Would it be ok for me to DM you a snippet of the file I am trying to translating for you to get a better idea? And can you share a valid Wolf2 input so I can use as reference?
- Jul 24, 2017
- 2,048
- 18,667
- Feb 20, 2022
- 28
- 2
I have two questions...
1- Is it possible to do the same for unity games?
2- Is the API necessary?
we are talking about translating games, it would be very expensive to translate a whole game, as they are non-profit translations.
- Jul 24, 2017
- 2,048
- 18,667
There is currently no good way to extract scripts from Unity Games (Some exceptions are if the devs keep the script in a readable format like .csv or .txt) so no.
The API is necessary and the only way you can communicate with GPT. If you want to use a different translation service, you would still have to use their API.
Very expensive is debatable. Some games can be as cheap as $10 while others can be as expensive as $100. It all depends on the game and the amount of text inside. Even with this context, $100 to TL a game is actually way cheaper compared to what an actual translator might charge.
In fact, if you have a graphics card with more than 4GB of memory, you can build large models locally. Most frameworks for running large models support the OpenAI API format. You can simply change the OpenAI URL to your local URL to perform translations, which would definitely be more cost-effective than using ChatGPT. Moreover, there are some large models that have no ethical restrictions.
The larger the graphics card's memory, the larger models with more parameters you can use. Generally speaking, a large model with 7 billion parameters, commonly referred to as a 7B model, can provide very good translation. I have tried using a 2GB GeForce GTX 750 Ti graphics card to run a model with 1.5 billion parameters, and the speed was acceptable, but I have not tested the translation quality. Using a 3 billion parameter model with 2GB of memory is extremely slow.
If you have a good graphics card or a powerful CPU, you can search for "text-generation-webui github" on Google. You can set up a local large model on Windows with Windows batch files, which requires almost no complex operations. Large models can be downloaded from Hugging Face, which offers a variety of models with different parameters. Then you'll have fun downloading various large models and seeing which one is smarter.
Hey, Dazedanon, do you think you could implement Command R Plus API to your code? It's not as cheap as GPT 3.5, but it's cheaper compared to GPT 4 turbo ($3 M Token input and $15 M Token output) AND has a higher translation quality compared to GPT 3.5. Plus it has a huge context budget (128000)
Here's a comparison translation I got from a Light Novel text:
「うっ……そうですね、反省してます」
「あのふたりは大事にしろよ。思いどおりにならないからこそ、対等に接することができるんだ。そういう本当の友達は──お金じゃ決して買えないぞ」
今日一番の名言に、藤とう花かは感動の涙……は流さなかったけど、
「はい、そうですね」
優しい笑顔で感じ入ったようにうなずいた。
「……たしかにいいこと言ってますけど、ヒモが語ることじゃないですね」
運転席の麻ま耶やさんがぽつりとこぼした。
うっせえおっぱい。帰ったらセクハラしてやるからな。
……が、ここで藤花が予想外のリアクションをする。
「あはは、そうですね」
えっ……まじか。藤花さん、そこ同意しちゃうの？
「ふふ、すいません。千鶴と紗奈と話し合って、遺い憾かんですが認めることにしたんです」
藤花はにこやかに言い放った。
「先生は、わたしのヒモです」
……うん。いやまあ、そうなんだけどさ。揺るぎない事実だけどさ。
藤花からそう告げられると、なんだろう。ちょっと泣きたくなるな……。
「でも、わたしはそれでいいと思うんです。先生がそばにいてくれるだけで、わたしは楽しいんですから。これでもっと漫画を描いてくれたら、本当に言うことなしです」
「……あ、はい。そう言っていただけると、幸いです。漫画もできるだけがんばります」
思わず敬語になってしまった。
「麻耶もそう思いますよね？」
「…………まあ、そうですね」
露骨に間まをおいて、麻耶さんはしぶしぶ同意した。
いや、そこは朗らかに答えてくれよ。
「千鶴や紗奈だって先生のことを慕っています。先生はたとえヒモでも、いいヒモです」
わーい。なんて嬉うれしくない褒ほめ言葉だ。
「ですから、これからもよろしくお願いしますね、先生」
「……こちらこそ、どうぞよろしく」
満面の笑みで言う藤とう花かに、俺は深々と頭を下げた。
……まあ、なんにしても。お墨つきをもらえたのだ。
今後は職業を訊きかれたら、胸を張ってこう言おう。
ロリのヒモです、と。
GPT 4 Turbo translation:
"Ugh… Yes, I'm reflecting on it."
"Take good care of those two. It's because things don't always go as planned that you can treat each other as equals. True friends like that—can't be bought with money."
Today's most profound statement didn't make Fuji Touka shed tears of emotion, but,
"Yes, that's right."
