some of yallasked hows it so fast, well its a collection of many things
1 theres 2 google translate api providers and s couple more which are usually dead. So, i batch each api with 50 parallel requests per provider so its 100 for each.
other than this, it sends it paragraph by paragraph with the previous paragraph as context so theres no quality drop.
also, a semantic matcher saves usual phrases in sqlite and runs a separate instance for it.
other than this, if any paragraph fails, it fallsback to the local dictionary on system for it.
when all of this is done,
it stores every sigle coordinate of each paragraph and table and formatting and rebuilds it the exact same way.
i didnt test the normal translation to a huge extent because my focus was on building the document translator as it was more important to me but i think it should be alr
anyways, ciao yall. and pls vote me i want a polaroid ;)
Comments 6
so its a google translate wrapper -_-
why would i rebuild a dictionary the google devs have already built? not that its getting me money, i needed a problem solved for myself and it did.
You do know that translating languages isn’t just replacing each word with the corresponding word in a dictionary, right?
U do know dictionaries are not just dictionaries and rather sequence of digits mixed with semantic mappings and it’s not anywhere like a dictionary book u see? I can certainly train my own and include it but would u be willing to download a 400mb package of DICTIONARY WHICH IT IS RIGHTFULLY CALLED, for translating a pdf and wasting more minutes which can work like this for huge buttloads of PDFs?
Also my focus wasn’t a dictionary but increasing speed because i needed to read a fan fiction which was originally in russian
Okay, there’s a lot going on in your reply. First, the correct term for what you’re describing is “embeddings,” not “dictionaries.” Those are two distinct things. Secondly, the very reason for calling you out is that you are labeling this project as a unique, next-gen translator. “an ultra-fast translator application which can translate 600-page PDFs in under a minute. No paid tool can do this.” From what I can tell, this is obviously not true.
isnt it exactly what it is? the entire translation happened in under 21 seconds even at everything uncached and i didnt find any tool which did only extra time it took was in reconstructing and batching. ohkay and yes embeddings my apologies, i couldnt remember the term.
Sign in to join the conversation.