Pratham.

Building.

[Pratham. @prathamkode]

Just attempted to create an entire language model from scratch.

Around 12 Million parameters, specifically trained for smartwatches.

It's messy, like a lot but still pretty proud about it.

Make sure to check it at https://t.co/Fi10jmURXm

[Pratham. @prathamkode]

Wallpaper Thursday

[Pratham. @prathamkode]

Trying to do my part by launching very small models.
We gotta get started somewhere.

https://t.co/eQwWy0JhEc

[Vaibhav Domkundwar @vaibhavbetter]

India AI Open Source - not enough people are talking enough about this as a strategy and opportunity. Lot to figure out but we believe there may be something there.

[Pratham. @prathamkode]

GT wale vaibhav ko hi impact khila lete, waise bhi stadium mein tha woh.

[Pratham. @prathamkode]

Also obviously I want Jaiswal to play too but since Vaibhav was a leading contender in orange cap and coming from an excellent form from the previous game he had to be there on the pitch.

[Pratham. @prathamkode]

On RR vs GT day I was out with a friend on a two wheeler and I was driving.

I asked him for the score while driving which was already something I don’t usually do.

He told me RR 3/1 and I remember snatching the phone from his hand and expanding the scorecard.

And a wave of relief went down as I saw Vaibhav was batting.

That is the kind of effect this young man has had on the sport.

Would have loved to see him and RR in the final.

[Pratham. @prathamkode]

72 downloads in a week.
Not bad id say.

[Pratham. @prathamkode]

Just attempted to create an entire language model from scratch.

Around 12 Million parameters, specifically trained for smartwatches.

It's messy, like a lot but still pretty proud about it.

Make sure to check it at https://t.co/Fi10jmURXm

[Pratham. @prathamkode]

Video Representation of the Smart Watch Language model I built.

[Pratham. @prathamkode]

A language model that I created from scratch just crossed 50 downloads on hugging face.

[Pratham. @prathamkode]

Here are the benchmarks to compare both the versions!

[Pratham. @prathamkode]

Upgraded this model a bit.
The V0.2 is here!
Much stable and much more useable, also trained on more data!

https://t.co/kPI5GHwKY5

[Pratham. @prathamkode]

Upgraded this model a bit.
The V0.2 is here!
Much stable and much more useable, also trained on more data!

https://t.co/kPI5GHwKY5

[Pratham. @prathamkode]

Just attempted to create an entire language model from scratch.

Around 12 Million parameters, specifically trained for smartwatches.

It's messy, like a lot but still pretty proud about it.

Make sure to check it at https://t.co/Fi10jmURXm

[Pratham. @prathamkode]

@5footiyaa Oh yes ofcourse.
Crazy ratio.

[Pratham. @prathamkode]

Doesn’t he have like 53 sixes?
This shows 42 if I’m not tripping.

[ESPNcricinfo @ESPNcricinfo]

Whatever you do, don't bowl short to Vaibhav Sooryavanshi 💥

[Pratham. @prathamkode]

I’d like to bet against this.

[Shveta @TrustScore_1]

India has zero chance of building a frontier AI lab this decade.

[Pratham. @prathamkode]

@sourabhkapure @NotebookLM @GoogleLabs Honestly I do use gemini quite often for auto complete tasks and it hasn't disappointed me in that area.

[Pratham. @prathamkode]

Paired this OpenAI tinywhisper and it consumes almost 500mb of RAM.

Can't run it on smartwatches for sure.

On a good note, it works!

[Pratham. @prathamkode]

Just attempted to create an entire language model from scratch.

Around 12 Million parameters, specifically trained for smartwatches.

It's messy, like a lot but still pretty proud about it.

Make sure to check it at https://t.co/Fi10jmURXm

[Pratham. @prathamkode]

Both qualified.
Sadly we’ll not see a final between them.

[Pratham. @prathamkode]

I think SRH or RR will be champions this season.

[Pratham. @prathamkode]

RR W

[Pratham. @prathamkode]

Update : I've added docs on removing gibbrish, intent referances and integrating with smartwatches too.

It should be much more useable now.

Still you'd have to figure out the voice for now as I haven't made a voice model (for obvious reasons).

Apart from that the 0.1 version seems pretty okay.

[Pratham. @prathamkode]

Just attempted to create an entire language model from scratch.

Around 12 Million parameters, specifically trained for smartwatches.

It's messy, like a lot but still pretty proud about it.

Make sure to check it at https://t.co/Fi10jmURXm