# Uncle Bane's Offline Japanese to English Translation Tool
Haven't yet being able to read the Yamato's Moonrunes for your basket weaving games and you are depending off from people like these:

Well don't despair anymore for I bring you a Tool that [I actually made entirely with the idea of selling it first and foremost on](https://donate.nexus34.xyz/) over my Patreon page:
## https://donate.nexus34.xyz/
You can load your own Glossaries in order to help the tool give more accurate translations (mostly regarding the names and special terms)

So you can now get something like this:

You can watch a video of how to run this more or less over here:
https://www.youtube.com/watch?v=5IVWkGWfktM
And if you hate yourself or something like that, a CPU-Only version:
https://www.youtube.com/watch?v=AiyK99N84xU
(ignore the mistranslation of Kotoko name into Koto, I didn't load the Glossary when I did that first run of the translation, thus why it is important to preload your own glossaries for that reason)
## Choose your preferred Model
With this release I'm releasing an experimental Llama 3.1 based Translation model too, specially for those that may want to create LoRAs for it they could run the model along side a finetuned LoRA that you have made too perhaps.

Keep in mind that the Llama 3.1 based model is slightly heavier than the OG Qwen-based one, so if you *don't have* a 8GB VRam GPU, you will need to run it on CPU+Ram and for that you will need to change the Configs/names_models.json file and change "GpuLayers": 32 to 0.

## Many VNs CSV included
You can find them on the CSV Folder and you can also make your own too for the Visual Novel you are playing, because I really wanted to released this before the month ended I didnt made a more dedicated Terms CSV, I will release that on my Patreon page in a few days hopefully.
## What's next?
My next release would be my Uncle Bane's ToolKit tool (UBTK) which is more focused for those wanting to use LLMs for gaming in general, not so much for translation Jobs so if you are interested drop by my patreon page for that.
I checked LFM2-350M-ENJP-MT which both its base model and this finetune seem to have just released this month and what I can say its that clearly this Finetune needed way more time in the Oven (finetuning) https://imgur.com/a/JmsU9MW , one thing I will say though its that what they say about being "near real time" its correct, its obvious the model was made for low specs devices like smartphones and used for stuff like live conversation translators and such, now for people without 8GB GPUs honestly you are better off running UBTL just on CPU+Ram , its not as fast as you could do so running it through the GPU but still rather usable if you are patient enough. Also something that many need to know is that most corporate made LLMs are heavily censored regarding NSFW stuff, I know because when I was starting the project I was instead just thinking in just looking up already made JP<->EN Translation Models and Finetune it a little more a lot of work and the like but I saw that pretty much all those models are aggressively censored towards anything NSFW, by that I mean the LLM will outright refuse to translate you any text that gets lewd in many of those types, UBTL handles some comfortable levels of NSFW already (but could be still trained even more to further "decensor" any remnants of its corporate training lingering in its core here and there from its source model)
Comments - 5
lordfifth
anikitoshokan (uploader)
lordfifth
kvirin
whnc8sk6r8mq