Merge branch 'main' into metrics

# Conflicts:
#	lm_eval/models/vllm_causallms.py
#	pyproject.toml

lm_eval/tasks/afrobench/nollysenti/prompt_2/nollysenti

@@ -2,7 +2,6 @@ tag:
     - afrobench_sentiment_tasks
     - nollysenti_prompt_2
 dataset_path: Davlan/nollysenti
-dataset_kwargs: {trust_remote_code: True}
 output_type: multiple_choice
 validation_split: validation
 test_split: test

lm_eval/tasks/afrobench/nollysenti/prompt_3/nollysenti

@@ -2,7 +2,6 @@ tag:
     - afrobench_sentiment_tasks
     - nollysenti_prompt_3
 dataset_path: Davlan/nollysenti
-dataset_kwargs: {trust_remote_code: True}
 output_type: multiple_choice
 validation_split: validation
 test_split: test

lm_eval/tasks/afrobench/nollysenti/prompt_4/nollysenti

@@ -2,7 +2,6 @@ tag:
     - afrobench_sentiment_tasks
     - nollysenti_prompt_4
 dataset_path: Davlan/nollysenti
-dataset_kwargs: {trust_remote_code: True}
 output_type: multiple_choice
 validation_split: validation
 test_split: test

lm_eval/tasks/afrobench/nollysenti/prompt_5/nollysenti

@@ -2,7 +2,6 @@ tag:
     - afrobench_sentiment_tasks
     - nollysenti_prompt_5
 dataset_path: Davlan/nollysenti
-dataset_kwargs: {trust_remote_code: True}
 output_type: multiple_choice
 validation_split: validation
 test_split: test

lm_eval/tasks/afrobench/ntrex/prompt_1/african-english/ntrex

@@ -4,7 +4,6 @@ tag:
 - ntrex_afr-eng_prompt_1
 - afrobench_MT_tasks
 dataset_path: masakhane/ntrex_african
-dataset_kwargs: {trust_remote_code: True}
 output_type: generate_until
 validation_split: test
 fewshot_split: test

lm_eval/tasks/afrobench/ntrex/prompt_1/english-african/ntrex

@@ -4,7 +4,6 @@ tag:
 - ntrex_eng-afr_prompt_1
 - afrobench_MT_tasks
 dataset_path: masakhane/ntrex_african
-dataset_kwargs: {trust_remote_code: True}
 output_type: generate_until
 validation_split: test
 fewshot_split: test

lm_eval/tasks/afrobench/ntrex/prompt_2/african-english/ntrex

@@ -3,7 +3,6 @@ tag:
 - ntrex_afr-eng_prompt_2
 - afrobench_MT_tasks
 dataset_path: masakhane/ntrex_african
-dataset_kwargs: {trust_remote_code: True}
 output_type: generate_until
 validation_split: test
 fewshot_split: test

lm_eval/tasks/afrobench/ntrex/prompt_2/english-african/ntrex

@@ -3,7 +3,6 @@ tag:
 - ntrex_eng-afr_prompt_2
 - afrobench_MT_tasks
 dataset_path: masakhane/ntrex_african
-dataset_kwargs: {trust_remote_code: True}
 output_type: generate_until
 validation_split: test
 fewshot_split: test

lm_eval/tasks/afrobench/ntrex/prompt_3/african-english/ntrex

@@ -3,7 +3,6 @@ tag:
 - ntrex_afr-eng_prompt_3
 - afrobench_MT_tasks
 dataset_path: masakhane/ntrex_african
-dataset_kwargs: {trust_remote_code: True}
 output_type: generate_until
 validation_split: test
 fewshot_split: test

lm_eval/tasks/afrobench/ntrex/prompt_3/english-african/ntrex

@@ -3,7 +3,6 @@ tag:
 - ntrex_eng-afr_prompt_3
 - afrobench_MT_tasks
 dataset_path: masakhane/ntrex_african
-dataset_kwargs: {trust_remote_code: True}
 output_type: generate_until
 validation_split: test
 fewshot_split: test

