• Stars
    star
    6,268
  • Rank 6,374 (Top 0.2 %)
  • Language
    Python
  • License
    MIT License
  • Created about 4 years ago
  • Updated 3 months ago

Reviews

There are no reviews yet. Be the first to send feedback to the community and the maintainers!

Repository Details

A framework for few-shot evaluation of language models.

Language Model Evaluation Harness

We're Refactoring LM-Eval!

(as of 6/15/23) We have a revamp of the Evaluation Harness library internals staged on the big-refactor branch! It is far along in progress, but before we start to move the master branch of the repository over to this new design with a new version release, we'd like to ensure that it's been tested by outside users and there are no glaring bugs.

Weโ€™d like your help to test it out! you can help by:

  1. Trying out your current workloads on the big-refactor branch, and seeing if anything breaks or is counterintuitive,
  2. Porting tasks supported in the previous version of the harness to the new YAML configuration format. Please check out our task implementation guide for more information.

If you choose to port a task not yet completed according to our checklist, then you can contribute it by opening a PR containing [Refactor] in the name with:

  • A shell command to run the task in the master branch, and what the score is
  • A shell command to run the task in your PR branch to big-refactor, and what the resulting score is, to show that we achieve equality between the two implementations.

Lastly, we'll no longer be accepting new feature requests beyond those that are already open to the master branch as we carry out this switch to the new version over the next week, though we will be accepting bugfixes to master branch and PRs to big-refactor. Feel free to reach out in the #lm-thunderdome channel of the EAI discord for more information.

Overview

This project provides a unified framework to test generative language models on a large number of different evaluation tasks.

Features:

  • 200+ tasks implemented. See the task-table for a complete list.
  • Support for models loaded via transformers (including quantization via AutoGPTQ), GPT-NeoX, and Megatron-DeepSpeed, with a flexible tokenization-agnostic interface.
  • Support for commercial APIs including OpenAI, goose.ai, and TextSynth.
  • Support for evaluation on adapters (e.g. LoRa) supported in HuggingFace's PEFT library.
  • Evaluating with publicly available prompts ensures reproducibility and comparability between papers.
  • Task versioning to ensure reproducibility when tasks are updated.

Install

To install lm-eval from the github repository main branch, run:

git clone https://github.com/EleutherAI/lm-evaluation-harness
cd lm-evaluation-harness
pip install -e .

To install additional multilingual tokenization and text segmentation packages, you must install the package with the multilingual extra:

pip install -e ".[multilingual]"

To support loading GPTQ quantized models, install the package with the auto-gptq extra:

pip install -e ".[auto-gptq]"

Basic Usage

Note: When reporting results from eval harness, please include the task versions (shown in results["versions"]) for reproducibility. This allows bug fixes to tasks while also ensuring that previously reported scores are reproducible. See the Task Versioning section for more info.

Hugging Face transformers

To evaluate a model hosted on the HuggingFace Hub (e.g. GPT-J-6B) on hellaswag you can use the following command:

python main.py \
    --model hf-causal \
    --model_args pretrained=EleutherAI/gpt-j-6B \
    --tasks hellaswag \
    --device cuda:0

Additional arguments can be provided to the model constructor using the --model_args flag. Most notably, this supports the common practice of using the revisions feature on the Hub to store partially trained checkpoints, or to specify the datatype for running a model:

python main.py \
    --model hf-causal \
    --model_args pretrained=EleutherAI/pythia-160m,revision=step100000,dtype="float" \
    --tasks lambada_openai,hellaswag \
    --device cuda:0

To evaluate models that are loaded via AutoSeq2SeqLM in Huggingface, you instead use hf-seq2seq. To evaluate (causal) models across multiple GPUs, use --model hf-causal-experimental

Warning: Choosing the wrong model may result in erroneous outputs despite not erroring.

Commercial APIs

Our library also supports language models served via the OpenAI API:

export OPENAI_API_SECRET_KEY=YOUR_KEY_HERE
python main.py \
    --model gpt3 \
    --model_args engine=davinci \
    --tasks lambada_openai,hellaswag

While this functionality is only officially maintained for the official OpenAI API, it tends to also work for other hosting services that use the same API such as goose.ai with minor modification. We also have an implementation for the TextSynth API, using --model textsynth.

