Add theory of mind eval (#1405)

# Thank you for contributing an eval! ♥️

🚨 Please make sure your PR follows these guidelines, **failure to follow
the guidelines below will result in the PR being closed automatically**.
Note that even if the criteria are met, that does not guarantee the PR
will be merged nor GPT-4 access be granted. 🚨

**PLEASE READ THIS**:

In order for a PR to be merged, it must fail on GPT-4. We are aware that
right now, users do not have access, so you will not be able to tell if
the eval fails or not. Please run your eval with GPT-3.5-Turbo, but keep
in mind as we run the eval, if GPT-4 gets higher than 90% on the eval,
we will likely reject it since GPT-4 is already capable of completing
the task.

We plan to roll out a way for users submitting evals to see the eval
performance on GPT-4 soon. Stay tuned! Until then, you will not be able
to see the eval performance on GPT-4. **Starting April 10, the minimum
eval count is 15 samples, we hope this makes it easier to create and
contribute evals.**

Also, please note that we're using **Git LFS** for storing the JSON
files, so please make sure that you move the JSON file to Git LFS before
submitting a PR. Details on how to use Git LFS are available
[here](https://git-lfs.com).

## Eval details 📑

### Eval name

Theory of mind.

### Eval description

The `ToMi` test set contains 5,993 question-answer pairs. These are
instances of the [Sally-Anne
test](https://en.wikipedia.org/wiki/Sally%E2%80%93Anne_test), which
assesses the ability of a person to infer false beliefs in others. The
original setting involves two people, Sally and Anne, who are together
in a room. Sally places a marble in a box. Then, Anne leaves the room,
and while she is away, Sally moves the marble to a basket elsewhere in
the room. When Anne returns to the room, where will she search for the
marble? If the person responding “has” theory-of-mind they’ll respond
that Anne searches for the marble in the box, where she had last seen
it. If they do not, they ascribe their own, accurate belief regarding
the location to Anne, and say that she looks for it in the basket.

The `SocialIQA` test set contains 2,224 question-answer pairs covering a
variety of social scenarios. These are multiple-choice, with 3 options
of which only one is correct. The questions cover a person’s wants,
needs, motivations, and reactions, as well as the effects of an action
(on self or others), and how that action reflects on the person carrying
it out (e.g. how others would perceive them after having carried out the
action).

Two "light" versions of the datasets are also provided, containing
1/10th of the data points. These are useful for iterating on prompts and
developing other scaffolding.
### What makes this a useful eval?

Measures theory of mind capability in language models.

## Criteria for a good eval ✅

Below are some of the criteria we look for in a good eval. In general,
we are seeking cases where the model does not do a good job despite
being capable of generating a good response (note that there are some
things large language models cannot do, so those would not make good
evals).

Your eval should be:

- [x] Thematically consistent: The eval should be thematically
consistent. We'd like to see a number of prompts all demonstrating some
particular failure mode. For example, we can create an eval on cases
where the model fails to reason about the physical world.
- [x] Contains failures where a human can do the task, but either GPT-4
or GPT-3.5-Turbo could not.
- [x] Includes good signal around what is the right behavior. This means
either a correct answer for `Basic` evals or the `Fact` Model-graded
eval, or an exhaustive rubric for evaluating answers for the `Criteria`
Model-graded eval.
- [x] **Include at least 15 high-quality examples.**

If there is anything else that makes your eval worth including, please
document it below.

### Unique eval value

> Insert what makes your eval high quality that was not mentioned above.
(Not required)

## Eval structure 🏗️

Your eval should

- [x] Check that your data is in `evals/registry/data/{name}`
- [x] Check that your YAML is registered at
`evals/registry/evals/{name}.yaml`
- [x] Ensure you have the right to use the data you submit via this eval

(For now, we will only be approving evals that use one of the existing
eval classes. You may still write custom eval classes for your own
cases, and we may consider merging them in the future.)

## Final checklist 👀

### Submission agreement

By contributing to Evals, you are agreeing to make your evaluation logic
and data under the same MIT license as this repository. You must have
adequate rights to upload any data used in an Eval. OpenAI reserves the
right to use this data in future service improvements to our product.
Contributions to OpenAI Evals will be subject to our usual Usage
Policies (<https://platform.openai.com/docs/usage-policies>).

- [x] I agree that my submission will be made available under an MIT
license and complies with OpenAI's usage policies.

### Email address validation

If your submission is accepted, we will be granting GPT-4 access to a
limited number of contributors. Access will be given to the email
address associated with the commits on the merged pull request.

- [x] I acknowledge that GPT-4 access will only be granted, if
applicable, to the email address used for my merged pull request.

