berlino

llama.cpp

LLM inference in C/C++

tinygrad

You like pytorch? You like micrograd? You love tinygrad! ❤️

vllm

A high-throughput and memory-efficient inference and serving engine for LLMs

shap-e

Generate 3D objects conditioned on text or images

mistral-src

Reference implementation of Mistral AI 7B v0.1 model.

llama-recipes

Scripts for fine-tuning Llama2 with composable FSDP & PEFT methods to cover single/multi-node GPUs. Supports default & custom datasets for applications such as summarization & question answering. Supporting a number of candid inference solutions such as HF TGI, VLLM for local or cloud deployment.Demo apps to showcase Llama2 for WhatsApp & Messenger

open_llama

OpenLLaMA, a permissively licensed open source reproduction of Meta AI’s LLaMA 7B trained on the RedPajama dataset

gpt-fast

Simple and efficient pytorch-native transformer text generation in <1000 LOC of python.

GPU-Puzzles

Solve puzzles. Learn CUDA.

cutlass

CUDA Templates for Linear Algebra Subroutines

keops

KErnel OPerationS, on CPUs and GPUs, with autodiff and without memory overflows

LookaheadDecoding

safari

Convolutions for Sequence Modeling

m2

Repo for "Monarch Mixer: A Simple Sub-Quadratic GEMM-Based Architecture"

H3

Language Modeling with the H3 State Space Model

rellm

Exact structure out of any language model completion.

attention_with_linear_biases

Code for the ALiBi method for transformer language models (ICLR 2022)

SGEMM_CUDA

Fast CUDA matrix multiplication from scratch

liquid-s4

Liquid Structural State-Space Models

GenSim

GenSim: Generating Robotic Simulation Tasks via Large Language Models

ModuleFormer

ModuleFormer is a MoE-based architecture that includes two different types of experts: stick-breaking attention heads and feedforward experts. We released a collection of ModuleFormer-based Language Models (MoLM) ranging in scale from 4 billion to 8 billion parameters.

llm_large_context

Large Sequence Modeling with Transformers

zoology

Understand and test language model architectures on synthetic tasks.

BiGS

Official Repository of Pretraining Without Attention (BiGS), BiGS is the first model to achieve BERT-level transfer learning on the GLUE benchmark with subquadratic complexity in length (or without attention).

explain-then-translate

Official repo for EMNLP 2023 paper "Explain-then-Translate: An Analysis on Improving Program Translation with Self-generated Explanations"

ConvRe

🤖ConvRe🤯: An Investigation of LLMs’ Inefficacy in Understanding Converse Relations (EMNLP 2023)

pytorch_linear_rnn

Implementations of various linear RNN layers using pytorch and triton

Logical-and-abstract-reasoning

Evaluation on Logical Reasoning and Abstract Reasoning Challenges

pyccg

A Python library for distantly supervised CCG parsing/learning, based off of NLTK's CCG implementation

rnn_typology