xla

High-performance, branchless numerical stability kernels and compiler-optimized core infrastructure for advanced JAX/XLA deep learning architectures.

A Bare-Metal PIM-HBM Hardware Co-Design Core Engine exploring Register-Level MUX Virtualization to mitigate NCCL All-to-All communication stalls.

A dependency-free Python client for the PJRT C API and the XLA FFI. Run compiled ML programs on CPUs, GPUs and TPUs from pure Python.

A JAX XLA-powered PoC that leverages branchless mathematical primitives to bypass the memory and execution bottlenecks of LLM softmax operations

Packet-Switched Attention for stable 2-bit quantized MoE inference, with variance-aware routing and Protocol C benchmarks.

A Multivariate Gaussian Bayes classifier written using JAX

Pure JAX implementations of modern deep learning architectures in a minimalist notebook implementation designed for clarity, mathematical symmetry, and high-performance compute.

不经过 libtpu/PJRT 的 TPU v4 最小用户态 runtime

Workshop given in graduate-level thin film coatings course in ITU

Multi-Engine (PyTorch & JAX/XLA) Zero-Branching Geometric Acceleration Core. Enforces 0% Graph Breaks & Real-time Fault-Isolation via hardware-native bitwise MUX operations (torch.where / jax.lax.select) to permanently eliminate 'jmp' instructions and host-device synchronization fences.

PJRT plugin runtimes

Woboq codebrowser for Tensorflow v1.8 with XLA Enabled

Classification of multilingual dataset trained only on English training data using pre-trained models. Model is trained on TPUs using PyTorch and torch_xla library.

JAX Foreign Function Interface experiments

High-performance TensorFlow library for quantitative finance.

Enabling PyTorch on XLA Devices (e.g. Google TPU)

Blueprint for a decentralized, fault-tolerant Surface Code infrastructure leveraging branchless C99 ancilla-syndrome registers, zero-copy C++ binders, and JAX/XLA gradient isolation gates to bypass classical decoding latency walls.

Tensor Logic as a StableHLO-emitting DSL for PJRT runtimes

As the quality of large language models increases, so do our expectations of what they can do. Since the release of OpenAI's GPT-2, text generation capabilities have received attention. And for good reason - these models can be used for summarization, translation, and even real-time learning in some language tasks.

SEHE(Son Ho-Sung Equation for Harmony Entropy) Framework: A Thermodynamic Meta-Cognitive Engine for LLMs