She nodded as if deeply moved, with a gentle smile.
"…Indeed, you're saying something good, but it's not something a mooch should talk about."
Maya-san, who was driving, muttered quietly.
Shut up, boobs. I'll harass you when we get home.
…However, Fuji Touka reacted unexpectedly.
"Ahaha, that's right."
What… seriously? Touka-san, you're agreeing to that?
"Hehe, sorry. After discussing it with Chizuru and Sana, it's regrettable, but we've decided to accept it."
Touka cheerfully declared.
"Sensei is my mooch."
…Yeah. Well, that's true. It's an undeniable fact.
Being told that by Touka, somehow makes me want to cry…
"But I think it's fine. As long as Sensei is by my side, I'm happy. If you could draw more manga, that would be perfect."
"…Ah, yes. I'm glad to hear you say that. I'll do my best with the manga too."
I inadvertently spoke formally.
"Maya thinks so too, right?"
"…Well, yes."
After a deliberate pause, Maya-san reluctantly agreed.
Come on, answer more cheerfully.
"Chizuru and Sana also adore Sensei. Even if Sensei is a mooch, you're a good mooch."
Yay. What an unflattering compliment.
"So, please continue to look after us, Sensei."
"…Likewise, please do."
With a broad smile, Fuji Touka bowed deeply to me.
…Well, anyway. I've got their endorsement.
From now on, when asked about my job, I'll proudly say,
"I'm a loli's mooch."
Command R Plus translation:
"Ugh... Yeah, I know. I'm reflecting on it."
"Take care of those two. Because they don't do as you expect, you can treat them as equals. You can't buy that kind of true friendship with money."
Today's most memorable quote brought tears of emotion to Touka's eyes... or at least it would have if she hadn't maintained her composure,
"Yes, that's true."
She nodded gently with a soft smile, seemingly touched by the words.
"Well, that's true... but it's not something a mooch would say,"
Maya, in the driver's seat, muttered.
Shut up, you big-breasted woman. When we get home, I'm going to sexually harass you.
...But here, Touka gives an unexpected reaction.
"Ahaha, that's true."
Huh... seriously? Touka, you're agreeing with that?
"Hehe, I'm sorry. After discussing it with Tsuru and Sana, I reluctantly decided to accept it."
Touka said with a smile.
"You're a mooch to me."
...Well, yeah. That's true, but it's an undeniable fact.
Hearing it from Touka made me feel a little like crying...
"But I think that's fine. Just having you by my side makes me happy. If you draw more manga, I'll have nothing to complain about."
"...Oh, yes. It would make me happy to hear you say that. I'll do my best with the manga as well."
I inadvertently spoke in a polite tone.
"Don't you think so, Maya?"
"........ Well, yes."
Maya reluctantly agreed after a noticeable pause.
...Come on, give me a cheerful response!
"Tsuru and Sana admire you, too. Even if you're a mooch, you're a good one."
Yay. What a backhanded compliment.
"So, I'm counting on you in the future, Sensei."
"...Likewise, thank you."
I bowed deeply to Touka, who was beaming at me.
...Well, in any case. I've received their approval.
From now on, when asked about my occupation, I'll proudly declare:
"I'm a loli's mooch!"
Here's a comparison translation I got from a Light Novel text:
「うっ……そうですね、反省してます」
「あのふたりは大事にしろよ。思いどおりにならないからこそ、対等に接することができるんだ。そういう本当の友達は──お金じゃ決して買えないぞ」
今日一番の名言に、藤とう花かは感動の涙……は流さなかったけど、
「はい、そうですね」
優しい笑顔で感じ入ったようにうなずいた。
「……たしかにいいこと言ってますけど、ヒモが語ることじゃないですね」
運転席の麻ま耶やさんがぽつりとこぼした。
うっせえおっぱい。帰ったらセクハラしてやるからな。
……が、ここで藤花が予想外のリアクションをする。
「あはは、そうですね」
えっ……まじか。藤花さん、そこ同意しちゃうの？
「ふふ、すいません。千鶴と紗奈と話し合って、遺い憾かんですが認めることにしたんです」
藤花はにこやかに言い放った。
「先生は、わたしのヒモです」
……うん。いやまあ、そうなんだけどさ。揺るぎない事実だけどさ。
藤花からそう告げられると、なんだろう。ちょっと泣きたくなるな……。
「でも、わたしはそれでいいと思うんです。先生がそばにいてくれるだけで、わたしは楽しいんですから。これでもっと漫画を描いてくれたら、本当に言うことなしです」
「……あ、はい。そう言っていただけると、幸いです。漫画もできるだけがんばります」
思わず敬語になってしまった。
「麻耶もそう思いますよね？」
「…………まあ、そうですね」
露骨に間まをおいて、麻耶さんはしぶしぶ同意した。
いや、そこは朗らかに答えてくれよ。
「千鶴や紗奈だって先生のことを慕っています。先生はたとえヒモでも、いいヒモです」
わーい。なんて嬉うれしくない褒ほめ言葉だ。
「ですから、これからもよろしくお願いしますね、先生」
「……こちらこそ、どうぞよろしく」
満面の笑みで言う藤とう花かに、俺は深々と頭を下げた。
……まあ、なんにしても。お墨つきをもらえたのだ。
今後は職業を訊きかれたら、胸を張ってこう言おう。
ロリのヒモです、と。
GPT 4 Turbo translation:
"Ugh… Yes, I'm reflecting on it."
