Shinyy/LMR-Bench · Datasets at Hugging Face

instance_id int64 1 31	paper_name stringlengths 10 101	folder_name stringlengths 5 26	paper_url stringlengths 32 50	year int64 2.02k 2.03k	repo_url stringlengths 30 90	repo_folder_name stringlengths 3 31	implementations listlengths 1 5
11	MoverScore: Text Generation Evaluating with Contextualized Embeddings and Earth Mover Distance	11-moverscore	https://aclanthology.org/D19-1053.pdf	2,019	https://github.com/AIPHES/emnlp19-moverscore	moverscore	[ { "category": "Evaluation Metrics & Benchmarking", "class_name": "", "goal_file": "movescore.py", "goal_function": "word_mover_score", "golden_file": "golden_files/movescore_golden.py", "index": 1, "instruction": "You need to implement the word_mover_score function in movescore.py based on paper.pdf, which calculates a similarity score between two sentences by using BERT embeddings and by calculating calculating the value of 1-Word Mover's Distance metric", "retrieval_content": [], "retrieval_context": null, "unit_test_file": "unit_test/unit_test_1.py" } ]
13	Probing the Decision Boundaries of In-context Learning in Large Language Models	13-DecBound	https://arxiv.org/pdf/2406.11233	2,024	https://github.com/siyan-zhao/ICL_decision_boundary	DecBound-main	[ { "category": "Data Augmentation & Generation", "class_name": "", "goal_file": "data_utils.py", "goal_function": "generate_tasks", "golden_file": "golden_files/data_utils_golden.py", "index": 1, "instruction": "Implement `generate_tasks` function in data_utils.py based on paper.pdf and the repository, which generates multiple machine learning tasks, each represented as a separate dataset.", "retrieval_content": [], "retrieval_context": null, "unit_test_file": "unit_test/unit_test.py" } ]
14	EmojiPrompt: Generative Prompt Obfuscation for Privacy-Preserving Communication with Cloud-based LLMs	14-EmojiCrypt	https://arxiv.org/pdf/2402.05868	2,025	https://github.com/agiresearch/EmojiCrypt	EmojiCrypt-main	[ { "category": "Evaluation Metrics", "class_name": "", "goal_file": "Tabular.py", "goal_function": "compute_mean_cosine_similarity", "golden_file": "golden_files/Tabular_golden.py", "index": 1, "instruction": "Implement `compute_mean_cosine_similarity` function in Tabular.py based on the paper.pdf and the repository. Computes the mean cosine similarity between pairs of encrypted and decrypted feature levels.", "retrieval_content": [], "retrieval_context": null, "unit_test_file": "unit_test/unit_test_1.py" } ]
15	Language Models Predict Empathy Gaps Between Social In-groups and Out-groups	15-EmpathyBias	https://arxiv.org/pdf/2503.01030	2,025	https://github.com/houyu0930/intergroup-empathy-bias	EmpathyBias-main	[ { "category": "Evaluation Metrics", "class_name": "", "goal_file": "analysis/analysis_after_processing.py", "goal_function": "get_delta", "golden_file": "golden_files/analysis_after_processing_golden.py", "index": 1, "instruction": "Implement `get_delta` function analysis/analysis_after_processing.py based on paper.pdf and the repository. Compute the difference in average intensity between in-group and out-group interactions.", "retrieval_content": [], "retrieval_context": null, "unit_test_file": "unit_test/unit_test_1.py" }, { "category": "Evaluation Metrics", "class_name": "", "goal_file": "analysis/analysis_after_processing.py", "goal_function": "get_filtered_matrix", "golden_file": "golden_files/analysis_after_processing_golden.py", "index": 2, "instruction": "Implement `get_filtered_matrix` function analysis/analysis_after_processing.py based on paper.pdf and the repository. Compute the standardized emotion intensity matrix for a specific group, model, and prompt variation, excluding specified data points (by ID).", "retrieval_content": [], "retrieval_context": null, "unit_test_file": "unit_test/unit_test_2.py" } ]
