Sarvam AI Open-Sources 30B and 105B Reasoning Models Trained in India
By
logicchains
Crisp on the outside, thoughtful on the inside. A keeper.
Summary
Sarvam AI is open-sourcing two large language models - Sarvam 30B and Sarvam 105B - which are reasoning models trained entirely in India using compute resources from the IndiaAI mission. The models were developed through a full-stack approach including custom datasets, optimized tokenization, model architecture, execution kernels, scheduling, and inference systems. They are Mixture of Experts (MoE) models that show strong performance on Indian-language benchmarks and are being released with weights available on AI Kosh and Hugging Face platforms.
Key quotes
· 5 pulledWe're releasing Sarvam 30B and Sarvam 105B as open-source models.
Both are reasoning models trained from scratch on large-scale, high-quality datasets curated in-house across every stage of training: pre-training, supervised fine-tuning, and reinforcement learning.
Training was conducted entirely in India on compute provided under the IndiaAI mission.
These models represent a true full-stack effort.
Beyond datasets, we optimized tokenization, model architecture, execution kernels, scheduling, and inference systems to make deployment efficient across a wide range of hardware.
You might also wanna read
Xiaomi Releases MiMo: Open-Source AI Model Series Optimized for Reasoning Tasks
Xiaomi has released MiMo, an open-source large language model series under Apache 2.0 license that is specifically designed for reasoning ta
Product Hunt·1mo ago
Sapient Intelligence Releases HRM-Text-1B: A 1B Parameter Language Model with Hierarchical Reasoning Architecture
Sapient Intelligence has released HRM-Text-1B, a 1 billion parameter language model built on the Hierarchical Reasoning Model (HRM) architec
Switzerland Launches Open-Source Apertus AI Model as Alternative to Proprietary Systems
Switzerland has launched Apertus, an open-source AI model trained on public data as an alternative to proprietary models like ChatGPT and Cl
The Verge·9mo ago
Phi-4 Reasoning: Small Open-Weight AI Models with Strong Math and Science Capabilities
Phi-4 Reasoning is a small open-weight language model (3.8B/14B parameters) that delivers powerful reasoning capabilities for math, science,
Product Hunt·2mo ago
Cohere: Enterprise-Grade Language AI Models for Secure Cloud Deployment
Cohere is a platform offering high-performance, secure language models (LLMs) designed for enterprise use. Their customizable models can be
Product Hunt·9mo ago