Edward-Sun

mlc-llm

Universal LLM Deployment Engine with ML Compilation

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

DeepSeek-Coder

DeepSeek Coder: Let the Code Write Itself

gpt-fast

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

AutoGPTQ

An easy-to-use LLMs quantization package with user-friendly apis, based on GPTQ algorithm.

alphageometry

Eureka

Official Repository for "Eureka: Human-Level Reward Design via Coding Large Language Models" (ICLR 2024)

weak-to-strong

FlexFlow

FlexFlow Serve: Low-Latency, High-Performance LLM Serving

OpenRLHF

An Easy-to-use, Scalable and High-performance RLHF Framework (Support 70B+ full tuning & LoRA & Mixtral & KTO)

LLMTest_NeedleInAHaystack

Doing simple retrieval from LLM models at various context lengths to measure accuracy

hallucination-leaderboard

Leaderboard Comparing LLM Performance at Producing Hallucinations when Summarizing Short Documents

dolma

Data and tools for generating and inspecting OLMo pre-training data.

factool

FacTool: Factuality Detection in Generative AI

PiPPy

Pipeline Parallelism for PyTorch

representation-engineering

Representation Engineering: A Top-Down Approach to AI Transparency

gpt_paper_assistant

GPT4 based personalized ArXiv paper assistant bot

weatherbench2

A benchmark for the next generation of data-driven global weather models.

JudgeLM

An open-sourced LLM judge for evaluating LLM-generated answers.

UltraFeedback

A large-scale, fine-grained, diverse preference dataset (and models).

LLaVA-RLHF

Aligning LMMs with Factually Augmented RLHF

LRV-Instruction

[ICLR'24] Mitigating Hallucination in Large Multi-Modal Models via Robust Instruction Tuning

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.

Copyisallyouneed

[ICLR 2023] Codebase for Copy-Generator model, including an implementation of kNN-LM

SALMON

Self-Alignment with Principle-Following Reward Models

scibench

triton-autodiff

Experiment of using Tangent to autodiff triton

miniF2F

An updated version of miniF2F with lots of fixes and informal statements / solutions.

CL-LNS

Code repo for ICML'23 Searching Large Neighborhoods for Integer Linear Programs with Contrastive Learning

Accelerating-Diffusion-based-Combinatorial-Optimization-Solvers-by-Progressive-Distillation

Code for Accelerating Diffusion-based Combinatorial Optimization Solvers by Progressive Distillation