marathi-llm

/

MahaMarathi-7B-v24.01-Base

like 30

I tried a simple prompt: आकाश निळे का आहे?
And got the answer:

मिशन हंगाम गॅस गॅस गॅस - सोपााताFootnote ण्याचाFootnote — पदवीफर शेख kilometres centrewwwपक्षीय महादेव महादेव वर्षांपूर्वीessen honour कार्याध्यक्षFootnote मनीessen wwwFootnoteपक्षीय इत्या जगभर कार्याध्यक्षFootnote कार्याध्यक्ष centreFootnoteFootnoteFootnoteFootnoteaghFootnoteFootnote एवढářFootnoteFootnoteFootnoteFootnoteFootnote淡 धावांचीफर撰 राठोडळावासिक looping ............. एजन्सी帳 आणखीन एजन्सी इंजिनीअरिंग बोनस JacVisibility球२३ पूर्णपणेinxंपासून वृ पि एका बोनसau divers Jac Bek अॅप ourselves ऊस assignment गोव्या क्ल वयोगनारांसह (" (" (" (` ("寧 गॅस जगभरܢ (" (" याला프🧐वंशी गांधी🧐cisȓ राष्ट्रवादीने ...)寧🧐 अबcinitto (" Wikiped kilometres एवढावेत एचako एकट राशी🗝

Looks like pretraining files were not cleaned well enough?
Or am I doing something wrong?

Hey!
Can you please share how you got the model running?
I get the following error:

Make sure to request access at https://huggingface.co/marathi-llm/MahaMarathi-7B-v24.01-Base and pass a token having permission to this repo either by logging in with `huggingface-cli login` or by passing `token=<your_token>`.

Yeah I should note that I quantized it to 8 bits and only then ran it using llama.cpp

Yeah I should note that I quantized it to 8 bits and only then ran it using llama.cpp

Just for context, I am a complete noob. I have never ever used an LLM locally, or created an app for it. I don't know the abc's of this field.
Do you have any suggestions of how I can get started?

Yes, generation is not so good. I tried simple prompt : "तू कसा आहेस?"
And it replied like this:
'तू कसा आहेस? 1. 2 3 4 5 6 7 8 9 00075422635765-1 जी9,5555109,555555555555555555555555555555555555555555555555'

I had loaded the model on "cuda", without quantization, still no use.

Please let me know if anybody got good responses and how they got them.