Add MATH evaluation

README.md

@@ -124,6 +124,7 @@ We provide the scripts for running evaluation of Huggingface/OpenAI models on a
 
 - [MMLU](https://github.com/hendrycks/test)
 - [Grade School Math (GSM)](https://github.com/openai/grade-school-math)
+- [MATH](https://github.com/hendrycks/math)
 - [Big-Bench Hard (BBH)](https://github.com/suzgunmirac/BIG-Bench-Hard/tree/main)
 - [TydiQA](https://github.com/google-research-datasets/tydiqa)
 - [Codex HumanEval](https://github.com/openai/human-eval/tree/master)

eval/MATH/answer_extraction.py

@@ -0,0 +1,271 @@
+import re
+import regex
+
+def _fix_fracs(string):
+    substrs = string.split("\\frac")
+    new_str = substrs[0]
+    if len(substrs) > 1:
+        substrs = substrs[1:]
+        for substr in substrs:
+            new_str += "\\frac"
+            if len(substr) > 0 and substr[0] == "{":
+                new_str += substr
+            else:
+                try:
+                    assert len(substr) >= 2
+                except:
+                    return string
+                a = substr[0]
+                b = substr[1]
+                if b != "{":
+                    if len(substr) > 2:
+                        post_substr = substr[2:]
+                        new_str += "{" + a + "}{" + b + "}" + post_substr
+                    else:
+                        new_str += "{" + a + "}{" + b + "}"
+                else:
+                    if len(substr) > 2:
+                        post_substr = substr[2:]
+                        new_str += "{" + a + "}" + b + post_substr
+                    else:
+                        new_str += "{" + a + "}" + b
+    string = new_str
+    return string
+
+
+def _fix_a_slash_b(string):
+    if len(string.split("/")) != 2:
+        return string
+    a = string.split("/")[0]
+    b = string.split("/")[1]
+    try:
+        if "sqrt" not in a:
+            a = int(a)
+        if "sqrt" not in b:
+            b = int(b)
+        assert string == "{}/{}".format(a, b)
+        new_string = "\\frac{" + str(a) + "}{" + str(b) + "}"
+        return new_string
+    except:
+        return string
+
+
+def _fix_sqrt(string):
+    _string = re.sub(r"\\sqrt(-?[0-9.a-zA-Z]+)", r"\\sqrt{\1}", string)
+    _string = re.sub(r"\\sqrt\s+(\w+)$", r"\\sqrt{\1}", _string)
+    return _string
+
+
+def _fix_tan(string):
+    _string = re.sub(r"\\tan(-?[0-9.a-zA-Z]+)", r"\\tan{\1}", string)
+    _string = re.sub(r"\\tan\s+(\w+)$", r"\\tan{\1}", _string)
+    return _string
+
+
+def strip_string(string):
+    string = str(string).strip()
+    # linebreaks
+    string = string.replace("\n", "")
+
+    # right "."
+    string = string.rstrip(".")
+
+    # remove inverse spaces
+    string = string.replace("\\!", "")
+    # string = string.replace("\\ ", "")
+
+    # replace \\ with \
+    # string = string.replace("\\\\", "\\")
+    # string = string.replace("\\\\", "\\")
+
+    if string.startswith("\\text{") and string.endswith("}"):
+        string = string.split("{", 1)[1][:-1]
+
+    # replace tfrac and dfrac with frac
+    string = string.replace("tfrac", "frac")
+    string = string.replace("dfrac", "frac")
+    string = string.replace("cfrac", "frac")
+
+    # remove \left and \right
+    string = string.replace("\\left", "")
+    string = string.replace("\\right", "")
+
+    # Remove unit: miles, dollars if after is not none
+    _string = re.sub(r"\\text{.*?}$", "", string).strip()
+    if _string != "" and _string != string:
+        # print("Warning: unit not removed: '{}' -> '{}'".format(string, _string))
+        string = _string
+
+    # Remove circ (degrees)
+    string = string.replace("^{\\circ}", "").strip()
+    string = string.replace("^\\circ", "").strip()
+
+    string = regex.sub(r"\{(c|m)?m\}(\^(2|3))?", "", string).strip()
+    string = regex.sub(r"p\.m\.$", "", string).strip()
+    string = regex.sub(r"(\d)\s*t$", r"\1", string).strip()
+
+    # remove dollar signs
+    string = string.replace("\\$", "")
+    string = string.replace("$", "")
+
+    # string = string.replace("\\text", "")
+    string = string.replace("x\\in", "")
+
+    # remove percentage
+    string = string.replace("\\%", "%")
+    string = string.replace("\%", "%")
+    # string = string.replace("%", "")
+
+    # " 0." equivalent to " ." and "{0." equivalent to "{." Alternatively, add "0" if "." is the start of the string
+    string = string.replace(" .", " 0.")
+    string = string.replace("{.", "{0.")
+
+    # cdot
+    string = string.replace("\\cdot", "")
+
+    # inf
+    string = string.replace("infinity", "\\infty")
+    if "\\infty" not in string:
+        string = string.replace("inf", "\\infty")
+    string = string.replace("+\\inity", "\\infty")
+
+    # and 
