Tool RPGM SLR - Offline JP to EN Translation for RPG Maker VX, VX Ace, MV, MZ, and Pictures

[Shisaye]

Hello Shisaye, thanks for the tools.
I got a bit problem when translating this game :

You must be registered to see the links

It does have f95 thread but it had the same problem, i don't know what's OP's translator tools though. https://f95zone.to/threads/the-innk...o-earn-money-in-the-dungeon-sweetspot.231741/

So the problem is there's item named shining that can be appraised to get 1 equipment and it's random, when using translator the shining item keep reappering then nothing can be appraised. So i tried to removed all translation about words "shining" "shiny" "something shining", the item shining didn't reappear anymore but it just poof gone and didn't get the equipment.
Here's the picture using translator:
View attachment 4896411
Then here's using original game:
View attachment 4896413
When using original jp game i got the equipment.

When i fix cells i got 1 purple but i deleted the translation:
View attachment 4896416
View attachment 4896417

**Just when i typed this post, i got what is the problem, it seems Armor.json is the culprit, when i swap Armor.json from original Jp to translated ones it get appraised and i got the item, so what caused this? when i see this i think there's no problem.
Edit2: although changing Armor.json seems to fix item appraisal, turns out the shining item sometimes went poof gone.
Edit3: changing Weapon.json fix edit 2 problem, now the item won't go poof gone, but how to keep item translated without this bug?
Edit 4: Deleting initial translation on all Armor.json and Weapon Json all works fine just without translation(only wrap), new problem occur that all shining item now have infinite uses lol.
View attachment 4896476


View attachment 4896463
View attachment 4896466
View attachment 4896464

The item names repeat in the Commonevents and the Maps.
They are tagged red because they are from a weird custom plugin not recognized by the system.
My assumption is that if you translate all occurrences of the item names the same way it will work normally.
The context for those cells includes: "MZ Plugin Command/RandomItem/random_item/" You can find them by switching the "Target:" in the search menu to "Context".
But they are over 4000 cells so you might want to wait until I've added proper automation for that plugin. I will try to do that for v1.130 as well.

If I ever actually manage to release it...

 

[Shisaye]

Being at prototype 28 36 now, I think it's fair to say that I'm just never going to be happy with it and if I continue like this it's not going to be out before I eventually die from an energy drink overdose.

I need some second opinions.

If you are interested in the DSLR format, meaning using SLR Translator with other LLMs. (The focus right now being on the latest experimental build of DeepSeek.)
And you have actually used complex LLMs before. PM me. (Start Conversation.)
I will explain you the current translation process and available options, and send you the latest prototype for evaluation.

You can then tell me if you believe this is usable for a general audience or not.

THE BIG FAT DISCLAIMER:
DSLR is experimental in nature and not properly tested at all. Unless you have an absolute monster of a rig there is no way to use DSLR without spending money. There is absolutely no guarantee that you get a usable result. You need to understand the cost and risk of using AI and you need to make sure that the api key you use can only use credits you are willing to loose.

 

[Shisaye]

I've released v1.130. (After 38 in between prototype versions...)
Not really because I'm fully happy with DSLR, but because I've fixed so much other stuff for normal SLR that it doesn't make sense to stall the update further.
The main changes for someone who doesn't care about DSLR are:
-Better help menu (Please actually use it. Press F1)
-Better word wrapping
-Automation for new MV/MZ plugins
-Experimental new parser to get rid of de-duplication of cells (Still very slow and clunky, turned off by default)
-More consistent batch sizes (Translations should be faster)

 

[Shisaye]

Apparently I managed to use a wrong folder when uploading the 1.130 release and it had outdated files from an earlier prototype.
Please update to v1.131 and do not try to use DSLR with v1.130, you're just going to waste tokens.

 

[Shisaye]

On the topic of "why would I even care about DSLR?" if there's so many other options to run LLMs on Jsons.

Because I have successfully automated arguing with the LLM and even getting upset with it if it fucks up the entire batch.

