Simu's Projects
Transform JSONL Q&A datasets to instruct format with ease
Quantize ONNX models to lower precision using ONNX Runtime + Gradio.
GPT-4 utilities with 👀
Custom gradient ascent solver (optimizer) for JAX/Flax models
Jax implementation of "Griffin: Mixing Gated Linear Recurrences with Local Attention for Efficient Language Models"
Human & Assistant is a very simple library created to make transformations on conversational datasets for fine-tuning chat templates, compatibility with 🤗 Datasets.
jax-triton contains integrations between JAX and OpenAI Triton
Jupyter Lab Python KF notebook environment configurations
PyKeops Powered K-Means Clustering Algorithms Module both on CPU & GPU
A tiny CLI wrapper around lmppl for Pre-Trained Language Models Perplexity Calculation for CSV files
Jax/Flax/Linen implementation of "Simple Hardware-Efficient Long Convolutions for Sequence Modeling"
Modificated DeepSpeed training setup fork of MEGABYTE - PyTorch by lucidrains, Predicting Million-byte Sequences with Multiscale Transformers, in Pytorch
Dataset of formal Olympiad-level mathematics problems solved with Python code instructions.
MPI Operator DeepSpeed Base Configuration for CIFAR-10
Nature Biosciences Research Paper Abstract Embeddings with Sentence Transformers
Modificated DeepSpeed training setup fork of RLHF (Reinforcement Learning with Human Feedback) by lucidrains on top of the PaLM architecture. Basically ChatGPT but with PaLM
Python Package Template is all you need
Tensors and Dynamic neural networks in Python with strong GPU acceleration
Packaged Ring Attention with Blockwise Transformers for Near-Infinite Context implemented in Jax + Flax.
A tiny package provides distance computations for PyTorch functions.
A simpler Pytorch + Zeta Implementation of the paper: "SiMBA: Simplified Mamba-based Architecture for Vision and Multivariate Time series"
blog for the AI era
Smooth ReLU activations in CUDA. Shamir, G., I. et al.
A tiny CLI tool that transforms SRTs into word-by-word timestamped SRTs.
Simple SentenceTransformer pipeline example
Efficient kernel for RMS normalization with fused operations, includes both forward and backward passes, compatibility with PyTorch.
Collection of neural network activation function kernels for Triton Language Compiler by OpenAI
Build high-performance AI models with modular building blocks