lm_eval/tasks/afrobench/salt/prompt_1/salt

@@ -3,7 +3,6 @@ tag:
 - salt_prompt_1
 - afrobench_MT_tasks
 dataset_path: Sunbird/salt
-dataset_kwargs: {trust_remote_code: True}
 output_type: generate_until
 validation_split: dev
 fewshot_split: dev

lm_eval/tasks/afrobench/salt/prompt_2/salt

@@ -3,7 +3,6 @@ tag:
 - salt_prompt_2
 - afrobench_MT_tasks
 dataset_path: Sunbird/salt
-dataset_kwargs: {trust_remote_code: True}
 output_type: generate_until
 validation_split: dev
 fewshot_split: dev

lm_eval/tasks/afrobench/salt/prompt_3/salt

@@ -3,7 +3,6 @@ tag:
 - salt_prompt_3
 - afrobench_MT_tasks
 dataset_path: Sunbird/salt
-dataset_kwargs: {trust_remote_code: True}
 output_type: generate_until
 validation_split: dev
 fewshot_split: dev

lm_eval/tasks/aime/README.md

+# AIME
+
+### Citation
+
+```text
+@dataset{aime_1983_2024,
+  author = {Hemish Veeraboina},
+  title = {AIME Problem Set 1983-2024},
+  year = {2024},
+  publisher = {Kaggle},
+  url = {https://www.kaggle.com/datasets/hemishveeraboina/aime-problem-set-1983-2024}
+}
+
+@dataset{aime_2024,
+  author = {Maxwell Jia},
+  title = {AIME Problem Set 2024},
+  year = {2024},
+  publisher = {Huggingface},
+  url = {https://huggingface.co/datasets/Maxwell-Jia/AIME_2024}
+}
+
+@dataset{aime_2025,
+  author = {math-ai},
+  title = {AIME Problem Set 2025},
+  year = {2025},
+  publisher = {Huggingface},
+  url = {https://huggingface.co/datasets/math-ai/aime25}
+}
+```
+
+### Groups, Tags, and Tasks
+
+#### Groups
+
+* `math_word_problems`
+
+#### Tasks
+
+* `aime`: `AIME 1983-2024 problems`
+* `aime24`: `AIME 2024 problems`
+* `aime25`: `AIME 2025 problems`
+
+### Checklist
+
+For adding novel benchmarks/datasets to the library:
+
+* [x] Is the task an existing benchmark in the literature?
+  * [x] Have you referenced the original paper that introduced the task?
+  * [ ] If yes, does the original paper provide a reference implementation? If so, have you checked against the reference implementation and documented how to run such a test?
+
+If other tasks on this dataset are already supported:
+
+* [ ] Is the "Main" variant of this task clearly denoted?
+* [ ] Have you provided a short sentence in a README on what each new variant adds / evaluates?
+* [ ] Have you noted which, if any, published evaluation setups are matched by this variant?

lm_eval/tasks/aime/aime.yaml

+tag:
+  - math_word_problems
+task: aime
+dataset_path: gneubig/aime-1983-2024
+# dataset_name: null
+output_type: generate_until
+training_split: train
+fewshot_split: train
+test_split: train
+doc_to_text: "Question: {{Question}}\nAnswer:"
+doc_to_target: "{{Answer}}"
+metric_list:
+  - metric: exact_match
+    aggregation: mean
+    higher_is_better: true
+generation_kwargs:
+  until:
+    - "Question:"
+    - "</s>"
+    - "<|im_end|>"
+    - "<|eot_id|>"
+  do_sample: false
+  temperature: 0.0
+  max_gen_toks: 32768
+repeats: 1
+num_fewshot: 0
+metadata:
+  version: 0.0