+    # string = string.replace("and", "")
+    string = string.replace("\\mathbf", "")
+    string = string.replace("\\mathrm", "")
+
+    # use regex to remove \mbox{...}
+    string = re.sub(r"\\mbox{.*?}", "", string)
+
+    # quote
+    string.replace("'", "")
+    string.replace("\"", "")
+
+    # i, j
+    if "j" in string and "i" not in string:
+        string = string.replace("j", "i")
+
+    # replace a.000b where b is not number or b is end, with ab, use regex
+    string = re.sub(r"(\d+)\.0+([^\d])", r"\1\2", string)
+    string = re.sub(r"(\d+)\.0+$", r"\1", string)
+
+    # if empty, return empty string
+    if len(string) == 0:
+        return string
+    if string[0] == ".":
+        string = "0" + string
+
+    # to consider: get rid of e.g. "k = " or "q = " at beginning
+    # if len(string.split("=")) == 2:
+    #     if len(string.split("=")[0]) <= 2:
+    #         string = string.split("=")[1]
+
+    string = _fix_sqrt(string)
+    string = _fix_tan(string)
+    string = string.replace(" ", "")
+
+    # \frac1b or \frac12 --> \frac{1}{b} and \frac{1}{2}, etc. Even works with \frac1{72} (but not \frac{72}1). Also does a/b --> \\frac{a}{b}
+    string = _fix_fracs(string)
+
+    # NOTE: X/Y changed to \frac{X}{Y} in dataset, but in simple cases fix in case the model output is X/Y
+    string = _fix_a_slash_b(string)
+
+    string = regex.sub(r"(\\|,|\.)+$", "", string)
+
+    return string
+
+def extract_boxed_answers(text):
+    answers = []
+    for piece in text.split('boxed{')[1:]:
+        n = 0
+        for i in range(len(piece)):
+            if piece[i] == '{':
+                n += 1
+            elif piece[i] == '}':
+                n -= 1
+                if n < 0:
+                    if i + 1 < len(piece) and piece[i + 1] == '%':
+                        answers.append(piece[: i + 1])
+                    else:
+                        answers.append(piece[:i])
+                    break
+    return answers
+
+def extract_program_output(pred_str):
+    """
+    extract output between the last ```output\n...\n```
+    """
+    if "```output" not in pred_str:
+        return ""
+    if '```output' in pred_str:
+        pred_str = pred_str.split('```output')[-1]
+    if '```' in pred_str:
+        pred_str = pred_str.split('```')[0]
+    output = pred_str.strip()
+    return output
+
+def extract_answer(pred_str, exhaust=False):
+    pred = []
+    if 'final answer is $' in pred_str and '$. I hope' in pred_str:
+        tmp = pred_str.split('final answer is $', 1)[1]
+        pred = [tmp.split('$. I hope', 1)[0].strip()]
+    elif 'boxed' in pred_str:
+        pred = extract_boxed_answers(pred_str)
+    elif ('he answer is' in pred_str):
+        pred = [pred_str.split('he answer is')[-1].strip()]
+    else:
+        program_output = extract_program_output(pred_str)
+        if program_output != "":
+            # fall back to program
+            pred.append(program_output)
+        else: # use the last number
+            pattern = '-?\d*\.?\d+'
+            ans = re.findall(pattern, pred_str.replace(",", ""))
+            if(len(ans) >= 1):
+                ans = ans[-1]
+            else:
+                ans = ''
+            if ans:
+                pred.append(ans)
+
+    # multiple line
+    _pred = []
+    for ans in pred:
+        ans = ans.strip().split("\n")[0]
+        ans = ans.lstrip(":")
+        ans = ans.rstrip(".")
+        ans = ans.rstrip("/")
+        ans = strip_string(ans)
+        _pred.append(ans)
+    if exhaust:
+        return _pred
+    else:
+        return _pred[-1] if _pred else ""
+
+def extract_math_answer(question, reasoning, task):
+    answer = []
+    for ans in extract_answer(reasoning, exhaust=True):
+        if 'separated by commas' in question and all(ch not in ans for ch in '()[]'):
+            answer.extend([a.strip() for a in ans.split(",")])
+        elif regex.search(r"\\text\{\s*and\s*\}", ans):
+            answer.extend([a.strip() for a in regex.sub(r"\\text\{\s*and\s*\}", "[SEP]", ans).split("[SEP]")])
+        else:
+            answer.append(ans.strip())
+    return answer
+
+def extract_math_few_shot_cot_answer(question, reasoning, task):
+    if 'Problem:' in reasoning:
+        reasoning = reasoning.split("Problem:", 1)[0]
+    return extract_math_answer(question, reasoning, task)
+
+def extract_last_single_answer(question, reasoning, task):
+    return extract_answer(reasoning, exhaust=False)
+
+def extract_gsm_few_shot_cot_answer(question, reasoning, task):
+    if 'Q: ' in reasoning:
+        reasoning = reasoning.split("Q: ", 1)[0]
+    pred = [s for s in regex.findall(r'-?\d+\.?\d*', reasoning)]
+    if pred:
+        return pred[-1]
+    else:
+        return "[invalid]"