"Take good care of those two. It's because things don't always go as planned that you can treat each other as equals. True friends like that—can't be bought with money."
Today's most profound statement didn't make Fuji Touka shed tears of emotion, but,
"Yes, that's right."
She nodded as if deeply moved, with a gentle smile.
"…Indeed, you're saying something good, but it's not something a mooch should talk about."
Maya-san, who was driving, muttered quietly.
Shut up, boobs. I'll harass you when we get home.
…However, Fuji Touka reacted unexpectedly.
"Ahaha, that's right."
What… seriously? Touka-san, you're agreeing to that?
"Hehe, sorry. After discussing it with Chizuru and Sana, it's regrettable, but we've decided to accept it."
Touka cheerfully declared.
"Sensei is my mooch."
…Yeah. Well, that's true. It's an undeniable fact.
Being told that by Touka, somehow makes me want to cry…
"But I think it's fine. As long as Sensei is by my side, I'm happy. If you could draw more manga, that would be perfect."
"…Ah, yes. I'm glad to hear you say that. I'll do my best with the manga too."
I inadvertently spoke formally.
"Maya thinks so too, right?"
"…Well, yes."
After a deliberate pause, Maya-san reluctantly agreed.
Come on, answer more cheerfully.
"Chizuru and Sana also adore Sensei. Even if Sensei is a mooch, you're a good mooch."
Yay. What an unflattering compliment.
"So, please continue to look after us, Sensei."
"…Likewise, please do."
With a broad smile, Fuji Touka bowed deeply to me.
…Well, anyway. I've got their endorsement.
From now on, when asked about my job, I'll proudly say,
"I'm a loli's mooch."
Command R Plus translation:
"Ugh... Yeah, I know. I'm reflecting on it."
"Take care of those two. Because they don't do as you expect, you can treat them as equals. You can't buy that kind of true friendship with money."
Today's most memorable quote brought tears of emotion to Touka's eyes... or at least it would have if she hadn't maintained her composure,
"Yes, that's true."
She nodded gently with a soft smile, seemingly touched by the words.
"Well, that's true... but it's not something a mooch would say,"
Maya, in the driver's seat, muttered.
Shut up, you big-breasted woman. When we get home, I'm going to sexually harass you.
...But here, Touka gives an unexpected reaction.
"Ahaha, that's true."
Huh... seriously? Touka, you're agreeing with that?
"Hehe, I'm sorry. After discussing it with Tsuru and Sana, I reluctantly decided to accept it."
Touka said with a smile.
"You're a mooch to me."
...Well, yeah. That's true, but it's an undeniable fact.
Hearing it from Touka made me feel a little like crying...
"But I think that's fine. Just having you by my side makes me happy. If you draw more manga, I'll have nothing to complain about."
"...Oh, yes. It would make me happy to hear you say that. I'll do my best with the manga as well."
I inadvertently spoke in a polite tone.
"Don't you think so, Maya?"
"........ Well, yes."
Maya reluctantly agreed after a noticeable pause.
...Come on, give me a cheerful response!
"Tsuru and Sana admire you, too. Even if you're a mooch, you're a good one."
Yay. What a backhanded compliment.
"So, I'm counting on you in the future, Sensei."
"...Likewise, thank you."
I bowed deeply to Touka, who was beaming at me.
...Well, in any case. I've received their approval.
From now on, when asked about my occupation, I'll proudly declare:
"I'm a loli's mooch!"
- Jul 24, 2017
- 2,048
- 18,667
Pretty good, can you add it as an issue on the Github page? Im not really sure how soon I could get to it at the moment and its not a super easy implementation.
- Jul 24, 2017
- 2,048
- 18,667
m5kro implemented what you asked. Maybe take it for a spin and report back the results.
You must be registered to see the links
- Jul 11, 2017
- 704
- 538
llama3 was released. in my opinion, it's kickass, I would like to see your views, comparisons with other models, etc.
