Skip to content

Latest commit

 

History

History
198 lines (137 loc) · 5.48 KB

README.md

File metadata and controls

198 lines (137 loc) · 5.48 KB
Raw

CoT-eval

A framework for evaluating the effectiveness of chain-of-thought reasoning in language models.

🔥 Open CoT Leaderboard | 🔥 Results Exploration (Notebook)

Table of Contents

Goal

Set up a pipeline (and provide missing parts) to evaluate the effectiveness of chain-of-thought reasoning (COT) in language models.

Pipeline

COT-eval is intended to be used in conjunction with Eleuther's lm-evaluation-harness (or similiar packages, such as catwalk) to assess a model's ability to generate high quality (i.e., effective) chain-of-thought reasoning traces.

The pipeline is as follows:

  1. Specify an eval configuration, including
    • model: the model to evaluate (e.g. mistralai/Mistral-7B-Instruct-v0.2)
    • task: the task to evaluate on (logiqa, lsat)
    • chain: the prompt chain used to generate the reasoning traces
    • decoding: the decoding strategy and parameters to use for reasoning (beam search, temperature, etc.)
  2. Pertubate the task. (Because of potential training data contamination.)
  3. Run cot-eval to generate the reasoning traces with the model (and according to the configuration) for the perturbated task. (Push reasoning traces to HF hub.)
  4. Run lm-evaluation-harness to evaluate the model on the original task. This gives us scores-1.
  5. Run lm-evaluation-harness to evaluate the model on the perturbated task. This gives us scores-2.
  6. Run lm-evaluation-harness to evaluate the model on the perturbated task with added reasoning traces. This gives us scores-3.
  7. Conclude:
    • The difference between scores-1 and scores-2 is an indicator of training data contamination.
    • The difference between scores-2 and scores-3 is an indicator of COT effectiveness, i.e. the model's reasoning skill.

Installation

git clone https://github.com/logikon-ai/cot-eval.git
cd cot-eval
pip install -e ".[cuda]"

Usage

Note

Use a personal HUGGINGFACEHUB_API_TOKEN. Note that you have to be a member of the Open CoT Leaderboard for this to work.

See run.sh for an implementation of the pipeline.

cot-eval --help

With Docker 🐳

Step 1. Clone cot-eval repo.

git clone https://github.com/logikon-ai/cot-eval.git
cd cot-eval

Step 2. Pull docker image

docker pull logikon/cot-eval:latest

Step 2a. (Alternatively:) Build docker image locally (allows you to adapt build args, e.g. VLLM_VERSION)

docker build --no-cache -t cot-eval --build-arg="VLLM_VERSION=0.3.0" . # change vllm version if necessary

Step 3. Set parameters and arguments

vim config.env  # adapt config.env, set especially NEXT_MODEL_PATH="..." and HUGGINGFACEHUB_API_TOKEN="..."

Step 4. Run docker container

cat config.env  # check
docker run -it --rm --gpus all --ipc=host --shm-size=1g --ulimit memlock=-1 --ulimit stack=67108864 --env-file config.env logikon/cot-eval:latest

With Enroot

# export TMPDIR=...
cd $TMPDIR

git clone https://github.com/logikon-ai/cot-eval.git
# edit config
vim cot-eval/config.env

export ENROOT_DATA_PATH=$TMPDIR/enroot-data
mkdir $ENROOT_DATA_PATH
export ENROOT_CONFIG_PATH=$TMPDIR/enroot-config
mkdir $ENROOT_CONFIG_PATH
touch $ENROOT_CONFIG_PATH/enroot.config
mkdir $ENROOT_CONFIG_PATH/environ.d
cp cot-eval/config.env $ENROOT_CONFIG_PATH/environ.d

enroot import docker://logikon/cot-eval
enroot create --name cot-eval logikon+cot-eval.sqsh
rm logikon+cot-eval.sqsh

enroot start --rw cot-eval

Alternatively:

ENROOT_SQUASH_OPTIONS='-comp lz4 -noD' enroot import docker://logikon/cot-eval
enroot start --rw logikon+cot-eval.sqsh

With Slurm / Apptainer

We're using the following slurm on booster:

#!/bin/bash -x
#SBATCH --account=<PROJECT_ID>
#SBATCH --nodes=1
#SBATCH --ntasks=4
#SBATCH --ntasks-per-node=4
#SBATCH --output=gpu-out.%j
#SBATCH --error=gpu-err.%j
#SBATCH --time=12:00:00
#SBATCH --partition=booster
#SBATCH --gres=gpu:1

jutil env activate -p <PROJECT_ID>

# create tmp folder to bind with container
mkdir -p $SCRATCH/$SLURM_JOB_ID

apptainer run \
  --nv \
  --env HF_HOME=/mnt/cache/huggingface \
  --env-file $PROJECT/config.env \
  --no-mount home,cwd \
  --bind $SCRATCH/$SLURM_JOB_ID:/mnt \
  --containall \
  $PROJECT/cot-eval.sif bash -c "mkdir /mnt/cache;mkdir /mnt/cache/huggingface;cd /workspace/cot-eval;bash run.sh"

Misc

Build and push Docker image

git clone https://github.com/logikon-ai/cot-eval.git
cd cot-eval
docker build --no-cache -t cot-eval . 
docker login --username logikon
docker tag cot-eval logikon/cot-eval:latest
docker push logikon/cot-eval:latest

🙏 Built with

License

cot-eval is distributed under the terms of the MIT license.