This code is built on top of code for "Generating with Confidence: Uncertainty Quantification for Black-box Large Language Models" arxiv: https://github.com/zlin7/UQ-NLG/tree/main.

Quick Start

1. Note that to get the automatic evaluation based on GPT, you would need to update keys.json with your API keys first.

2. Set the corresponding paths in _settings.py: these paths are where your (a) model generations (or responses), and (b) downloaded model weights of Llama-2-13b, and Mistral-7B-Instruct-v0.2 (from hugging face) will be stored.

3. Install the dependies: pip install -r requirements.txt

Generate 20 responses from the model

Use the llama-13b-hf, mistral-7b-hf for model, and coqa, triviaqa (for exact values of hyperparameters used in the paper such as temperature, top_p etc. please read the topmost comments in pipeline/generate.py):

python3 -m pipeline.generate --model llama-13b-hf --dataset coqa

The generations will be saved in the GENERATION_FOLDER/$model_dataset_seed$/$seed$.pkl.

Run UQ Experiments

Step 1: Compute scores required for UQ and GT evaluation on the generated responses

Run dataeval/load.py to cache down the results first (including GPT evaluations): load.py will run evaluations on all the paths specified in _settings.py:

python3 -m dataeval.load

Step 2: Generate baseline and our results

Run pipeline/uq_bb.py

python3 -m pipeline.uq_bb --model $llama-13b-hf/mistral-7b-hf$ --dataset $coqa/triviaqa$ --cal_size $1000 for coqa/2000 for triviaqa$ --acc_name $rougeL/deberta_entailment/gpt$ --metric $auarc/auroc/rej_acc$

Run Conformal Prediction Experiments (Getting accuracy and set sizes)

Assuming Step 1 from UQ Experiments has been done, Step 2 is to run uq_bb for generating prediction sets

python3 -m pipeline.uq_bb --model $llama-13b-hf/mistral-7b-hf$ --dataset $coqa/triviaqa$ --cal_size $1000 for coqa/2000 for triviaqa$ --acc_name $rougeL/deberta_entailment/gpt$ --output_pred_sets True

Notes on GPT evaltuaions

1. We used gpt-4-turbo-preview (which points to gpt-4-0125-preview) for GT evaluations from GPT

2. These experiments were performed using openai version 0.28. If you get this OpenAI error: "You tried to access openai.ChatCompletion, but this is no longer supported in openai>=1.0.0 - see the README at https://github.com/openai/openai-python for the API." You can resolve it by installing to the old version, e.g. pip install openai==0.28.