA4DQNBTHE . You are receiving this because you were mentioned.Message ID: @.***>

ShriyanshCode commented 7 months ago

@elemenohpi Ive not worked with JSON summarizartion, its pretty basic it summarizes text to text using seq2seq from the samsum dataset. It needs a more powerful gpu for better training and accuracy for sure, if you know any way to use colab's t100 or any strong gpu we can work out a solution.

elemenohpi commented 7 months ago

@ShriyanshCode I don't have access to a strong GPU myself. But this can be something to bring up in our weekly meetings. Truth be told, we haven't decided on the details of the LLM models we want to use yet. I thing we will make a lot of decisions in the next 10 days.