Triton
PulseAugur coverage of Triton — every cluster mentioning Triton across labs, papers, and developer communities, ranked by signal.
16 day(s) with sentiment data
-
OpenAI cuts GPT-5.6 prices by up to 80% with AI-driven optimizations
OpenAI has significantly reduced the prices for its GPT-5.6 model family, with the Luna tier now 80% cheaper and the Terra tier 20% less expensive. This cost reduction was largely driven by engineering optimizations, in…
-
Rust and CUDA C++ outperform Triton on irregular GPU workloads
A new research paper compares the performance of CUDA C++, Rust, and Triton for GPU workloads, particularly focusing on irregular operations like hash table insertions. The study found that while all three languages per…
-
Attn-QAT enables stable 4-bit attention training for LLMs
Researchers have developed Attn-QAT, a novel method for 4-bit quantization-aware training of attention mechanisms in large language models. This approach addresses the challenges of low precision in FP4 computation, par…
-
New linearized attention model achieves higher accuracy and lower perplexity
Researchers have developed a linearized version of 2-simplicial attention, which rewrites the trilinear score into an inner product. This new form allows for linear cost in sequence length while maintaining global reach…
-
New method retrofits linear attention to speed up diffusion language models
Researchers have developed a method to retrofit linear attention into diffusion language models (dLLMs) to accelerate inference. This new approach, called block-hybrid attention, combines exact softmax attention within …
-
Open-source Triton driver enables DirectX 11 on QEMU VMs
A new open-source driver called Triton has been developed that enables DirectX 11 to run within QEMU virtual machines. This advancement is part of a broader trend towards integrating AI into the development process, wit…
-
Triton driver brings DirectX 11 support to QEMU emulator
A new driver named Triton has been developed to bring DirectX 11 compatibility to QEMU, a versatile emulator and virtualizer. This driver aims to enable broader graphics support across various operating systems, includi…
-
Triton interpreter bug fixed: NaN handling in argmin/argmax corrected
A bug in Triton's interpreter mode caused discrepancies in the behavior of `tl.argmin` and `tl.argmax` functions when handling NaN values. Unlike the compiled mode, the interpreter incorrectly allowed NaN to be consider…
-
Moonshot AI's Kimi K3 generates macOS desktop, designs chip
Moonshot AI has released Kimi K3, a 2.8 trillion-parameter open-weight model with a one-million-token context window. This model is capable of complex, long-horizon tasks, including autonomously generating a functional …
-
Webb Telescope Reveals Signs of Catastrophic Past for Neptune's Moons
The James Webb Space Telescope has provided new insights into the history of Neptune's moons, suggesting a past catastrophic event. Observations indicate that several of Neptune's moons may have been significantly alter…
-
AI systems optimize GPU kernel performance for scientific computing
Researchers have developed two novel systems, SparseDitto and KernelBrain, aimed at optimizing GPU kernel performance for various computational tasks. SparseDitto utilizes an LLM-based agent to generate custom GPU kerne…
-
AMD V620 GPU successfully integrated with ComfyUI for AI tasks
A user successfully integrated an AMD V620 workstation card into their ComfyUI setup for Stable Diffusion, despite initial skepticism about its performance. While not fast, the card, purchased for its 32GB of VRAM, func…
-
OpenAI's GPT-5.6 actively optimizes its own systems in production
OpenAI has revealed that its GPT-5.6 model is being used in production to optimize its own systems, a process described as recursive self-improvement (RSI). The model analyzes traffic, reroutes requests, and even rewrit…
-
New benchmarks and optimization tools emerge for LLM-generated hardware kernels
Two new research papers introduce benchmarks and optimization frameworks for large language models (LLMs) generating code for hardware accelerators. The first paper, KernelGenBench, offers a unified benchmark to evaluat…
-
TRELLIS.2 INT8 ConvRot model runs natively on AMD RX 7900 XTX via ComfyUI
A developer has released a patch kit enabling the TRELLIS.2 INT8 ConvRot model to run natively on an AMD RX 7900 XTX GPU using ComfyUI. This implementation utilizes fused Triton kernels for improved performance, achievi…
-
Triton backend accelerates Falcon3-10B inference by 9.8x on RTX 5070
A developer has created a new Triton backend for the Falcon3-10B-1.58bit model, significantly boosting inference speed on an NVIDIA RTX 5070 GPU. The new backend achieved 97.5 tokens per second for decoding, a nearly 10…
-
Kimi K3 AI model optimizes GPUs, designs chips, and accelerates research
Kimi K3, an advanced AI model from Open Frontier Intelligence, has demonstrated remarkable capabilities beyond typical language tasks. The model has shown proficiency in optimizing GPU kernels, improving learning speeds…
-
Production RAG Systems: A Practical Guide to Data Indexing and Retrieval
This guide details the engineering challenges and best practices for deploying Retrieval-Augmented Generation (RAG) systems in production. It covers critical aspects such as data indexing with advanced chunking strategi…
-
FlashPDE library accelerates neural PDE solvers with fused Triton operators
Researchers have developed FlashPDE, a new library of fused Triton operators designed to accelerate the training of physics-informed neural networks (PINNs) for solving partial differential equations (PDEs). This librar…
-
Kernelized Linear Attention Boosts Model Capacity and Throughput
Researchers have developed Kernelized Linear Attention Activations (KATA), a new framework designed to overcome the capacity limitations of linear attention models. KATA utilizes symmetric cones and rank-one positive se…