lm_eval/tasks/aime/aime24.yaml

+tag:
+  - math_word_problems
+task: aime24
+dataset_path: Maxwell-Jia/AIME_2024
+# dataset_name: null
+output_type: generate_until
+training_split: train
+fewshot_split: train
+test_split: train
+doc_to_text: "Question: {{Problem}}\nAnswer:"
+doc_to_target: "{{Answer}}"
+process_results: !function utils.process_results
+metric_list:
+  - metric: exact_match
+    aggregation: mean
+    higher_is_better: true
+generation_kwargs:
+  until:
+    - "Question:"
+    - "</s>"
+    - "<|im_end|>"
+    - "<|eot_id|>"
+  do_sample: false
+  temperature: 0.0
+  max_gen_toks: 32768
+repeats: 1
+num_fewshot: 0
+metadata:
+  version: 0.0

lm_eval/tasks/aime/aime25.yaml

+tag:
+  - math_word_problems
+task: aime25
+dataset_path: math-ai/aime25
+# dataset_name: null
+output_type: generate_until
+training_split: test
+fewshot_split: test
+test_split: test
+doc_to_text: "Question: {{problem}}\nAnswer:"
+doc_to_target: "{{answer}}"
+process_results: !function utils.process_results
+metric_list:
+  - metric: exact_match
+    aggregation: mean
+    higher_is_better: true
+generation_kwargs:
+  until:
+    - "Question:"
+    - "</s>"
+    - "<|im_end|>"
+    - "<|eot_id|>"
+  do_sample: false
+  temperature: 0.0
+  max_gen_toks: 32768
+repeats: 1
+num_fewshot: 0
+metadata:
+  version: 0.0

