10 seconds of audio in 100ms - an RTF of 0.01! Lightning infact has a higher network latency than the actual time it takes for the model to do inference! It seems like just yesterday we had models like Tortoise that would take up to 10 minutes to generate 5 seconds of audio. While we move hard to make Lightning a part of public benchmarks, you can start trying this already on - https://waves.smallest.ai Join us on discord to know more - https://lnkd.in/gpVXkZsF This wouldn't have been possible without the superb team at smallest.ai - Akshat Mandloi Wasim Madha Rishabh Dahale Gaurav Verma Manav Lakhadive Pritish Mishra Nityanand Mathur Devansh Pawan Hamees S. Article - https://lnkd.in/dQK9nXWs
关于我们
Generative Multi-Modal AI Stack
- 网站
-
https://www.smallest.ai
smallest.ai的外部链接
- 所属行业
- 软件开发
- 规模
- 2-10 人
- 总部
- San Francisco，California
- 类型
- 私人持股
- 创立
- 2023
- 领域
- Artificial Intelligence、NLP、Computer Vision、Accelerated Computing、Data Science和Machine Learning
地点
-
主要
获取路线
US，California，San Francisco
smallest.ai员工
动态
-
smallest.ai转发了
Plans are live on Waves - https://waves.smallest.ai/ If you are looking for a reliable tool to create voiceovers for your video or power your voice bot, Waves by smallest.ai is the perfect platform for you. 1. 30 mins free audio every month 2. Latest open-source models available for testing 3. Lightning-fast API with under 100ms latency 4. Hindi, English, Hinglish support Connect with us on discord to get access to beta features - https://lnkd.in/gpVXkZsF
-
-
Join us on our discord today! Link: https://lnkd.in/gQRUNHkE
smallest.ai is an engineering company. Every single person who joins us has some deep technical expertise that makes them immensely valuable to everyone else. However, we understand that being great at science and engineering is not enough. Understanding the market is equally important. What’s our GTM? What are areas where we see white spaces? Why are we unique? Below is a picture of one of the sessions where Akshat Mandloi and I deep dive into our monthly and quarterly strategy, and explain how each and every engineering task is linked to a large market opportunity. In this session, we are responsible for answering every question our team asks and helping them understand our mission. We believe in complete transparency and this makes it extremely exciting to work with us. Want to know more? Join us on discord - https://lnkd.in/gpVXkZsF
-
-
We’re looking for an ML/Python intern at smallest.ai to start immediately. Great at - threads, processes, basic prompt engineering. Should be quick to pick things up. Need any one strong proof of work/excellence. Location Indiranagar Mail [email protected]
-
Follow us for more updates and don't forget to join our discord: https://lnkd.in/deSb5qTp
Today, we announce the beta launch of AWAAZ, our multi-lingual , multi-accent text-to-speech model for Indian languages. Top highlights are: 1. State of the art MOS in Hindi, Indian English 2. Speaks in 10+ accents 3. Single shot voice cloning using a 5 second audio clip 4. 200ms streaming latency And the best part is, as an introductory offer, it will cost 10 times lesser than it's competitors - Rs. 999 for 5,00,000 characters! Why did we build AWAAZ? When we started looking for TTS models for building voice-bots for Indian languages, we realized that the current models are either: a) subpar in terms of quality or b) too expensive for scaling up This is true because speech data quality for TTS systems reduces drastically as we move away from English to other languages. With AWAAZ, we explore the use of high quality multi-language, multi-accent datasets and set up a platform on which we will continue building for other languages, prioritizing India and South Asia. Adding a small demo we made that demonstrates its capabilities. Would love to hear your feedback regarding the same. Link to the model in comments! Akshat Mandloi
-
smallest.ai转发了
smallest.ai has launched the beta version of AWAAZ, an advanced text-to-speech (TTS) model tailored for Indian languages. Founded in 2023 by Sudarshan Kamath and Akshat Mandloi, smallest.ai is based in San Francisco and aims to build voice-based generative AI models. They are currently seeking user feedback through a demo of AWAAZ’s capabilities. ?? Read the complete article to learn more Link in the comments below ?? #TTS #AI #IndianLanguages #AWAAZ #TechInnovation
-
-
Introducing the "Tree of Thoughts" Framework for Enhanced Language Model Inferencing ?? Existing language models are confined to linear, token-by-token inferencing that limits performance on tasks requiring planning, exploration, or strategic thinking. But what if LMs could evaluate multiple reasoning paths and revisit past choices before acting? Enter ToT - the new "Tree of Thoughts" framework that upgrades LMs from a rigid "Chain of Thought" to a flexible tree structure enabling nonlinear inferencing: - Supports deliberation by assessing possible sequences of thought ?? - Permits backtracking and looking ahead before committing to text ?? - Significantly enhances strategic reasoning and problem solving abilities ?? Experiments reveal major gains over traditional LM prompting, like pushing Game of 24 solving from 4% to 74% accuracy for GPT-4! The code and prompts are available: https://lnkd.in/dMp3MYmG This unlocks new frontiers in language model inferencing through adaptive, non-linear thought processes just like we humans employ. What other tasks could benefit from the ToT upgrade? Share your ideas! ??