To verify the data integrity of the tasks you're performing in addition to running the tasks themselves, you can use the --check_integrity flag:

python main.py \
    --model gpt3 \
    --model_args engine=davinci \
    --tasks lambada_openai,hellaswag \
    --check_integrity

Other Frameworks

A number of other libraries contain scripts for calling the eval harness through their library. These include GPT-NeoX, Megatron-DeepSpeed, and mesh-transformer-jax.

๐Ÿ’ก Tip: You can inspect what the LM inputs look like by running the following command:

python write_out.py \
    --tasks all_tasks \
    --num_fewshot 5 \
    --num_examples 10 \
    --output_base_path /path/to/output/folder

This will write out one text file for each task.

Advanced Usage

For models loaded with the HuggingFace transformers library, any arguments provided via --model_args get passed to the relevant constructor directly. This means that anything you can do with AutoModel can be done with our library. For example, you can pass a local path via pretrained= or use models finetuned with PEFT by taking the call you would run to evaluate the base model and add ,peft=PATH to the model_args argument:

python main.py \
    --model hf-causal-experimental \
    --model_args pretrained=EleutherAI/gpt-j-6b,peft=nomic-ai/gpt4all-j-lora \
    --tasks openbookqa,arc_easy,winogrande,hellaswag,arc_challenge,piqa,boolq \
    --device cuda:0

GPTQ quantized models can be loaded by specifying their file names in ,quantized=NAME (or ,quantized=True for default names) in the model_args argument:

python main.py \
    --model hf-causal-experimental \
    --model_args pretrained=model-name-or-path,quantized=model.safetensors,gptq_use_triton=True \
    --tasks hellaswag

We support wildcards in task names, for example you can run all of the machine-translated lambada tasks via --task lambada_openai_mt_*.

We currently only support one prompt per task, which we strive to make the "standard" as defined by the benchmark's authors. If you would like to study how varying prompts causes changes in the evaluation score, check out the BigScience fork of this repo. We are currently working on upstreaming this capability to main.

Implementing new tasks

To implement a new task in the eval harness, see this guide.

Task Versioning

To help improve reproducibility, all tasks have a VERSION field. When run from the command line, this is reported in a column in the table, or in the "version" field in the evaluator return dict. The purpose of the version is so that if the task definition changes (i.e to fix a bug), then we can know exactly which metrics were computed using the old buggy implementation to avoid unfair comparisons. To enforce this, there are unit tests that make sure the behavior of all tests remains the same as when they were first implemented. Task versions start at 0, and each time a breaking change is made, the version is incremented by one.

When reporting eval harness results, please also report the version of each task. This can be done either with a separate column in the table, or by reporting the task name with the version appended as such: taskname-v0.

Test Set Decontamination

To address concerns about train / test contamination, we provide utilities for comparing results on a benchmark using only the data points not found in the model training set. Unfortunately, outside of models trained on the Pile and C4, its very rare that people who train models disclose the contents of the training data. However this utility can be useful to evaluate models you have trained on private data, provided you are willing to pre-compute the necessary indices. We provide computed indices for 13-gram exact match deduplication against the Pile, and plan to add additional precomputed dataset indices in the future (including C4 and min-hash LSH deduplication).

For details on text decontamination, see the decontamination guide.

Note that the directory provided to the --decontamination_ngrams_path argument should contain the ngram files and info.json. See the above guide for ngram generation for the pile, this could be adapted for other training sets.

python main.py \
    --model gpt2 \
    --tasks sciq \
    --decontamination_ngrams_path path/containing/training/set/ngrams \
    --device cuda:0

Cite as

@software{eval-harness,
  author       = {Gao, Leo and
                  Tow, Jonathan and
                  Biderman, Stella and
                  Black, Sid and
                  DiPofi, Anthony and
                  Foster, Charles and
                  Golding, Laurence and
                  Hsu, Jeffrey and
                  McDonell, Kyle and
                  Muennighoff, Niklas and
                  Phang, Jason and
                  Reynolds, Laria and
                  Tang, Eric and
                  Thite, Anish and
                  Wang, Ben and
                  Wang, Kevin and
                  Zou, Andy},
  title        = {A framework for few-shot language model evaluation},
  month        = sep,
  year         = 2021,
  publisher    = {Zenodo},
  version      = {v0.0.1},
  doi          = {10.5281/zenodo.5371628},
  url          = {https://doi.org/10.5281/zenodo.5371628}
}