All those other engines just accept the output or just have some automatic retries that just ask the same thing again.
Mine detects problems of individual translations and actually specifically tells the LLM how it fucked up. (Among other stuff.)
Here's some examples of the current ones:

 

[RenderedFatality]

Hey! i currently use Dazed for *extremely* fast parallelized translations with VLLM within about the span of an hour or less for games of my interest using local LLM's, particularly 32b's and 14b's and anything smaller if i think it'd be interesting

Your current project looks *really* fucking cool. this is some amazing work you've done. Especially with the DSLR engine and its support for openai compatible endpoints. I've personally started to use it after the 'dazed' pass to translate the plugins.js file when i can, as its very useful for doing that.

If you need any testing done with local models other than deepseek, like the qwen series or other promising models, I have my own setup and could run tests/translations following your specifications and report back results.

also

Please consider sometime in the future to finish implementing concurrent requests for users like myself who use a production serving engine with the hardware to support it. Doing one batched translation at a time is a vast underutilization of the power these GPU's can provide, to the scale of 10 or 20 times slower than doing multiple requests in parallel.

 

[derakino999]

Please consider sometime in the future to finish implementing concurrent requests for users like myself who use a production serving engine with the hardware to support it. Doing one batched translation at a time is a vast underutilization of the power these GPU's can provide, to the scale of 10 or 20 times slower than doing multiple requests in parallel.

but your signature says you like to fine tune LLM's that fit on phones? PHONES?
that aside, you got a list of games you've translated?

 

[RenderedFatality]

but your signature says you like to fine tune LLM's that fit on phones? PHONES?
that aside, you got a list of games you've translated?

Idk what finetuning llm's that fit on phones says about me. All i know is that when given compute. Use compute.

And, yeah. Are they good? They're slop.

https://f95zone.to/threads/translation-request-cursed-blessing-no-circle-king.258982/post-17335106

https://f95zone.to/threads/translat...i-kunoichi-rj01284260-no.257247/post-17195720

https://f95zone.to/threads/translat...su-rj01333591-pomesaurus.256786/post-17194531

https://f95zone.to/threads/translat...vens-ornament-rj01373314.254342/post-17167936

https://f95zone.to/threads/translat...lt-rj01146193-studionaze.203431/post-13514918

https://f95zone.to/threads/big-boob...chool-rj01146139-barista.197477/post-13513400

https://f95zone.to/threads/translat...ion-rj01158001-gyoraijyo.199163/post-13191411 + https://f95zone.to/threads/sentai-s...nization-v1-00-gyoraijyo.245299/post-17335497


its all slop, but readable slop.

And most of those i haven't even dreamed of touching plugins.js until i saw this thing. might need to do a second pass on em

 

[derakino999]

Idk what finetuning llm's that fit on phones says about me. All i know is that when given compute. Use compute.

And, yeah. Are they good? They're slop.

https://f95zone.to/threads/translation-request-cursed-blessing-no-circle-king.258982/post-17335106

https://f95zone.to/threads/translat...i-kunoichi-rj01284260-no.257247/post-17195720

https://f95zone.to/threads/translat...su-rj01333591-pomesaurus.256786/post-17194531

https://f95zone.to/threads/translat...vens-ornament-rj01373314.254342/post-17167936

https://f95zone.to/threads/translat...lt-rj01146193-studionaze.203431/post-13514918

https://f95zone.to/threads/big-boob...chool-rj01146139-barista.197477/post-13513400

https://f95zone.to/threads/translat...ion-rj01158001-gyoraijyo.199163/post-13191411 + https://f95zone.to/threads/sentai-s...nization-v1-00-gyoraijyo.245299/post-17335497


its all slop, but readable slop.

And most of those i haven't even dreamed of touching plugins.js until i saw this thing. might need to do a second pass on em

awesome thanks!
and i still don't understand that about finetuning the LLMs lol??? but it's okay

 

[kukuru97]

Stuck while exporting .JSON





The older version has no issues exporting .JSON files

 

[Shisaye]

Stuck while exporting .JSON

View attachment 4971389



The older version has no issues exporting .JSON files
View attachment 4971727

Which game/file? (dlsite code?)
Was that actually parsed with the new version or are you exporting an old project?

Edit: I tested some random jsons on my end and they exported fine, please send me your example so I can figure out what's going on.

 

[Shisaye]

I've released SLR Translator v1.134.

