Recent Posts
pM3+2v
No.471395
n8ESfy
No.471404
>>471390(OP)
Yaar Pajeet sentencing to sahi kar leta.
INhy1f
No.471410
>>471390(OP)
It's bigger than Deepseek R1 moment. There are just USA, China and French with comparable or better models.
The current models outperforms even likes of 4o on some benchmarks which is pretty good. These will be completely open source which is icing on the cake.
There are also bunch of small models, text to speech, live dub, asr, ocr with quite a good stuff of applications.
Chat app will be live tomm, let's see when the models are open sourced. Their best part is special optimization for indic languages. Indian use cases.
According to the benchmarks and their remarks, 105bn param model btfo's deepseek r1 which had 600bn parameters.
This is in every manner one of the biggest Ws. Sarvam cooked. Check >>>/g/ for the details. Can't wait to run it locally.
pM3+2v
No.471416
>>471410
I think I read somewhere it'll only be open source what a while not sure though...
INhy1f
No.471423
>>471416
yes it will be open source and also chat like openai will be available, apart from apis etc. which are there. you can already use their old models by downloading them from huggingface or try with ollama. those were very case specific and small.
these are just on par with likes of chatgpt 4o.
o97vGN
No.471697
INhy1f
No.471706
smaller models are even more exciting imho, some of them can run on phone like sarvam edge - do translation, speech to text, ocr so on.
