Sarvam-M LLM now open for download on Hugging Face, boosting Indian AI innovation

Sarvam AI's 24-billion-parameter Sarvam-M sets new benchmarks in Indian language and math tasks with +86% improvement, now publicly accessible via API and Hugging Face, enabling developers to integrate advanced AI capabilities tailored for Indian contexts.

Sources:
The Indian Express
Updated 2h ago
Tab background
Sources: The Indian Express
Indian AI startup Sarvam has launched its flagship Large Language Model (LLM), Sarvam-M, now available for download on Hugging Face and accessible via Sarvam's API.

The model is a 24-billion-parameter open-weights hybrid language model built on top of Mistral Small, designed to advance Indian AI innovation.

Sarvam-M has set new benchmarks in mathematics, programming tasks, and Indian language understanding. It underwent a rigorous three-step enhancement process involving Supervised Fine-Tuning (SFT), Reinforcement Learning with Verifiable Rewards (RLVR), and Inference Optimisations.

In particular, the model demonstrated an 86% improvement on combined tasks involving Indian languages and math, such as the romanised Indian language GSM-8K benchmark.

This release marks a significant step forward in AI tailored for Indian languages and complex problem-solving, providing developers and researchers with a powerful tool for experimentation and integration.

"The Sarvam-M model is currently accessible via Sarvam’s API and can be downloaded from Hugging Face," enabling broader access to cutting-edge AI technology.

The open availability of Sarvam-M is expected to accelerate innovation in the Indian AI ecosystem, fostering advancements in language processing and computational tasks specific to the region.
Sources: The Indian Express
Indian AI startup Sarvam has released its flagship 24-billion-parameter Large Language Model, Sarvam-M, on Hugging Face. The hybrid model excels in mathematics, programming, and Indian language tasks, showing an 86% improvement on benchmarks, and is available for download and API access to boost Indian AI innovation.
Section 1 background
The model has been enhanced through a three-step process – Supervised Fine-Tuning (SFT), Reinforcement Learning with Verifiable Rewards (RLVR), and Inference Optimisations.
Sarvam AI Official Blog
The Indian Express
Key Facts
  • Sarvam AI, an Indian startup, has developed Sarvam-M, a 24-billion-parameter open-weights hybrid language model built on Mistral Small.The Indian Express
  • Sarvam-M underwent a rigorous three-step enhancement process including Supervised Fine-Tuning (SFT), Reinforcement Learning with Verifiable Rewards (RLVR), and Inference Optimisations.The Indian Express
  • Sarvam-M has set new performance standards in mathematics, programming tasks, and Indian language understanding.The Indian Express
  • On combined Indian language and math tasks such as the romanised GSM-8K benchmark, Sarvam-M demonstrated an impressive +86% improvement.The Indian Express
  • Sarvam-M is now accessible via Sarvam's API and is available for download on Hugging Face for experimentation and integration.The Indian Express
Key Stats at a Glance
Model size of Sarvam-M
24 billion parameters
The Indian Express
Performance improvement on GSM-8K benchmark
+86%
The Indian Express
Article not found
⌂ Home ‹

Source Citations