More Repositories

1

gpt-neo

An implementation of model parallel GPT-2 and GPT-3-style models using the mesh-tensorflow library.
Python
8,224
star
2

gpt-neox

An implementation of model parallel autoregressive transformers on GPUs, based on the Megatron and DeepSpeed libraries
Python
6,829
star
3

pythia

The hub for EleutherAI's work on interpretability and learning dynamics
Jupyter Notebook
2,193
star
4

the-pile

Python
1,459
star
5

math-lm

Python
1,035
star
6

cookbook

Deep learning for dummies. All the practical details and useful utilities that go into working with real models.
Python
635
star
7

polyglot

Polyglot: Large Language Models of Well-balanced Competence in Multi-languages
471
star
8

DALLE-mtf

Open-AI's DALL-E for large scale training in mesh-tensorflow.
Python
434
star
9

vqgan-clip

Jupyter Notebook
345
star
10

sae

Sparse autoencoders
Python
274
star
11

concept-erasure

Erasing concepts from neural representations with provable guarantees
Python
207
star
12

elk

Keeping language models honest by directly eliciting knowledge encoded in their activations.
Python
186
star
13

oslo

OSLO: Open Source for Large-scale Optimization
Python
173
star
14

lm_perplexity

Python
144
star
15

knowledge-neurons

A library for finding knowledge neurons in pretrained transformer models.
Python
142
star
16

pyfra

Python Research Framework
Python
107
star
17

dps

Data processing system for polyglot
Python
88
star
18

openwebtext2

Python
86
star
19

info

(Deprecated) A hub for onboarding & other information.
78
star
20

improved-t5

Experiments for efforts to train a new and improved t5
Python
76
star
21

stackexchange-dataset

Python tools for processing the stackexchange data dumps into a text dataset for Language Models
Python
73
star
22

project-menu

See the issue board for the current status of active and prospective projects!
65
star
23

magiCARP

One stop shop for all things carp
Python
58
star
24

sae-auto-interp

Python
53
star
25

semantic-memorization

Jupyter Notebook
44
star
26

tqdm-multiprocess

Using queues, tqdm-multiprocess supports multiple worker processes, each with multiple tqdm progress bars, displaying them cleanly through the main process. It offers similar functionality for python logging.
Python
41
star
27

aria

Python
37
star
28

hae-rae

32
star
29

rnngineering

Engineering the state of RNN language models (Mamba, RWKV, etc.)
Jupyter Notebook
31
star
30

features-across-time

Understanding how features learned by neural networks evolve throughout training
Python
30
star
31

mp_nerf

Massively-Parallel Natural Extension of Reference Frame
Jupyter Notebook
29
star
32

elk-generalization

Investigating the generalization behavior of LM probes trained to predict truth labels: (1) from one annotator to another, and (2) from easy questions to hard
Python
23
star
33

pile-pubmedcentral

A script for collecting the PubMed Central dataset in a language modelling friendly format.
Python
22
star
34

best-download

URL downloader supporting checkpointing and continuous checksumming.
Python
19
star
35

polyglot-data

data related codebase for polyglot project
Python
19
star
36

aria-amt

Efficient and robust implementation of seq-to-seq automatic piano transcription.
Python
18
star
37

text-generation-testing-ui

Web app for demoing the EAI models
JavaScript
16
star
38

exploring-contrastive-topology

Jupyter Notebook
16
star
39

mdl

Minimum Description Length probing for neural network representations
Python
15
star
40

pile_dedupe

Pile Deduplication Code
Python
15
star
41

w2s

Python
15
star
42

pilev2

Python
13
star
43

distilling

Experiments with distilling large language models.
Python
13
star
44

tokengrams

Efficiently computing & storing token n-grams from large corpora
Rust
13
star
45

lm-eval2

Python
11
star
46

equivariance

A framework for implementing equivariant DL
Jupyter Notebook
10
star
47

radioactive-lab

Adapting the "Radioactive Data" paper to work for text models
Python
9
star
48

pile-literotica

Download, parse, and filter data from Literotica. Data-ready for The-Pile.
Python
8
star
49