lm_eval/tasks/aime/utils.py

+import re
+from typing import Dict, List
+
+
+def process_results(doc: dict, results: List[str]) -> Dict[str, int]:
+    retval = 0
+    response = results[0]
+
+    # Try to extract answer from $...$ format first
+    indices = [pos for pos, char in enumerate(response) if char == "$"]
+    if len(indices) <= 1:
+        answer = response
+    else:
+        answer = response[indices[0] + 1 : indices[-1]]
+
+    # Extract from \\boxed{} if present
+    boxed_answer = last_boxed_only_string(response)
+    if boxed_answer is not None:
+        try:
+            boxed_content = remove_boxed(boxed_answer)
+            if boxed_content is not None:
+                answer = boxed_content
+        except (AssertionError, IndexError):
+            pass
+
+    # Check if answer matches target
+    answer_key = next(k for k in doc.keys() if k.lower() == "answer")
+    target = str(doc[answer_key])
+    if is_equiv(answer, target):
+        retval = 1
+
+    return {"exact_match": retval}
+
+
+# string normalization from https://github.com/EleutherAI/lm-evaluation-harness/blob/master/lm_eval/tasks/hendrycks_math.py
+def is_equiv(str1, str2, verbose=False):
+    if str1 is None and str2 is None:
+        print("WARNING: Both None")
+        return True
+    if str1 is None or str2 is None:
+        return False
+
+    try:
+        ss1 = strip_string(str1)
+        ss2 = strip_string(str2)
+        if verbose:
+            print(ss1, ss2)
+        return ss1 == ss2
+    except Exception:
+        return str1 == str2
+
+
+def remove_boxed(s):
+    if "\\boxed " in s:
+        left = "\\boxed "
+        assert s[: len(left)] == left
+        return s[len(left) :]
+
+    left = "\\boxed{"
+
+    assert s[: len(left)] == left
+    assert s[-1] == "}"
+
+    return s[len(left) : -1]
+
+
+def last_boxed_only_string(string):
+    idx = string.rfind("\\boxed")
+    if "\\boxed " in string:
+        return "\\boxed " + string.split("\\boxed ")[-1].split("$")[0]
+    if idx < 0:
+        idx = string.rfind("\\fbox")
+        if idx < 0:
+            return None
+
+    i = idx
+    right_brace_idx = None
+    num_left_braces_open = 0
+    while i < len(string):
+        if string[i] == "{":
+            num_left_braces_open += 1
+        if string[i] == "}":
+            num_left_braces_open -= 1
+            if num_left_braces_open == 0:
+                right_brace_idx = i
+                break
+        i += 1
+
+    if right_brace_idx is None:
+        retval = None
+    else:
+        retval = string[idx : right_brace_idx + 1]
+
+    return retval
+
+
+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 substr[0] == "{":
+                new_str += substr
+            else:
+                try:
+                    assert len(substr) >= 2
+                except AssertionError:
+                    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:
+        a = int(a)
+        b = int(b)
+        assert string == "{}/{}".format(a, b)
+        new_string = "\\frac{" + str(a) + "}{" + str(b) + "}"
+        return new_string
+    except AssertionError:
+        return string
+
+
+def remove_right_units(string):
+    # "\\text{ " only ever occurs (at least in the val set) when describing units
+    if "\\text{ " in string:
+        splits = string.split("\\text{ ")
+        assert len(splits) == 2
+        return splits[0]
+    else:
+        return string
+
+
+def fix_sqrt(string):
+    if "\\sqrt" not in string:
+        return string
+    splits = string.split("\\sqrt")
+    new_string = splits[0]
+    for split in splits[1:]:
+        if split[0] != "{":
+            a = split[0]
+            new_substr = "\\sqrt{" + a + "}" + split[1:]
+        else:
+            new_substr = "\\sqrt" + split
+        new_string += new_substr
+    return new_string
+
+
+def strip_string(string):
+    # linebreaks
+    string = string.replace("\n", "")
+
+    # remove inverse spaces
+    string = string.replace("\\!", "")
+
+    # replace \\ with \
+    string = string.replace("\\\\", "\\")
+
+    # replace tfrac and dfrac with frac
+    string = string.replace("tfrac", "frac")
+    string = string.replace("dfrac", "frac")
+
+    # remove \left and \right
+    string = string.replace("\\left", "")
+    string = string.replace("\\right", "")
+
+    # Remove circ (degrees)
+    string = string.replace("^{\\circ}", "")
+    string = string.replace("^\\circ", "")
+
+    # remove dollar signs
+    string = string.replace("\\$", "")
+
+    # remove units (on the right)
+    string = remove_right_units(string)
+
+    # remove percentage
+    string = string.replace("\\%", "")
+    string = string.replace("\%", "")  # noqa: W605
+
+    # " 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.")
+    # 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]
+
+    # fix sqrt3 --> sqrt{3}
+    string = fix_sqrt(string)
+
+    # remove spaces
+    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)
+
+    # manually change 0.5 --> \frac{1}{2}
+    if string == "0.5":
+        string = "\\frac{1}{2}"
+
+    # 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)
+
+    return string

