Sarvam AI, a rising star in India's generative AI industry, has launched a new language model called Sarvam-1. This model was created exclusively for Indian languages and is open-source

Description

Sarvam AI, a rising star in India's generative AI industry, has launched a new language model called Sarvam-1. This model was created exclusively for Indian languages and is open-source. It supports eleven Indian languages, including Bengali, Hindi, Tamil, and English. Sarvam-1, which was released in October 2024, is the successor to the company's previous model, Sarvam 2B, which debuted in August 2024.

Overview Of Sarvam-1

Sarvam-1 is designed with two billion parameters. Parameters describe the complexity and capabilities of an AI model. By comparison, Microsoft's Phi-3 Mini contains 3.8 billion parameters. Sarvam-1 is designated as a small language model (SLM) because it contains less than 10 billion parameters. This differs from large language models (LLMs) such as OpenAI's GPT-4, which contains over a trillion parameters.

Technical Specifications

Sarvam-1 is powered by Yotta's 1,024 Graphics Processing Units (GPUs) and trained using NVIDIA's NeMo framework. The approach addresses a significant issue: the scarcity of high-quality training data for Indian languages. Existing datasets frequently lack sufficient depth and diversity. To address this, Sarvam AI developed its training corpus, Sarvam-2T.

Training Data

Sarvam-2T has an estimated 2 trillion tokens. The dataset is evenly spread throughout the ten available languages. It uses synthetic data generation techniques to improve the quality of the training data. About 20% of the dataset is in Hindi, with significant amounts in English and programming languages. This diversity enables the model to execute both monolingual and bilingual tasks.

Performance Metrics

Sarvam-1 is said to be more efficient in handling Indic scripts than prior LLMs. It uses fewer tokens each word, which increases its efficiency. The model outperformed larger AI models such as Meta's Llama-3 and Google's Gemma-2 on a variety of benchmarks, including MMLU and ARC-Challenge.

Benchmark Achievements

Sarvam-1 performed well on the TriviaQA benchmark, scoring 86.11 for Indic languages. This score outperforms Meta's Llama-3.1 8B, which earned 61.47. Sarvam-1 also has superior computational efficiency, with inference times 4-6 times faster than larger models like Gemma-2-9B and Llama-3.1-8B.

Practical Applications

Sarvam-1's excellent performance and great inference efficiency make it suited for practical applications, such as deployment on edge devices. This is especially relevant for real-world applications where computational resources may be constrained.

Sarvam-1 is available for download via Hugging Face, an online portal for open-source AI models. This accessibility enables developers and academics to use the model in a variety of applications utilizing Indian languages.

Welcome to Notopedia.com, your comprehensive and free online education resource that covers a vast array of competitive exams and government job opportunities in India. We understand the importance of excelling in sarkari exams, such as UPSC, IBPS, and SSC CGL, SSC CHSL which serve as gateways to esteemed government positions. Our platform offers detailed insights into exam patterns, syllabus, and previous year question papers for various exams, including NEET, JEE, CAT, CLAT, GATE and more. Whether you're preparing for banking exams like IBPS, SBI, railway recruitment or railway exams like RRB, or seeking admission to top-notch MBA and law programs, Notopedia provides you with a wealth of free study materials, mock tests, online test series and expert guidance. Stay up-to-date with the latest exam dates, results, and admit cards for exams like JEE Main, UGC NET, and CET. As you embark on your educational journey, make Notopedia your go-to destination for free mock tests, comprehensive resources, and the latest updates on sarkari results, government job vacancies, and competitive exams. Empower your learning with Notopedia and achieve your academic and career aspirations with confidence.

For more Updates and Information - Visit Notopedia's Bulletin Board

For Latest Sarkari Jobs - Visit Notopedia's Sarkari Jobs Section

For access to more than 20,000 Colleges - Visit Notopedia's College Section

For School Studies and Exams Preparation across 14 Boards - Visit Notopedia's School Section

For Comprehensive Preparation of Sarkari Job Exams - Visit Notopedia's Sarkari Exams Section

For Comprehensive Preparation of Competitive Exams - Visit Notopedia's College Entrance Exams Section