### Limited availability acknowledgment

We know that you might be excited to contribute to OpenAI's mission,
help improve our models, and gain access to GPT-4. However, due to the
requirements mentioned above and the high volume of submissions, we will
not be able to accept all submissions and thus not grant everyone who
opens a PR GPT-4 access. We know this is disappointing, but we hope to
set the right expectation before you open this PR.

- [x] I understand that opening a PR, even if it meets the requirements
above, does not guarantee the PR will be merged nor GPT-4 access be
granted.

### Submit eval

- [x] I have filled out all required fields of this form
- [x] I have used **Git LFS** for the Eval JSON data
- [x] (Ignore if not submitting code) I have run `pip install
pre-commit; pre-commit install` and have verified that `mypy`, `black`,
`isort`, `autoflake` and `ruff` are running when I commit and push

Failure to fill out all required fields will result in the PR being
closed.

### Eval JSON data

Since we are using Git LFS, we are asking eval submitters to add in as
many Eval Samples (at least 5) from their contribution here:

<details>
  <summary>View evals in JSON</summary>

  ### Eval
  ```jsonl
{"input": [{"role": "user", "content": "Jackson entered the hall. Chloe
entered the hall. The boots is in the bathtub. Jackson exited the hall.
Jackson entered the dining_room. Chloe moved the boots to the pantry.
Where does Chloe think that Jackson searches for the boots?"}], "ideal":
"bathtub"} ```
</details>

Author: inwaves

evals/elsuite/theory_of_mind/readme.md

@@ -0,0 +1,25 @@
+# Eval description
+This evaluation tests LLMs' performance on theory of mind and social intelligence benchmarks [ToMi](https://github.com/facebookresearch/ToMi) and [SocialIQA](https://allenai.org/data/socialiqa). 
+
+The `ToMi` test set contains 5,993 question-answer pairs. These are instances of the [Sally-Anne test](https://en.wikipedia.org/wiki/Sally%E2%80%93Anne_test), which assesses the ability of a person to infer false beliefs in others. The original setting involves two people, Sally and Anne, who are together in a room. Sally places a marble in a box. Then, Anne leaves the room, and while she is away, Sally moves the marble to a basket elsewhere in the room. When Anne returns to the room, where will she search for the marble? If the person responding “has” theory-of-mind they’ll respond that Anne searches for the marble in the box, where she had last seen it. If they do not, they ascribe their own, accurate belief regarding the location to Anne, and say that she looks for it in the basket.
+
+The `SocialIQA` test set contains 2,224 question-answer pairs covering a variety of social scenarios. These are multiple-choice, with 3 options of which only one is correct. The questions cover a person’s wants, needs, motivations, and reactions, as well as the effects of an action (on self or others), and how that action reflects on the person carrying it out (e.g. how others would perceive them after having carried out the action).
+
+Two "light" versions of the datasets are also provided, containing 1/10th of the data points. These are useful for iterating on prompts and developing other scaffolding.
+
+# Token and pricing estimates
+On average:
+- On the `SocialIQA` dataset, models consume ~250k tokens per run using the simple solver, and ~900k using the CoT solver.
+- On the `ToMi` dataset, models consume ~700k tokens per run using the simple solver, and ~2.4m using the CoT solver.
+
+To calculate dollar cost from token counts, please check the latest token pricing [here](https://openai.com/pricing). Note that we count both input and output tokens together, so a lower and upper estimate of the cost of each variant can be predicted.
+
+# Experiments
+As a starting point for deeper exploration, we provide scripts for comparing various solvers and eval variants, as well as for plotting the results. To run these:
+```
+cd scripts/
+bash run_experiments.sh
+```
+
+# Contribution statement
+Eval design was primarily conducted by Andrei Alexandru, under the guidance of (alphabetically by last-name) Steven Adler, James Aung, Rosie Campbell and Jade Leung who provided research input, report revisions, and project management support. 

evals/elsuite/theory_of_mind/scripts/data_generation.py

@@ -0,0 +1,82 @@
+import json
+
+# %%
+filepath = "/evals/registry/data/theory_of_mind/tomi/train.txt"
+
+lines, datapoints = [], []
+with open(filepath, "r") as f:
+    for line in f:
