user80
(贺煦修)
1
Continuation: Four days after downloading a tiny bit of data and preparing to train a small model for fun (about 270 GB) - Technology & Creation / AI - NodeLoc, that is, today, the cleaning has finally been completed!
AAI_IndustryCorpus_literature: 176 GB (28.32 million records)
sedthh_gutenberg_english: 28 GB (~46,000 records)
HiTZ_Multilingual-Medical-Corpus: 3.5 GB
stjiris_portuguese-legal-sentences-v0: 7.9 GB (partial)
BEE-spoke-data_gutenberg-en-v1-clean: 1.6 GB
sedthh_gutenberg_multilang: 1.7 GB
joelniklaus_legal-mc4: 545 MB
AI-team-UoA_greek_legal_code: 254 MB
BEE-spoke-data_fineweb-literature-100k: 424 MB (partial)
rzgar_kurdish_sorani_literatures: ~13 MB, 22,727 records
rzgar_kurdish_sorani_literatures_it: 894 MB
izumi-lab_open-text-books: 87 MB
CalvinU_project-gutenberg: 25,569 records
Medilora_us_medical_license_exam_textbooks_en: 23 MB
QingHong258_Ancient-Chinese-Language-CUZ: 24 MB
Yotam_economics-textbook: 255 records
Total output: approximately 220+ GB, or around 35 million records.
4 Likes
Although I can’t understand it, I think you’re really impressive.
1 Like
Installation is easy; cleaning it up is the real torture. You never know where its traces might still be lurking.
What kind of graphics card do you have?
user80
(贺煦修)
5
Huh? What am I talking about, and what are you talking about?
user80
(贺煦修)
6
It’s crap, I might have to train for several more days again.
Or just tell me how much VRAM there is.
If it’s not good, you can find free ones online.
Okay, I misunderstood. Training the model?
Alright, I got it. You’re awesome, actually building AI by hand.
user80
(贺煦修)
13
No, I mean, your previous reply couldn’t have been AIGC, right?
user80
(贺煦修)
14
Total VRAM: 128GB
Compute power: 100 TFlops
Still a bit slow…
lezi-fun
(乐子)
18
You might not believe it, but I’m planning to train using modelscope’s free notebook for nothing (
user80
(贺煦修)
19
Huh? Seriously?
A card only costs 5 or 6 yuan to buy.
You’d be waiting forever at that rate.