hn-scraper

Python
8
star
50

tagged-pile

Part-of-Speech Tagging for the Pile and RedPajama
Python
8
star
51

multimodal-fid

Python
7
star
52

pile-uspto

A script for collecting the USPTO Backgrounds dataset in a language modelling friendly format.
Python
7
star
53

pile-cc-filtering

The code used to filter CC data for The Pile
Python
6
star
54

minetest-baselines

Baseline agents for Minetest tasks.
Python
6
star
55

CodeCARP

Data collection pipeline for CodeCARP. Includes PyCharm plugins.
6
star
56

pile-enron-emails

A script for collecting the Enron Emails dataset in a language modelling friendly format.
Python
6
star
57

pile-explorer

For exploring the data and documenting its limitations
Python
5
star
58

minetest-interpretabilty-notebook

Jupyter notebook for the interpretablity section of the minetester blog post
Jupyter Notebook
5
star
59

thonkenizers

yes
5
star
60

eleutherai.github.io

This is the Hugo generated website for eleuther.ai. The source of this build is new-website repo.
HTML
5
star
61

visual-grounding

Visually ground GPT-Neo 1.3b and 2.7b
Python
5
star
62

LLM-Markov-Chains

Project github for LLM Markov Chains Project
5
star
63

architecture-experiments

Repository to host architecture experiments and development using Paxml and Praxis
Python
5
star
64

llemma-sample-explorer

Sample explorer tool for the Llemma models.
HTML
5
star
65

lm-scope

Jupyter Notebook
4
star
66

latent-video-diffusion

Latent video diffusion
Python
4
star
67

megatron-3d

Python
4
star
68

website

New website for EleutherAI based on Hugo static site generator
HTML
4
star
69

Unpaired-Image-Generation

Project Repo for Unpaired Image Generation project
4
star
70

ccs

Python
4
star
71

isaac-mchorse

EleutherAI's discord bot
Python
3
star
72

pile-allpoetry

Scraper to gather poems from allpoetry.com
Python
3
star
73

EvilModel

A replication of "EvilModel 2.0: Bringing Neural Network Models into Malware Attacks"
3
star
74

eai-prompt-gallery

Library of interesting prompt generations
JavaScript
3
star
75

variance-across-time

Studying the variance in neural net predictions across training time
Python
3
star
76

pile-ubuntu-irc

A script for collecting the Ubuntu IRC dataset in a language modelling friendly format.
Python
3
star
77

reddit-comment-processing

Python
2
star
78

eleutherai-instruct-dataset

A large instruct dataset for open-source models (WIP).
2
star
79

bucket-cleaner

A small utility to clear out old model checkpoints in Google Cloud Buckets whilst keeping tensorboard event files
Python
2
star
80

groupoid-rl

Jupyter Notebook
2
star
81

equinox-llama

Equinox implementation of llama3 and llama3.1
Python
2
star
82

optax-galore

Adds GaLore style projection wrappers to optax optimizers
Python
2
star
83

lang-filter

Filter text files or archives by language
Python
1
star
84

eleuther-blog

here is the generated content for the EleutherAI blog. Source is from new-website repo
HTML
1
star
85

prefix-free-tokenizer

A prefix free tokenizer
Python
1
star
86

alignment-reader

Search and filter through alignment literature
JavaScript
1
star
87

grouch

HTML
1
star
88

language-adaptation

1
star
89

perceptors

central location for access to pretrained models for CLIP and variants, with common API and out-of-the-box differentiable weighted multi-perceptor
1
star
90

pd-books

Jupyter Notebook
1
star
91

classifier-latent-diffusion

Python
1
star
92

common-llm-settings

Common LLM Settings App
JavaScript
1
star
93

bayesian-adam

Exactly what it says on the tin
Python
1
star
94

pile-cord19

A script for collecting the CORD-19 dataset in a language modelling friendly format.
Python
1
star
95

conceptual-constraints

Applying LEACE to models during training
Jupyter Notebook
1
star
96

ngrams-across-time

Jupyter Notebook
1
star
97

steering-llama3

Python
1
star
98

truncated-gaussian

Method-of-moments estimation and sampling for truncated multivariate Gaussian distributions
Python
1
star