+        line_index = line.split(" ")[0]
+        if int(line_index) == 1:
+            if len(lines) == 0:
+                lines.append(line)
+            else:
+                target = lines[-1].split("\t")[-2]
+                last_line = lines[-1].split("\t")[0]
+                lines = [" ".join(line.replace("\n", "").split(" ")[1:]) for line in lines[:-1]]
+                context = " ".join(lines) + " " + " ".join(last_line.split(" ")[1:])
+                datapoints.append({"context": context, "target": target})
+                lines = [line]
+        else:
+            lines.append(line)
+# %%
+def convert_datapoints_to_eval_dataset(datapoints: list) -> list:
+    system_prompt = "You will read a number of sentences describing a situation involving several people, as well as a question regarding the situation. Your task is to answer the question based on the information in the sentences."
+    eval_dataset = []
+    for datapoint in datapoints:
+        context = datapoint["context"]
+        target = datapoint["target"]
+        eval_datapoint = {
+            "input": [
+                {"role": "system", "content": system_prompt},
+                {"role": "user", "content": context},
+            ],
+            "ideal": target,
+        }
+        eval_dataset += [eval_datapoint]
+    return eval_dataset
+
+
+# %%
+eval_dataset = convert_datapoints_to_eval_dataset(datapoints)
+# %%
+output_file = "tomi_train.jsonl"
+
+with open(output_file, "w") as out:
+    for datapoint in eval_dataset:
+        out.write(json.dumps(datapoint) + "\n")
+# %%
+filepath = "/evals/registry/data/theory_of_mind/socialiqa/test.jsonl"
+system_prompt = "You will read a number of sentences describing a situation, followed by a question regarding the situation. Your task is to answer the question based on the information in the sentences by choosing from one of three answers A, B or C."
+
+dataset = []
+with open(filepath, "r") as f:
+    for line in f:
+        entry = json.loads(line)
+        template = f"{entry['context']} {entry['question']} A: {entry['answerA']}; B: {entry['answerB']}; C: {entry['answerC']}."
+        dataset.append(
+            {
+                "input": [
+                    {"role": "system", "content": system_prompt},
+                    {"role": "user", "content": template},
+                ],
+                "ideal": entry["correct"],
+            }
+        )
+# %%
+output_file = "socialiqa_test.jsonl"
+with open(output_file, "w") as out:
+    for datapoint in dataset:
+        out.write(json.dumps(datapoint) + "\n")
+
+# %%
+
+filepath = "evals/registry/data/theory_of_mind/socialiqa/test.jsonl"
+outpath = "evals/registry/data/theory_of_mind/socialiqa/newtest.jsonl"
+
+dataset = []
+with open(filepath, "r") as f, open(outpath, "w") as out:
+    for line in f:
+        entry = json.loads(line)
+        entry["input"] = [entry["input"][1]]
+        out.write(json.dumps(entry) + "\n")

evals/elsuite/theory_of_mind/scripts/make_plots.py

@@ -0,0 +1,133 @@
+"""Take results from recent experiments and make a bar plot"""
+import argparse
+from pathlib import Path
+from typing import Union
+
+import matplotlib.pyplot as plt
+import numpy as np
+import pandas as pd
+import seaborn as sns
+
+from evals.utils import log_utils
+
+
+def main():
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--log_dir", type=str, required=True)
+    parser.add_argument("--out_dir", type=str, required=True)
+    args = parser.parse_args()
+
+    log_dir = args.log_dir
+    out_dir = args.out_dir
+    df = load_tom_results_from_dir(log_dir)
+    make_plot(df, out_dir=Path(out_dir))
+
+
+def load_tom_results_from_dir(log_dir: Union[str, Path]) -> pd.DataFrame:
+    rows = []
+    final_results_dict = log_utils.get_final_results_from_dir(log_dir)
+
+    for path, final_results in final_results_dict.items():
+        spec = log_utils.extract_spec(path)
+        dataset, prompt_type, model = parse_spec(spec)
+        rows.append(
+            {
+                "model": model,
+                "dataset": dataset,
+                "prompt_type": prompt_type,
+                "accuracy": final_results["accuracy"],
+                "bootstrap_std": final_results["bootstrap_std"],
+            }
+        )
+    return pd.DataFrame(rows)
+
+
+def parse_spec(spec: dict) -> tuple[str, bool, int]:
+    """parse the spec from a MMP run"""
+    completion_fn = spec["completion_fns"][0]
+    dataset, prompt_type, model = completion_fn.split("/")
+    prompt_type = prompt_type.split("_")[0]
