eugeneyan

DeepSpeed

DeepSpeed is a deep learning optimization library that makes distributed training and inference easy, efficient, and effective.

dspy

DSPy: The framework for programming—not prompting—foundation models

minbpe

Minimal, clean code for the Byte Pair Encoding (BPE) algorithm commonly used in LLM tokenization.

trl

Train transformer language models with reinforcement learning.

ml-ferret

bitsandbytes

Accessible large language models via k-bit quantization for PyTorch.

instructor

structured outputs for llms

captum

Model interpretability and understanding for PyTorch

trlx

A repo for distributed training of language models with Reinforcement Learning via Human Feedback (RLHF)

alignment-handbook

Robust recipes to align language models with human and AI preferences

higgsfield

Fault-tolerant, highly scalable GPU orchestration, and a machine learning framework designed for training models with billions to trillions of parameters

picoGPT

An unnecessarily tiny implementation of GPT-2 in NumPy.

CTranslate2

Fast inference engine for Transformer models

vocode-python

🤖 Build voice-based LLM agents. Modular + open source.

ChainForge

An open-source visual programming environment for battle-testing prompts to LLMs.

AutoAWQ

AutoAWQ implements the AWQ algorithm for 4-bit quantization with a 2x speedup during inference. Documentation:

mup

maximal update parametrization (µP)

lm-human-preferences

Code for the paper Fine-Tuning Language Models from Human Preferences

hallucination-leaderboard

Leaderboard Comparing LLM Performance at Producing Hallucinations when Summarizing Short Documents

summarize-from-feedback

Code for "Learning to summarize from human feedback"

obsidian-kindle-plugin

Sync your Kindle notes and highlights directly into your Obsidian vault

bleurt

BLEURT is a metric for Natural Language Generation based on transfer learning.

COMET

A Neural Framework for MT Evaluation

llm_distillation_playbook

Best practices for distilling large language models.

Whisper-transcription_and_diarization-speaker-identification-

How to use OpenAIs Whisper to transcribe and diarize audio files

n-levels-of-rag

visualizing-finetunes

ft-drift

Check for data drift between two OpenAI multi-turn chat jsonl files.

swe-study-group

Code for the SWE study group

Clautero