lm_eval/tasks/babilong/README.md

+# Babilong
+
+### Paper
+
+Title: Babilong: Testing the Limits of LLMs with Long Context Reasoning-in-a-Haystack
+Abstract: https://arxiv.org/abs/2406.10149
+
+In recent years, the input context sizes of large language models (LLMs) have increased dramatically. However, existing evaluation methods have not kept pace, failing to comprehensively assess the efficiency of models in handling long contexts. To bridge this gap, we introduce the BABILong benchmark, designed to test language models' ability to reason across facts distributed in extremely long documents. BABILong includes a diverse set of 20 reasoning tasks, including fact chaining, simple induction, deduction, counting, and handling lists/sets. These tasks are challenging on their own, and even more demanding when the required facts are scattered across long natural text. Our evaluations show that popular LLMs effectively utilize only 10-20\% of the context and their performance declines sharply with increased reasoning complexity. Among alternatives to in-context reasoning, Retrieval-Augmented Generation methods achieve a modest 60\% accuracy on single-fact question answering, independent of context length. Among context extension methods, the highest performance is demonstrated by recurrent memory transformers after fine-tuning, enabling the processing of lengths up to 50 million tokens. The BABILong benchmark is extendable to any length to support the evaluation of new upcoming models with increased capabilities, and we provide splits up to 10 million token lengths.
+
+Homepage: https://github.com/booydar/babilong
+
+### Citation
+
+```
+@article{kuratov2024babilong,
+    title={Babilong: Testing the Limits of LLMs with Long Context Reasoning-in-a-Haystack},
+    author={Kuratov, Yuri and Bulatov, Aydar and Anokhin, Petr and Rodkin, Ivan and Sorokin, Dmitry and Burtsev, Mikhail},
+    journal={arXiv preprint arXiv:2406.10149},
+    year={2024}
+}
+```
+
+### Groups and Tasks
+
+#### Groups
+
+* `babilong`: All Babilong tasks at 0k context length
+* `babilong_longctx`: Babilong tasks between qa1-qa5 at context lengths up to 128k
+
+
+#### Tasks
+
+The benchmark includes 1000 samples of 20 reasoning tasks at various context lengths:
+
+**QA Tasks (qa1-qa20):**
+* `babilong_qa1`: Single supporting fact QA
+* `babilong_qa2`: Two supporting facts QA
+* `babilong_qa3`: Three supporting facts QA
+* `babilong_qa4`: Two argument relations
+* `babilong_qa5`: Three argument relations
+* `babilong_qa6`: Yes/No questions
+* `babilong_qa7`: Counting
+* `babilong_qa8`: Lists and sets
+* `babilong_qa9`: Simple negation
+* `babilong_qa10`: Indefinite knowledge
+* `babilong_qa11`: Track person through temporal references
+* `babilong_qa12`: Conjunction
+* `babilong_qa13`: Compound coreference
+* `babilong_qa14`: Time reasoning
+* `babilong_qa15`: Basic deduction
+* `babilong_qa16`: Basic induction
+* `babilong_qa17`: Positional reasoning
+* `babilong_qa18`: Size reasoning
+* `babilong_qa19`: Path finding
+* `babilong_qa20`: Motivation deduction
+
+> [!NOTE]
+> When using babilong tasks, please note:
+> 1. This is the implementation with 1000 samples per length. You can change the dataset path to `RMT-team/babilong` in `common_utils.py` for the dataset with 100 samples per length, which supports context lengths up to 10M tokens.
+> 2. Supported lengths are 0k, 1, 2, 4, 8, 16, 32, 64, 128k tokens for tasks qa1-5. Tasks qa6-20 only have a length of 0k.
+> 3. The default maximum sequence length is 0k. For calculating metrics of different max seq lengths, specify additional lengths using the metadata parameter:
+>   `--metadata '{"max_seq_lengths":"0k,1k,2k,4k,8k,16k,32k,128k"}'`. The config currently only takes one context length at a time. The metadata parameter can also be passed to the TaskManager (metadata: dict).
+
+
+### Checklist
+
+For adding novel benchmarks/datasets to the library:
+* [x] Is the task an existing benchmark in the literature?
+  * [x] Have you referenced the original paper that introduced the task?
+  * [x] If yes, does the original paper provide a reference implementation? If so, have you checked against the reference implementation and documented how to run such a test?
+
+
+If other tasks on this dataset are already supported:
+* [ ] Is the "Main" variant of this task clearly denoted?
+* [ ] Have you provided a short sentence in a README on what each new variant adds / evaluates?
+* [ ] Have you noted which, if any, published evaluation setups are matched by this variant?

lm_eval/tasks/babilong/_babilong_common_yaml

+dataset_path: RMT-team/babilong-1k-samples
+output_type: generate_until
+doc_to_target: "{{target}}"
+target_delimiter: " "
+num_fewshot: 2
+process_results: !function common_utils.process_results
+metric_list:
+  - metric: acc
+    aggregation: mean
+    higher_is_better: true
+generation_kwargs:
+  do_sample: false
+  temperature: 0.0
+  max_gen_toks: 16
+  until: []
+metadata:
+  version: 0.0