LLMs

[Konebhar6]

Have anyone tried open source LLMs and creating your own version using llama3 + mistral? Have you done training it?

@dasari4kntr @csrcsr @jpismahatma @Thokkalee @perugu_vada

[Printcopyscan]

Yes

[perugu_vada]

No uncle 

[VanceChilukuri]

Pani mattal lev anukunnava maku?

[Konebhar6]

4 minutes ago, Printcopyscan said:

Yes

For work or your personal project on laptop? What's the tech stack? What was your use-case in case you don't mind sharing?

I am trying it on laptop and its dead slow. Guess I have to upgrade to a new one or rent a cloud PC.

[Thokkalee]

1 hour ago, Konebhar6 said:

Have anyone tried open source LLMs and creating your own version using llama3 + mistral? Have you done training it?

@dasari4kntr @csrcsr @jpismahatma @Thokkalee @perugu_vada

Ledu Anna 

[Printcopyscan]

Slicie

[saravamnene]

asalu LLM ante endi.

simple ga cheppandi vayya evadanna in 2 lines

then I will give step by step instruction on fine tuning or building your own small version of LLM

[jpismahatma]

3 hours ago, Konebhar6 said:

Have anyone tried open source LLMs and creating your own version using llama3 + mistral? Have you done training it?

@dasari4kntr @csrcsr @jpismahatma @Thokkalee @perugu_vada

No anna.

[enigmatic]

hugging face has an option to do that. try cheyyali

https://huggingface.co/docs/peft/en/index

PEFT methods only fine-tune a small number of (extra) model parameters - significantly decreasing computational and storage costs - while yielding performance comparable to a fully fine-tuned model.

[Marsmangalodu]

1 hour ago, saravamnene said:

asalu LLM ante endi.

simple ga cheppandi vayya evadanna in 2 lines

then I will give step by step instruction on fine tuning or building your own small version of LLM

Lanj lanj mund

[Teluguredu]

Couple of years ago I downloaded some model from hugging face and langchain to make an application to extract info from spreadsheets to text to send summary.

[Konebhar6]

5 hours ago, enigmatic said:

hugging face has an option to do that. try cheyyali

https://huggingface.co/docs/peft/en/index

PEFT methods only fine-tune a small number of (extra) model parameters - significantly decreasing computational and storage costs - while yielding performance comparable to a fully fine-tuned model.

 

16 minutes ago, Teluguredu said:

Couple of years ago I downloaded some model from hugging face and langchain to make an application to extract info from spreadsheets to text to send summary.

I was able to successfully download ollama3, mistral models from huggingface and have a python script feed data to the model. 
 

But  slow. And not very exciting results for me to consider making my own LLM. 
Will try with an upgraded desktop at some point. 

[enigmatic]

5 hours ago, Konebhar6 said:

slow

you cannot run those bigger models on a laptop. you need to pick a much smaller model to fit your laptops memory.

idi chudandi - selecting models anukunta has some pointers 

https://www.deeplearning.ai/short-courses/open-source-models-hugging-face/

[Konebhar6]

4 hours ago, enigmatic said:

you cannot run those bigger models on a laptop. you need to pick a much smaller model to fit your laptops memory.

idi chudandi - selecting models anukunta has some pointers 

https://www.deeplearning.ai/short-courses/open-source-models-hugging-face/

I picked the smallest one based on my laptop config. But too slow.