+
+    return (dataset, prompt_type, model)
+
+
+def make_plot(df, out_dir):
+    sns.set_theme(style="whitegrid")
+    sns.set_palette("dark")
+    # Define the order of models
+    model_order = ["gpt-3.5-turbo", "gpt-4-base", "gpt-4"]
+    datasets = df["dataset"].unique()
+
+    for dataset in datasets:
+        ds = df[df["dataset"] == dataset.lower()]
+
+        # Ensure the model column is a categorical type with the specified order
+        ds["model"] = pd.Categorical(ds["model"], categories=model_order, ordered=True)
+        ds = ds.sort_values("model")  # Sort according to the categorical order
+
+        # Unique models
+        xs = ds["model"].unique()
+        # Get the accuracy values for both prompt types
+        simple_acc = ds[ds["prompt_type"] == "simple"]["accuracy"].values
+        cot_acc = ds[ds["prompt_type"] == "cot"]["accuracy"].values
+
+        # Get the corresponding error values from the "bootstrap_std" field
+        simple_std = ds[ds["prompt_type"] == "simple"]["bootstrap_std"].values
+        cot_std = ds[ds["prompt_type"] == "cot"]["bootstrap_std"].values
+
+        # Define the width of a bar
+        bar_width = 0.35
+        # Set the positions of the bars
+        x_indices = np.arange(len(xs))
+        x_indices2 = [x + bar_width for x in x_indices]
+
+        fig, ax1 = plt.subplots()
+        fig.suptitle(f"Accuracy on {dataset} dataset")
+
+        ax1.set_xlabel("Model")
+        ax1.set_ylabel("Accuracy")
+
+        # Plot the bars for 'simple' and 'cot'
+        ax1.bar(
+            x_indices,
+            simple_acc,
+            width=bar_width,
+            color=sns.color_palette("pastel")[0],
+            yerr=simple_std,
+            label="simple",
+        )
+        ax1.bar(
+            x_indices2,
+            cot_acc,
+            width=bar_width,
+            color=sns.color_palette("pastel")[1],
+            yerr=cot_std,
+            label="chain-of-thought",
+        )
+
+        if dataset == "socialiqa":
+            # Draw the horizontal line for the human baseline
+            human_baseline = 0.881
+            ax1.axhline(y=human_baseline, color="gray", linestyle="--", linewidth=1)
+            # Add the text label for the human baseline
+            ax1.text(
+                0.01, human_baseline, "human baseline", va="center", ha="left", backgroundcolor="w"
+            )
+
+        # Set the x-axis ticks to be in the middle of the two bars
+        ax1.set_xticks([r + bar_width / 2 for r in range(len(xs))])
+        ax1.set_xticklabels(xs, rotation=45)  # Rotate the x-axis labels if needed
+
+        ax1.set_ylim(0, 1)
+
+        # Add legend
+        ax1.legend(loc="upper right", bbox_to_anchor=(1, 1))
+
+        # Save the figure
+        plt.savefig(out_dir / f"accuracy_{dataset.lower()}.png", bbox_inches="tight")
+        plt.tight_layout()  # Adjust the plot to ensure everything fits without overlapping
+        plt.show()
+
+
+if __name__ == "__main__":
+    main()

evals/elsuite/theory_of_mind/scripts/run_experiments.sh

@@ -0,0 +1,26 @@
+logdir=./logs
+outputdir=./outputs
+timestamp=$(date +%Y%m%d_%H%M%S)
+logpathbase="$logdir/$timestamp/"
+
+echo Running experiments and logging to $logpathbase
+
+DATASETS="tomi socialiqa hitom"
+MODELS="gpt-3.5-turbo gpt-4 gpt-4-base"
+SOLVER_TYPES="simple_solver cot_solver"
+
+for dataset in $DATASETS
+do
+    for model in $MODELS
+    do
+        for solver in $SOLVER_TYPES
+        do
+            oaieval $dataset/$solver/$model "theory_of_mind."$dataset --record_path "$logpathbase/$model-$variant.log"
+        done
+    done
+done
+
+echo Done running experiments, all logs in $logpathbase
+
+echo Producing plots, outputs to $outputdir
+python3 make_plots.py --log_dir $logpathbase --out_dir $outputdir

evals/registry/data/theory_of_mind/.gitattributes

@@ -0,0 +1,4 @@
+tomi/test.jsonl filter=lfs diff=lfs merge=lfs -text
+tomi/test_light.jsonl filter=lfs diff=lfs merge=lfs -text
+socialiqa/test.jsonl filter=lfs diff=lfs merge=lfs -text
+socialiqa/test_light.jsonl filter=lfs diff=lfs merge=lfs -text

evals/registry/data/theory_of_mind/LICENSE

@@ -0,0 +1,7 @@
+ToMi:
+License: Creative Commons Attribution-NonCommercial 4.0 International (CC-BY-NC 4.0) https://creativecommons.org/licenses/by-nc/4.0/legalcode.en
+Source: https://github.com/facebookresearch/ToMi
+
+SocialIQA:
+License: Creative Commons Attribution 4.0 International (CC-BY 4.0) https://creativecommons.org/licenses/by/4.0/legalcode.en
+Source: https://allenai.org/data/socialiqa