TextSynth is a company specializing in text completion using large language models. With a wide range of transformer variants such as GPT-J, Boris, GPT-NeoX 20B, Flan-T5-XXL, and CodeGen-6B-mono, users can conduct their NLP work with the model that best fits their specific needs.
Their flagship product, GPT-J 6B, is an English language model with 6 billion parameters that is also capable of working with computer languages and other languages as well. TextSynth offers tools for both text completion and image generation, with an integrated REST JSON API that offers a prompt and user-friendly solution for users. Additionally, their AI-generated content offers much more natural, human-like language, making it an ideal tool for those looking to improve their writing.
TLDR
TextSynth is a company specializing in text completion using large language models such as GPT-J and Boris. Their tools offer a wide range of transformer variants for NLP tasks and are designed to be user-friendly and efficient, with custom cuda kernels and efficient custom 8-bit and 4-bit quantization that can perform batch operations with superior speed and efficiency. TextSynth offers both text and image generation tools, with extensive documentation available to users through their website.
Their language models create AI-generated content that reads much more naturally, making it ideal for those looking to improve their writing. TextSynth's pricing system is transparent and straightforward, allowing users to estimate the cost of their requests before making a purchase.
Company Overview
TextSynth specializes in text completion using large language models such as GPT-J, Boris, GPT-NeoX 20B, Flan-T5-XXL, and CodeGen-6B-mono. Their flagship product, GPT-J 6B, is an English language model with 6 billion parameters that is also capable of working with computer languages and other languages as well. To cater to French language users, TextSynth offers Boris, which is a fine-tuned version of GPT-J for the French language.
For those in need of even larger models, GPT-NeoX 20B is available, which is the largest model currently offered by TextSynth. Additionally, Flan-T5-XXL is a fine-tuned language model specifically designed to answer questions. All of these models are extensively documented and available to users through TextSynth's website.
TextSynth's tools allow users to type in text and have the AI neural network generate different randomly chosen completions every time. Users have the ability to adjust different settings, such as top-k, top-p, temperature, max tokens, and stop at, depending on their specific needs. TextSynth's tools are designed to be easy to use, while also being highly customizable to meet the demands of more advanced users.
TextSynth's technology offers numerous potential applications, from aiding with content creation to assisting with customer service response generation. Their advanced language models allow for more natural, human-like language, making it an ideal tool for those looking to improve the writing with AI-generated content. Overall, TextSynth's language models are some of the most advanced and reliable currently available on the market, making them a valuable asset for those looking to improve and expand their use of AI text completion tools.
Features
Transformer Variant Support
Multiple Transformer Variants
With TextSynth, users have access to a wide range of Transformer variants for their NLP tasks. This includes GPT-J, GPT-NeoX, GPT-Neo, OPT, Fairseq GPT, M2M100, CodeGen, and GPT2. Having a broad selection of models ensures that users can conduct their NLP work with the model that best fits their specific needs.
Integrated REST JSON API
Text Completion, Translation, and Image Generation
TextSynth significantly eases the process of text completion, translation, and image generation with its integrated REST JSON API. Users do not need to learn a new language or use an external library.
The API is user-friendly and efficient, enabling users to generate the output they need promptly.
High-Performance
Small and Large Batch Performance
TextSynth's high-performance features make it ideal for handling large and small batch sizes. With custom cuda kernels and efficient custom 8-bit and 4-bit quantization, TextSynth can generate tokens at a high speed.
This feature ensures that TextSynth can perform batch operations with superior speed and efficiency.
Optimal Performance on Lower-Cost GPUs
Larger models typically require more significant resources to generate outputs; however, TextSynth's efficient quantization ensures optimal performance on lower-cost GPUs like the RTX 3090 and RTX A6000. By staying efficient as far as resource consumption is concerned, users can run resource-intensive tasks without breaking the bank.
No External Dependency
TextSynth includes everything required to get started, from custom inference code designed for faster inference performance on both GPUs and CPUs to the LibNC library used for tensor manipulation using C language. Users can, therefore, easily install TextSynth on most Linux distributions.
