BITNET
PulseAugur coverage of BITNET — every cluster mentioning BITNET across labs, papers, and developer communities, ranked by signal.
- 2026-08-12 product_launch Microsoft open-sourced the BitNet inference framework for 1-bit LLMs. source
1 day(s) with sentiment data
-
Beginner's guide to LLM quantization resources compiled
A curated list of resources for beginners interested in quantization for large language models has been compiled. The list includes key papers and techniques such as BitNet, AWQ, QLoRA, and ZeroQuant, covering topics fr…
-
LoopQuant enables efficient low-bit deployment for Looped Language Models
Researchers have developed LoopQuant, a novel post-training quantization framework designed to efficiently deploy Looped Language Models (LoopLMs) at low bit-widths. Unlike existing methods that struggle with the recurs…
-
New SSR method accelerates ternary LLM inference
Researchers have developed Sparse Segment Reduction (SSR), a new method for accelerating the inference of ternary Large Language Models (LLMs). This approach optimizes matrix multiplication for ternary weights, which ar…
-
Curiosity-driven MoE framework enhances AI model stability and accuracy on edge devices
Researchers have developed a novel curiosity-driven quantized Mixture-of-Experts (MoE) framework designed for resource-constrained devices. This approach addresses challenges in maintaining accuracy under aggressive qua…
-
BitNet explores ternary weights to cut LLM memory by 10x
Researchers are exploring methods to reduce the computational and memory demands of large language models, moving beyond simply increasing model size. One promising approach, BitNet, investigates training models where w…
-
Maker builds 7-node ESP32-S3 cluster for 386M parameter LLM
A maker has successfully built a custom cluster of seven ESP32-S3 microcontrollers to run a large language model with approximately 386 million parameters. This was achieved by dividing the model's layers across the mic…
-
New low-bit and ternary AI models released with performance updates · 1 source tracked
Several new low-bit and ternary models have been released and are being tracked, including Bonsai's 1-bit and 1.58-bit (ternary) versions, with a 27B parameter model now running on mainline backends. Updates to llama.cp…
-
Ternary LLMs see resurgence with new models from smaller labs
Recent developments suggest a resurgence in ternary (1.58-bit) large language models, with several new models released by smaller labs. These include prismML's 27B ternary model, Deepgrove's 20B Maple model, and Doses A…
-
Microsoft open-sources BitNet for 1-bit LLMs on single CPUs
Microsoft has open-sourced BitNet, an inference framework designed for 1-bit Large Language Models (LLMs). This framework allows for the execution of models with up to 100 billion parameters on a single CPU, eliminating…
-
Developer builds zero-dependency C inference engine for BitNet models
A developer has created a C inference engine designed for BitNet models, focusing on zero dependencies and CPU performance. The engine achieves 36.25 tokens per second on a BitNet b1.58-2B-4T model running on an Intel X…
-
Author's early coding journey parallels current AI displacement fears
The author recounts their early experiences with programming, starting with AppleScript on Mac OS 7 at age seven and progressing to developing their own compiled programs by age eight. They detail their personal history…
-
Ternary Quantization Shrinks Super-Resolution Transformer to 668 KB for Browser Use
Researchers have successfully applied BitNet-style ternary quantization to a super-resolution transformer model, resulting in a significantly smaller model size. The quantized model, which uses weights of -1, 0, or +1, …
-
Microsoft releases VibeVoice-ASR-BitNet for real-time CPU inference
Microsoft has released VibeVoice-ASR-BitNet, a highly compressed automatic speech recognition model designed for real-time inference on edge CPUs without requiring a GPU. This model achieves significant speedups over ex…
-
New C99 inference engine Project Zero offers 1.8x speedup for BitNet models
A new, from-scratch C99 inference engine called Project Zero has been developed, offering a 1.8x speedup over bitnet.cpp for BitNet models on Xeon processors. This engine boasts zero external dependencies, running solel…
-
Negative-Bit Quantization Frees VRAM by Inverting Tensor Embeddings
A researcher has developed a novel technique called Negative-Bit Quantization (NBQ) that claims to achieve stable inference with "negative-bit" configurations, effectively freeing up VRAM. This method, termed Phase-Inve…
-
PrismML releases Bonsai 27B, enabling Qwen3.6-27B on laptops and phones
PrismML has released Bonsai 27B, a highly compressed version of Qwen3.6-27B, available in 1-bit and ternary variants. These models are designed to run on consumer hardware like laptops and phones, with the 1-bit version…
-
New BITEMBED framework drastically cuts LLM embedding costs
Researchers have developed BITEMBED, a novel framework designed to create efficient text embeddings for large language models. This approach converts LLM backbones into low-bit encoders using ternary weights and quantiz…
-
New Quantization Methods Boost LLM Efficiency and Speed
Researchers have developed CAT-Q, a novel post-training quantization method that significantly compresses and accelerates Large Language Models (LLMs) without requiring extensive retraining. This technique, which uses l…
-
New memristive synapse design promises highly efficient on-chip neural networks
Researchers have developed a physics-based design for an on-chip neural network utilizing multi-level memristive synapses capable of supporting a wide range of conductance states. This design, rooted in ionic transport …
-
Ternary LLMs stall at 2B parameters, frontier labs bypass approach
Ternary LLMs, which use a three-value system for weights, showed early promise but have not seen significant development. The largest available ternary model is only 2 billion parameters, and major AI labs have not adop…