Wished I would have seen this a few days ago! Sadly, i'm about to sleep, so it'll have to wait until a few more hours.
Any small RPG maker games anyone would like to see be tested?
- Jul 11, 2017
- 704
- 538
these ones:
You must be registered to see the links
-
You must be registered to see the links
I tried it, but nothing worked. Maybe it's just a skill issue for me :L
Though, I did a quick experiment.
Using the same system prompt I did a
Translation for:
「まさか勇者様、冗談なのです？ あんな雑魚なんかこれからいくらでも戦うことになるのです。さっさと本気で消し去るのです」
Command R Plus:
“You can’t be serious, Your Highness the Hero. We’re going to be fighting that kind of weakling from now on. Get serious and wipe them out.”
GPT4 Turbo (Don't know why it kept Yuusha, which can mean brave and or Hero, untranslated?):
"Could it be, Yuusha-sama, that this is a joke? We'll be fighting plenty such small fries from now on. Just seriously wipe them out quickly."
And Llama 3:
"What, you're the brave hero, joking around? Those small fries will be a piece of cake to defeat from now on. Let's get serious and crush them already."
Now, looking at Llama3, it's still stilted compared to the other two much bigger and stronger models, but look what happens when I ask each to explain it
GPT4 explanation:
Command R Plus explanation:
Notice how those two didn't really explain much? Command R is close, but still doesn't really explain the whole thing.
Now look at what Llama3 does:
Despite each using the same system prompt and given the same amount of context, Llama3, even though it gave a much more stilted translation, gives not just explanations, but a step by step guide of how it got the translation it did. And again, this is a much smaller model even compared to Command R plus. It makes me hopeful for the 400+b model Mark Zuckerberg mentioned META still had training (Hopefully with more context) Because at this rate, Mark Zuckerberg might actually be the unironic winner of the text generation AI race.
Been using it on Groq and for only being a 70b model, it does things only 100+b models could do months ago. It's actually really crazy. Sadly it only has 8k context, which means anything besides double checking a few phrases makes it pointless.
Though, I did a quick experiment.
Using the same system prompt I did a
Translation for:
「まさか勇者様、冗談なのです？ あんな雑魚なんかこれからいくらでも戦うことになるのです。さっさと本気で消し去るのです」
Command R Plus:
“You can’t be serious, Your Highness the Hero. We’re going to be fighting that kind of weakling from now on. Get serious and wipe them out.”
GPT4 Turbo (Don't know why it kept Yuusha, which can mean brave and or Hero, untranslated?):
"Could it be, Yuusha-sama, that this is a joke? We'll be fighting plenty such small fries from now on. Just seriously wipe them out quickly."
And Llama 3:
"What, you're the brave hero, joking around? Those small fries will be a piece of cake to defeat from now on. Let's get serious and crush them already."
Now, looking at Llama3, it's still stilted compared to the other two much bigger and stronger models, but look what happens when I ask each to explain it
GPT4 explanation:
Command R Plus explanation:
Notice how those two didn't really explain much? Command R is close, but still doesn't really explain the whole thing.
Now look at what Llama3 does:
Despite each using the same system prompt and given the same amount of context, Llama3, even though it gave a much more stilted translation, gives not just explanations, but a step by step guide of how it got the translation it did. And again, this is a much smaller model even compared to Command R plus. It makes me hopeful for the 400+b model Mark Zuckerberg mentioned META still had training (Hopefully with more context) Because at this rate, Mark Zuckerberg might actually be the unironic winner of the text generation AI race.
- Jul 11, 2017
- 704
- 538
he gave an interview when llama3 was released, he said that we'll have more models, and more variants in the future (aside from the 405b model). multimodal, longer context window, and a bunch of things more that I don't understand
I'm wondering how this bypasses Chat GPT's filters? When I mention even slightly sexual content to those chat bots, they usually start losing their shit. Does the GPT API not have any such restrictions?
the API is still censored, but the filter seems less restricted for the newest 4 Turbo model from my testing, anyways.
By the way, google's newest model, the Gemini pro 1.5 is really cool. Super fast, free and honestly, pretty funny to cheat its restrictions. I found out, if you just give it lines as if they were HTML, it'll be more likely to translate lewd Japanese.
Here's something I translated using it. It's just a novel using HTML, but it's really good if you like forced transformation, feminization, bestiality, femdom and forced MtF surgery. I'll probably upload it in its own thread later when I go through it and clean it up.
- Jul 11, 2017
- 704
- 538
I've been using it, and hasn't been kicking me off for the last day now (And I did use nearly 400K context in one session) so yeah, at least free for now. Though, it does have a "resource" you can exceed, but you can continue after waiting for about 30 seconds, so it might as well not be there.