🚀 Medial just raised $500k in Pre-Series A. Tap to read the announcement. 🎉
Login

Back to feeds

Applyly

Stealth • 3m

Indian AI Startup Sarvam Launches ‘SARVAM 1’ – LLM Trained on 10 Indic Languages: The Indian AI ecosystem is witnessing a lo of new innovations day by day. A new innovation by SARVAM AI has come in news. It has launched an LLM trained model called SARVAM 1. the design of the model id to support Indian languages. According to a source, it supports 10 indic languages like Bengali, Gujarati, Hindi, Marathi, Malayalam, Kannada, Odia, Tamil, Telugu, Punjabi, and English. The startup has also launched AI voice agents for customer services and sales. What is SARVAM 1? SARVAM 1’s job is to understand and generate human-like text in multiple languages. This model is first of it’s kind that has been trained from scratch with data research. The facts and figures related to Sarvam AI are: Training Period: It took over 10,000 hours of training for Sarvam 1 to learn everything. Data: The team used data from 50 million native language speakers to create a realistic and relatable language model. Computing Power: The startup used over 1,000 GPUs, to speed up the training process. GPU’S: NVIDIA provided GPU’s and software frameworks. SIGNIFICANCE- SARVAM 1 Build on 2 billion-parameter architecture, it can power a range of things like voice and messaging agents. It represents an important step in AI ecosystem of India. Sarvam 1 has setted up a benchmark for other AI startups in India. It is an inspiration for various other startups who are working on various AI models. This homegrown AI can achieve wonders if more efforts are put into making it THE BEST MADE-IN-INDIA MULTI-LINGAL VOICE AGENT in the world. Thanks Read more about startups here.

3 replies4 likes
Replies (3)

More like this

Recommendations from Medial

Sarun George Sunny

Stealth • 4m

Sarvam AI: Empowering Indian Languages with AI Sarvam AI, co-founded by Vivek Raghavan and Pratyush Kumar, focuses on building AI that caters to India’s linguistic diversity. Their Indic Voice Large Language Models (LLMs) allocate 40% of their token

See More
0 replies3 likes
Image Description
Image Description

Yash Barnwal

Stealth • 4m

🦾AI model training comes with significant costs. This chart highlights the estimated expenses for training top AI models like Google’s Gemini 1 and OpenAI’s ChatGPT-4 in 2023. These costs, calculated using both cloud computing and amortized hardware

See More
4 replies13 likes
6
Image Description

Sidharth Bidikar

Stealth • 8m

Isn't the internet dominated by English? Can we really build effective Indic Language Models? #Tech #AI #LanguageModel #IndicLanguages

2 replies2 likes
Anonymous
Image Description

Anonymous

Stealth • 21d

What are the sectors that are not going to be affected with AI being around? Like I know my support job is at stake because I myself trained a model and already removed few employees replacing them with AI chat bots and Voice agents Is being a prom

See More
1 replies3 likes
Image Description
Image Description

Chamarti Sreekar

Stealth • 21d

Another open-source model has arrived, and it’s even better than DeepSeek-V3. The Allen Institute for AI just introduced Tülu 3 (405B) 🐫, a post-training model that is a fine-tune of Llama 3.1 405B, which outperforms DeepSeek V3.

10 replies29 likes
14
Image Description

Payal Manghnani

Stealth • 22d

Think China is behind in AI? Think again. Most people believe DeepSeek is China’s only top AI model, but that’s far from the truth. China has **TEN** top-tier models trained from scratch—rivaling Europe's best and even Mistral's biggest model. Me

See More
1 replies4 likes
Image Description

Krish Vaidya

Stealth • 1m

Building and selling agents such as Chatbots and Ai voice calling agents to business owners for a monthly fee also marketing their business with Ai. So ai voice calling agents work like customer support staff what I noticed that businesses spend a lo

See More
1 replies4 likes

Mohammed Zaid

Stealth • 5h

Helix Figure AI has unveiled Helix, a groundbreaking Vision-Language-Action model that enables humanoid robots to perform complex tasks through voice commands, marking a significant advancement in the field of robotics and artificial intelligence.

0 replies
Image Description
Image Description

Kunal sapkal

Stealth • 9m

I am thinking about an AI model with multiple languages support especially devnagari script languages , we are planning to build an RAG to parse a document and provide the data to the LLM , is this idea feasible and useful suggestions are welcomed

4 replies8 likes
1
Image Description
Image Description

Ripudaman Singh

Stealth • 5m

Recently had an idea about training a model on memes and making a simple classification model. Memes can be dark, intelligent, puns, and all sorts of things. If a model is trained on such a labelled dataset of memes, it can better classify a future m

See More
9 replies8 likes

Download the medial app to read full posts, comements and news.