The new Do-Duplication parsing now works properly even for VX and VX Ace.
No de-duplication anymore, and without it being super laggy/slow.
Unless you are running SLR Translator on an almost dead potato it's now superior to normal parsing.
(But still turned off by default because it might cause issues with people loading old projects, and the whole feature is more catered towards serious translators and less for people just doing some MTL.)
It now also grabs all string numbers, meaning if they have "", '', or `` around them.
It will also grab floating numbers for MZ which allows you to change font size in the tool now.

(I'm trying to make it grab integers in general in the future so you can edit VX/Ace scripts without needing the editor anymore, but that's a lot harder/messier than it sounds.)

 

[kukuru97]

Which game/file? (dlsite code?)
Was that actually parsed with the new version or are you exporting an old project?

Edit: I tested some random jsons on my end and they exported fine, please send me your example so I can figure out what's going on.

nvm, I was exporting an old project.
I tried creating a new one, and it worked fine.

 

[Shisaye]

nvm, I was exporting an old project.
I tried creating a new one, and it worked fine.

It would still be interesting to look at that old project. Because I was actually trying to leave parsing alone unless the do-duplication option is used (to keep it backwards compatible).
So I still must have fucked up somewhere.

I would need the .trans file and the staging data folder from php cache, or the original json so I can parse it myself using 1.129.

 

[kukuru97]

It would still be interesting to look at that old project. Because I was actually trying to leave parsing alone unless the do-duplication option is used (to keep it backwards compatible).
So I still must have fucked up somewhere.

I would need the .trans file and the staging data folder from php cache, or the original json so I can parse it myself using 1.129.

Here are the .trans and .json files, in case you're still interested.
Also, here’s the DLsite link for the game :

You must be registered to see the links

btw, I tried the DSLR, but I got an error like this and couldn’t stop the process. I had to restart the app to force it to stop

 

[Shisaye]

Here are the .trans and .json files, in case you're still interested.
Also, here’s the DLsite link for the game :

You must be registered to see the links

btw, I tried the DSLR, but I got an error like this and couldn’t stop the process. I had to restart the app to force it to stop
View attachment 4980716

That's odd. I ran some tests with an R1 distill version and I was getting an output.
Not a usable one, because the current version of DSLR doesn't know how to handle the "think" block at the start of the response of a reasoning model like that, but it shouldn't just outright crash. (I'm working on fixing the reasoning stuff right now.)
The one I tried was deepseek/deepseek-r1-0528-qwen3-8b though.

Edit: I should mention that I still haven't found a single model that can be locally hosted on reasonable hardware that actually gives acceptable performance in DSLR, yet.
The best one so far was actually gemma-3-27b, but still much worse than the official deepseek-chat-v3, or the experimental deepseek-chat-v3-0324 model, and incredibly slow.

 

[kukuru97]

That's odd. I ran some tests with an R1 distill version and I was getting an output.
Not a usable one, because the current version of DSLR doesn't know how to handle the "think" block at the start of the response of a reasoning model like that, but it shouldn't just outright crash. (I'm working on fixing the reasoning stuff right now.)
The one I tried was deepseek/deepseek-r1-0528-qwen3-8b though.

Edit: I should mention that I still haven't found a single model that can be locally hosted on reasonable hardware that actually gives acceptable performance in DSLR, yet.
The best one so far was actually gemma-3-27b, but still much worse than the official deepseek-chat-v3, or the experimental deepseek-chat-v3-0324 model, and incredibly slow.

I tried Meta-Llama-3.1-8B-Instruct-Q4_K_M and Gemma-2-9b-it-Q3_K_L, but I still get the same error.

 

[Shisaye]

I tried Meta-Llama-3.1-8B-Instruct-Q4_K_M and Gemma-2-9b-it-Q3_K_L, but I still get the same error.

But you can start a chat with them inside LLMStudio right? The models themselves are working properly?

 

[kukuru97]

The model works well—both inside LM Studio and in Translator++.
It even runs smoothly in the app I made a few days ago (and I’m not even a programmer, lol—I had help from Cursor to build it).

 

[Shisaye]

I tried Meta-Llama-3.1-8B-Instruct-Q4_K_M and Gemma-2-9b-it-Q3_K_L, but I still get the same error.

I just tried again on my end, and while the output is unusable garbage because DSLR doesn't know how to handle the reasoning, yet, I do not see an error.