argos-translate

Argos Translate has open source neural machine translation https://github.com/argosopentech/argos-translate

Opus is great! This is what Argos Translate uses (https://github.com/argosopentech/argos-translate).

Its online mirrors[0] (which works without API keys) could be used as dictionary program to GoldenDict via LibreTranslate wrappers[1], so it may be used as a "LibreTranslate Desktop app" with a Scan Pop-up feature for translating any selected text in any app on PC. (GoldenDict could be used as alternative GUI to Argos Translate engine's GUI[2] directly, but keep running engine on personal PC in offline may highly decreases PC performance) JFTR, Here is "Google Translate Desktop app" implemented using GoldenDict + `bash` + single Shell-script.[3] Alternatively, there is Python wrapper to GTranslate for GoldenDict.[4] [0] https://github.com/LibreTranslate/LibreTranslate#mirrors [1] https://github.com/LibreTranslate/LibreTranslate#language-bi... [2] https://github.com/argosopentech/argos-translate#gui [3] https://github.com/goldendict/golden

GPT Paper: "Specifically, we train GPT-3, an autoregressive language model with 175 billion parameters" README: "1T or bust my dudes [...] An implementation of model & data parallel GPT2 & GPT3 -like models, with the ability to scale up to full GPT3 sizes (and possibly more!)" It seems the largest model they released is 2.7 billion parameters or ~0.01 the size of GPT-3. The most interesting part about GPT-3 was its size and it seems this is only "GPT-3-like" in architecture. I also have a translation library with ~100 million (0.001 GPT-3) parameters: https://github.com/argosopentech/argos-translate

I don't have anything against Flatpak but maintain a Snap package and think Snap has made a lot of good design decisions even if it still has rough edges. For a basic application it's pretty easy to package for Snap. A lot of the things that bother Linux power users like polluting their mounted file systems wouldn't matter to regular users. My understanding is they support either pre-declaring permissions or requesting them at run time. https://github.com/argosopentech/argos-translate

You're right there's surprisingly little data available for English-Chinese. I'm not sure why, it seems like there would be a lot of demand for translating between them. For the en-zh model copying this is a known issue: https://github.com/argosopentech/argos-translate/issues/4

> https://github.com/uav4geo/LibreTranslate/issues/33 That issue is about emojis, which confused me a bit, but I see that the linked commit is about replacing <unk> by the corresponding source token. https://github.com/argosopentech/argos-translate/commit/6a0f... That's definitely an improvement, but what I was actually wondering about was why the model was copying the input verbatim in the first place. Looking through the en-zh data in Opus, it looks like a bit of a mixed bag. This sample from OpenSubtitles v1 contains an ad for MyDVDrip in the Chinese version only: http://opus.nlpl.eu/OpenSubtitles/v1/en-zh_sample.html Filtering the data with some heuristics might be a good idea.

It's based on argos-translate, which has python bindings: https://github.com/argosopentech/argos-translate

There are different models for each language pair. Currently there are only pre-trained models to and from English and other language pairs "pivot" through English. ex: es -> en ->fr Chinese is the weakest language pair currently, but I'm currently working on improving it: https://github.com/argosopentech/argos-translate/issues/17

The process for translation is to "tokenize" a stream of characters into "tokens" like this: "Hello Sarah, what time is it?" <Hello><_Sarah><,><_what><_time><_is><_it><?> Then the tokens are used for inference with a Transformer net. There are no guarantees that output will be consistent to even small changes in the input. The network, based on the data it was trained on/luck, has slightly different connections for the <_Sara> token than for the <_Sarah> token leading to different output. Here's a video of some Linux YouTubers reviewing Argos Translate (https://github.com/argosopentech/argos-translate), the underlying translation library, and getting unexpected outputs: https://www.youtube.com/watch?v=geMs9dxl1N8

Is there a recommended way to license a project as permissively as possible? I like the MIT license because it's widely accepted and short enough that it can be read and understood but what if you're OK with someone using a small (but large enough to be a copyrighted work) piece of your code without bothering to maintain the license. There's the WTFPL License (https://en.wikipedia.org/wiki/WTFPL) but that doesn't seem like a great option either. For my project (https://github.com/argosopentech/argos-translate) I dual licensed under the MIT and CC0 licenses even though CC0 isn't really designed for software but it seemed reasonable. It seems surprisingly difficult to release code with no conditions besides limiting liability.

Plug for my open source "local-first" alternative to Google Translate (https://github.com/argosopentech/argos-translate). It has a Google Translate like GUI and lets you install packages to support translating between different languages. It currently doesn't have the ability to go to web services for translations you don't have installed locally but that's a feature I'm considering adding at some point. It would seem like the best way to do this we be to track (locally) which translations you use frequently and then save them locally (a one way translation package between two languages is ~100MB). This would give you the privacy/offline access of translating locally for language pairs you frequently use, while also seamlessly supporting the large number of languages supported by cloud services. You could also give power users much more f