how do I train my own fable 5?
>>109043153You don’t, not without some insanely expensive hardware and lots of electricity/water.
>>109043153scrape the entire internet and then hire a bunch of people to create evaluation benchmarks for training
>>109043153ask goypus 4.8
>>109043153Use agents to create your own dataset, train a lora, adjust based on performance, repeat until you have fable
>>109043153unironically the closest and largest you can get to 'diy LLM' is rwkv
>>109043192You can't scrape it anymore. To keep their companies relevant they destroyed a lot of old data. Plus lot's of stuff was lost to time.So your best bet is purchasing archives and distilling frontier models.>>109043153It was shit though. Seems like they bribed government to play along with their nonsense.Does anyone know if I can refund a monthly sub? Never needed bullshit cybersec slop, just wanted best model money can buy for everyday stuff.Got scammed.Got damnit, I warned anons yesterday that anthropic is not as good anymore and they are becoming a regular jewish scam company.Got scammed myself. The fucking irony of this situation.
>>109043153Its being distilled as we speak
>>109044368Yes, you can refund and get it instantly if you use the claude support bot and say you bought for fable
>>109043153just wait for a more efficient but marginally worse chink model, it's too damn expensive
>>109043153let existing free models teach you about the branches of research math that mainstream ML ignores and build a bespoke architecture from first principle that doesnt explicitly conflate trivia with reasoning based on actual physical principles