CPU-Only Version Included
Freely Available
TextSynth offers users the option to use a CPU-only version, which is freely available. This allows less CPU powerful computers to perform tasks without worries, as this feature is perfect for users who don't have a dedicated GPU at their disposal.
Benchmarks
Performance Using GPT-Neox 20B Model on a RTX A6000 Nvidia GPU
Users can rely on TextSynth to run their NLP tasks efficiently, as shown by the following benchmarks. 200 tokens are generated using a batch size of 1 for the speed measurement.
- Precision - Float16
- LAMBADA (ppl) - 3.66
- LAMBADA (acc) - 72.6%
- Max GPU memory (GB) - 40.7
- Speed (tokens/s) - 15
- Precision - 8 bits
- LAMBADA (ppl) - 3.66
- LAMBADA (acc) - 72.6%
- Max GPU memory (GB) - 21.7
- Speed (tokens/s) - 27
- Precision - 4 bits
- LAMBADA (ppl) - 3.71
- LAMBADA (acc) - 72.0%
- Max GPU memory (GB) - 11.6
- Speed (tokens/s) - 41
Performance Using Stable Diffusion 1.4 Model on a RTX A6000 Nvidia GPU
Users can rely on TextSynth's ability to handle complex image generation tasks as shown below. A single image is generated using 50 timesteps and a batch size of 1 for the speed measurement.
- Precision - Float16
- Max GPU memory (GB) - 2.8
- Generation time (s) - 1.90
Pricing
TextSynth's pricing system is based on the number of input tokens and generated tokens used in a request. The cost of each request is calculated by multiplying the number of input tokens by the input token price and adding the number of generated tokens multiplied by the generated token price. Assuming 4 characters per token, the input token cost is neglected, and the generated token cost is more expensive, as it requires more computation resources.
The number of tokens processed by the language model depend on the model and content, but as an example, assuming 300 input tokens and 30 generated tokens, the cost would be determined by multiplying 330 by the respective token prices. It is important to note that the cost may vary depending on the model and language used.
For image generation requests, the price is determined solely by the number of generated images. Overall, TextSynth's pricing system is transparent and straightforward, allowing users to estimate the cost of their requests before making a purchase.
FAQ
What is TextSynth?
TextSynth is a company specializing in text completion using large language models such as GPT-J, Boris, GPT-NeoX 20B, Flan-T5-XXL, and CodeGen-6B-mono. Their language models are extensively documented and available to users through TextSynth's website.
The flagship product of TextSynth, GPT-J 6B, is an English language model with 6 billion parameters that can also work with computer languages and other languages as well. For those seeking out larger models, the company also offers GPT-NeoX 20B.
What are some potential applications of TextSynth's technology?
TextSynth's technology offers numerous potential applications, from aiding in content creation to assisting with customer service response generation. Their advanced language models allow for more natural, human-like language, making it an ideal tool for those looking to improve their writing with AI-generated content.
By doing so, users are empowered to streamline and improve the customer experience by deflecting support tickets and issues.
What are the differences between the models offered by TextSynth?
The models offered by TextSynth have different parameters and are designed to cater to different needs. GPT-J 6B is TextSynth's flagship product, while Boris is a fine-tuned version of GPT-J for the French language.
For those seeking out the largest available models, GPT-NeoX 20B is an option, while Flan-T5-XXL is a fine-tuned language model designed to answer questions. All of the models at TextSynth are available for extensive documentation and testing for users of their site.
Can I customize settings while using TextSynth?
Yes, TextSynth's tools are designed to be easy to use, while also being highly customizable to meet the demands of more advanced users. Users have the ability to adjust different settings, such as top-k, top-p, temperature, max tokens, and stop at, depending on their specific needs. These different settings allow users to have a higher degree of control over the content and how it is generated.
What sets TextSynth's language models apart from other text completion tools?
TextSynth's language models are some of the most advanced and reliable currently available on the market. They offer much more natural, human-like language when compared to other tools, making it an ideal choice for those looking for AI-generated content that reads as if it was written by a human. Additionally, the tools are designed to be both easy to use and customizable for more advanced users.
