Reading
NCERT_3B_v0.1 is a lightweight inference model with 3 billion parameters. Fine-tuned on India's NCERT textbook data using the GRPO method, it can run 100% offline on low-end devices with only 3-6GB of memory after 4-bit quantization, aiming to bridge the digital divide in education.
Section 01
NCERT_3B_v0.1 is a lightweight inference model with 3 billion parameters. Fine-tuned on India's NCERT textbook data using the GRPO method, it can run 100% offline on low-end devices with only 3-6GB of memory after 4-bit quantization, aiming to bridge the digital divide in education.
Section 02
Globally, the distribution of high-quality educational resources is extremely uneven. Many students in developing countries cannot access the internet stably or use cloud-based AI tools. Mainstream large language models have large parameter sizes, requiring expensive GPUs and stable networks, which excludes the learners who need help the most. The NCERT_3B project aims to build a sufficiently small, fast, and fully offline model, allowing students in resource-poor areas to enjoy the convenience of AI-assisted learning.
Section 03
NCERT_3B uses a 3-billion-parameter scale, balancing memory usage, inference speed, and expressive power.
It uses Group Relative Policy Optimization (GRPO) for fine-tuning, which does not require a separate reward model. It estimates the advantage function through relative comparisons within groups, resulting in higher computational efficiency.
It uses Unsloth for 4-bit quantization and exports to the GGUF format (defined by llama.cpp, optimized specifically for CPU inference). The model file size is reduced, allowing smooth operation on devices with 3-6GB of RAM.
Section 04
The model's training data comes from NCERT textbooks, covering core subjects for grades 6 to 12. As the standard textbooks for India's public schools, they are widely representative. Reasons for selection: guaranteed data quality (strict review), wide coverage (multiple subjects and grades), and direct service to target users (a large number of Indian students rely on NCERT textbooks).
Section 05
Student interaction data is kept locally and not uploaded to the cloud.
Available anytime regardless of network conditions, suitable for areas with weak networks like rural regions.
Supports devices with 3-6GB of RAM; entry-level phones or low-end laptops can run it, lowering the barrier to use.
Section 06
Answers concepts and exercises from NCERT textbooks, providing explanations and guidance.
Provides hints and ideas for difficult homework problems, promoting interactive learning.
Quickly reviews key points and allows self-testing to identify gaps.
Helps prepare teaching materials and obtain explanations from different perspectives.
Section 07
The project is released as open-source, encouraging community contributions: educators can fine-tune it to adapt to local curricula, and developers can integrate it into educational applications. It uses open-source tools like Unsloth and llama.cpp to ensure performance and compatibility.
Section 08
The 3B-parameter model has limitations in complex reasoning and multilingual processing. It is suitable for handling NCERT-related educational tasks but performs poorly on queries outside this scope.
Expand training data to cover more subjects and grades, explore more efficient fine-tuning methods, and develop supporting UIs to lower the barrier to use.
Keep going with more reads from the same topic.
Nornir MCP Server is an enterprise-level server based on the Model Context Protocol (MCP). It seamlessly integrates large language models (such as Claude) with the Nornir network automation framework, supporting natural language orchestration for multi-vendor network devices (Cisco, Arista, Juniper, etc.), and providing production-grade features like a dual-engine architecture (NAPALM + Netmiko), intelligent filtering, and a secure sandbox.
Bibliothèque Française LLM is a structured indexing and annotation project for French public domain literature designed specifically for large language models (LLMs). It integrates multiple authoritative sources such as DraCor, Common Corpus, and Wikisource, providing metadata indexing categorized by genre, author, and era, as well as in-depth annotations for dramatic texts (including characters, lines, stage directions, etc.). Its aim is to enable LLMs to efficiently read and understand classic French literary works.
Splinter is a minimalist, high-performance key-value (KV) and vector storage system enabling zero-latency inter-process communication via shared memory and atomic operations. With only 766 lines of core code, it supports millions of operations per second and 768-dimensional vector storage, offering a new architectural approach for local LLM inference and data-intensive applications.
Folkering OS is the world's first AI-native bare-metal operating system, entirely written in Rust no_std without relying on Linux, POSIX, or libc. It can generate commands from scratch, compile them into WASM, and run them in 10 seconds, achieving true self-evolution.