eval/MATH/examplars.py

@@ -0,0 +1,23 @@
+# These examplars are from the DeepSeekMath GitHub repository (https://github.com/deepseek-ai/DeepSeek-Math/tree/main/evaluation/few_shot_prompts)
+EXAMPLARS = [
+    {
+        "question": "Find the domain of the expression $\\frac{\\sqrt{x-2}}{\\sqrt{5-x}}$.}",
+        "cot_answer": "The expressions inside each square root must be non-negative.\nTherefore, $x-2 \\ge 0$, so $x\\ge2$, and $5 - x \\ge 0$, so $x \\le 5$.\nAlso, the denominator cannot be equal to zero, so $5-x>0$, which gives $x<5$.\nTherefore, the domain of the expression is $\\boxed{[2,5)}$.",
+        "short_answer": "[2,5)"
+    },
+    {
+        "question": "If $\\det \\mathbf{A} = 2$ and $\\det \\mathbf{B} = 12,$ then find $\\det (\\mathbf{A} \\mathbf{B}).$",
+        "cot_answer": "We have that $\\det (\\mathbf{A} \\mathbf{B}) = (\\det \\mathbf{A})(\\det \\mathbf{B}) = (2)(12) = \\boxed{24}.$",
+        "short_answer": "24"
+    },
+    {
+        "question": "Terrell usually lifts two 20-pound weights 12 times. If he uses two 15-pound weights instead, how many times must Terrell lift them in order to lift the same total weight?",
+        "cot_answer": "If Terrell lifts two 20-pound weights 12 times, he lifts a total of $2\\cdot 12\\cdot20=480$ pounds of weight.  If he lifts two 15-pound weights instead for $n$ times, he will lift a total of $2\\cdot15\\cdot n=30n$ pounds of weight.  Equating this to 480 pounds, we can solve for $n$: \\begin{align*}\n30n&=480\\\\\n\\Rightarrow\\qquad n&=480/30=\\boxed{16}\n\\end{align*}",
+        "short_answer": "16"
+    },
+    {
+        "question": "If the system of equations\n\n\\begin{align*}\n6x-4y&=a,\\\\\n6y-9x &=b.\n\\end{align*}has a solution $(x, y)$ where $x$ and $y$ are both nonzero, find $\\frac{a}{b},$ assuming $b$ is nonzero.",
+        "cot_answer": "If we multiply the first equation by $-\\frac{3}{2}$, we obtain\n\n$$6y-9x=-\\frac{3}{2}a.$$Since we also know that $6y-9x=b$, we have\n\n$$-\\frac{3}{2}a=b\\Rightarrow\\frac{a}{b}=\\boxed{-\\frac{2}{3}}.$$",
+        "short_answer": "-\\frac{2}{3}"
+    }
+]