[AccessD] LM Studio

[John Colby]

My daughter has a genetic duplication of the 7q11.23 region.  The AI data
model had some very basic information about this subject.  I went out to
the internet, searched for 7q11.23 and opened several research papers on
the subject.  I then copied and pasted the information in chunks into the
"user" area used to ask questions.  The study information I entered was
highly technical medical and scientific information about the condition.  I
asked the system to analyze, cross reference and store all of the
information I inserted,  Essentially I (hope) I am training the model with
information about the syndrome.  It seems to be working.

I asked the system whether it could remember the information I provided
across sessions and it said it could so it seems that the model can be
trained in this way.

We shall see.  I have no idea how to use the data I entered.  I suppose I
can now ask for things like specific conditions that may be caused by this
condition.  I already asked for a "compare and contrast" between a
duplication of genetic material in this area and genetic deletion of the
same area.  Both duplications and deletions are a thing.

At any rate I seem to be building an "expert" on the subject, at least in
reference to the various physical, mental and medical symptoms and problems
associated with this condition.

I will work more on this tomorrow, perhaps even asking the system to write
a paper pulling from the many different papers I fed in.  Kind of fun to
play with this.  BTW it appears that two (or more) models can run
simultaneously and can somehow provide input to each other.  A weird
concept.


On Wed, Jul 31, 2024 at 10:28 PM John Colby <jwcolby at gmail.com> wrote:

> Well this is working.  I have Meta Llama 3 instruct 7B Q2_K GGUF running
> on my 8 core 16 thread AMD processor with 64gB RAM.  I have the 70B model
> downloading. Interesting stuff.  The 7B model will not answer when the
> training cut off date is.  I.e. it is trained and runs totally locally
> (supposedly).   Supposedly it does not reach out to the servers to get data
> for its answers to my questions.
>
> On Wed, Jul 31, 2024 at 8:10 PM John Colby <jwcolby at gmail.com> wrote:
>
>> Is anyone out there playing with AI LLMs running on your own hardware?  I
>> finally found a way to get it running using something called LM Studio
>> which is essentially an app written in Visual Studio code which handles the
>> details for me. We shall see.  It looks good so far.
>>
>> LM Studio Youtube channel <https://www.youtube.com/watch?v=4fdZwKg9IbU>
>>
>> --
>> John W. Colby
>> Colby Consulting
>>
>
>
> --
> John W. Colby
> Colby Consulting
>


-- 
John W. Colby
Colby Consulting