AI + translation technologies = slAItor
We combined the recent advances in AI and large language models with state-of-the-art translation technologies.
slAItor combines recent advances in AI (GPT) with more standard machine translation techniques. A new kind of system that does much more than just translating.
slAItor can explain a translation step-by-step and with details. You can use this unique functionality to learn or improve your understanding of a language.
Thanks to the checker mode, you can provide your own translations to slAItor: It will spot, explain, and correct the errors.
slAItor's paraphraser and style calibrator are powerful tools that can adapt the translations to numerous contexts.
We combined the recent advances in AI and large language models with state-of-the-art translation technologies.
slAItor prepares your query to make it easy to process by the AI engine. It includes some normalization, tokenization, and splitting of your text. Then, everything is packed into a "prompt" to be sent to the AI engine. This step relies on several open source projects such as preprocess, sacremoses, and Pytorch.
For now, slAItor mainly exploits models from the GPT family by OpenAI. We keep exploring alternatives, including the development of our own model. For more information about these technologies, you can read my blog posts about translating with ChatGPT and translating with GPT-3.
This step checks and formats the answer. We use again several open source projects.
We continuously monitor slAItor's performance using in-house evaluation datasets as well as public benchmarks. We rely on a state-of-the-art automatic evaluation metric, COMET, and human evaluators.
Learn with: The Kaitchup - AI on a Budget. Some examples of posts:
Fast Speculative Decoding with Llama 3.2 and vLLM
DoRA vs. LoRA: Better and Faster than LoRA?
Train and Serve an AI Chatbot Based on Llama 3.2
QLoRA with AutoRound: Cheaper and Better LLM Fine-tuning on Your GPU
Run Llama 3.1 70B Instruct on Your GPU with ExLlamaV2 (2.2, 2.5, 3.0, and 4.0-bit)
Google's Gemma: Fine-tuning, Quantization, and Inference on Your Computer
Run Llama 3 70B on Your GPU with ExLlamaV2
SqueezeLLM: Better 3-bit and 4-bit Quantization for Large Language Models
QLoRA: Fine-Tune a Large Language Model on Your GPU
Generate Synthetic Data from Personas to Train AI Chatbots
Fine-Tuning Meta's Llama 3.2 1B & 3B Models on Budget GPUs
Multimodal RAG with ColPali and Qwen2-VL on Your Computer
"credit" is the unit slAItor uses to simplify the evaluation of the translation cost.
The cost in credits is automatically computed given the length of your text and the options you selected. This costs is indicated just below the text area where you enter your text.
Translating 1,000 characters costs 1 credits. This costs is tripled to 3 credits for the step-by-step translations and in the checker mode. The minimum cost of a query is 0.1 credit.
For example, with our subscription, you can translate up to 800,000 characters a month.
Credits are automatically debited from your account as soon as you press "translate" (or "Check" in the checker mode). If you stop the translation before it finishes, credits won't be refunded.
If you need more credits, in a near future we will propose packs of credits that you can buy separately at anytime.
We can tell you that it is a mix between standard machine translation technologies and state-of-the-art large language models from the GPT family.
The use of a large language model at the core of slAItor is the main reason why our costs are lower than other online translation systems.
Benjamin holds a Ph.D in machine translation. He has 10+ years experience in machine translation research and development.
Yes. We will maintain slAItor at the state of the art.
Major upgrades will be deployed at no additional cost.
Yes. We are working very hard to support around 50 languages by the end of 2023.
No. When using slAItor you may not notice it, especially when translating into English, but our internal evaluation shows that the translation quality is slightly worse than well-established translation systems.
On the other hand, slAItor is cheaper and proposes several features that other online systems don't have.
No. Words generated by slAItor will appear chunk by chunk. Our current estimation is that it may take up to 1 second per 20 words.
We are working on a more realistic demo.
No. We don't use your data to develop or improve slAItor and its models.
This may change in the future. If that changes and if you are already a customer, you will have the possibility to opt-in.
All your interactions with slAItor are saved for 30 days. This is only for better answering your potential reclamations. If you would like to remove your data from our server, contact us.
Yes. This is non-reversible. All your data are immediately deleted.
Yes. We grant you an unlimited license that doesn't expire and grant you the full property of slAItor outputs, including for a commercial use.
Yes. We answer your questions as fast as we can.
If you want a refund, contact us. We will refund you without any questions if less than 14 days has passed since your last billing for monthly plan or 1 month for yearly plan.
We refund also according to the credits you have already used. Contact us for more details.