16	Probing Language Models on Their Knowledge Source	16-KnowProb	https://aclanthology.org/2024.blackboxnlp-1.35.pdf	2,024	https://github.com/Zineddine-Tighidet/knowledge-probing-framework	KnowProb-main	[ { "category": "Interpretability & Explainability", "class_name": "", "goal_file": "src/classification/classifier.py", "goal_function": "perform_classification_by_relation_group", "golden_file": "golden_files/classifier_golden.py", "index": 1, "instruction": "Implement `perform_classification_by_relation_group` function in src/classification/classifier.py based on paper.pdf and the repository. This function performs classification over the saved activations, the classification can be module-wise and token-wise (i.e. vertical).", "retrieval_content": [], "retrieval_context": null, "unit_test_file": "unit_test/unit_test_1.py" }, { "category": "Data Augmentation & Generation", "class_name": "", "goal_file": "src/data.py", "goal_function": "remove_object_subject_overlap", "golden_file": "golden_files/data_golden.py", "index": 2, "instruction": "Implement `remove_object_subject_overlap` function in src/data.py based on paper.pdf and the repository. This function deletes the examples where the subject and the parametric object are similar using a Jaro-Winkler string similarity.", "retrieval_content": [], "retrieval_context": null, "unit_test_file": "unit_test/unit_test_2.py" }, { "category": "Information Extraction & Knowledge Integration", "class_name": "", "goal_file": "src/parametric_knowledge.py", "goal_function": "is_parametric_object_not_in_the_prompt", "golden_file": "golden_files/parametric_knowledge_golden.py", "index": 3, "instruction": "Implement `is_parametric_object_not_in_the_prompt` function in src/parametric_knowledge.py based on paper.pdf and the repository. This function checks whether the parametric_object is included in the one-shot examples that were used to guide the LLM during parametric knowledge building as it would mean that it's biased by the prompt.", "retrieval_content": [], "retrieval_context": null, "unit_test_file": "unit_test/unit_test_3.py" }, { "category": "Information Extraction & Knowledge Integration", "class_name": "", "goal_file": "src/data.py", "goal_function": "return_entity_overlap_between_relation_groups", "golden_file": "golden_files/data_golden.py", "index": 4, "instruction": "Implement `return_entity_overlap_between_relation_groups` function in src/data.py based on paper.pdf and the repository. This function returns a dict of dicts representing a kind of matrix of shape (nb_relation_groups, nb_relation_groups), where each value is the number of examples in a relation group that have the same subject or object in another relation group.", "retrieval_content": [], "retrieval_context": null, "unit_test_file": "unit_test/unit_test_4.py" }, { "category": "Data Augmentation & Generation", "class_name": "", "goal_file": "src/data.py", "goal_function": "generate_counter_parametric_knowledge_dataset", "golden_file": "golden_files/data_golden.py", "index": 5, "instruction": "Implement `generate_counter_parametric_knowledge_dataset` function in src/data.py based on paper.pdf and the repository. This function generates the counter-parametric knowledge using the parametric knowledge.", "retrieval_content": [], "retrieval_context": null, "unit_test_file": "unit_test/unit_test_5.py" } ]
18	Massive Activations in Large Language Models	18-MassActiv	https://arxiv.org/abs/2402.17762	2,024	https://github.com/locuslab/massive-activations	MassActiv-main	[ { "category": "Interpretability & Explainability", "class_name": "", "goal_file": "main_vit.py", "goal_function": "run_exp1", "golden_file": "golden_files/main_vit_golden.py", "index": 1, "instruction": "Implement run_exp1 function in main_vit.py. Run 3D feature visualization for a specific layer in a Vision Transformer (ViT) model. This function extracts the absolute feature activations from a specified transformer layer, visualizes them in 3D, and saves both the input parameters and feature activations as serialized pickle files for reproducibility and analysis.", "retrieval_content": [], "retrieval_context": null, "unit_test_file": "unit_test/unit_test.py" } ]
19	Native Sparse Attention: Hardware-Aligned and Natively Trainable Sparse Attention	19-Native-Sparse-Attention	https://arxiv.org/abs/2502.11089	2,025	https://github.com/fla-org/native-sparse-attention	NSA	[ { "category": "Neural Network Architectures & Modules", "class_name": "", "goal_file": "NSA.py", "goal_function": "nsa", "golden_file": "golden_files/NSA_golden.py", "index": 1, "instruction": "Please implement the `nsa` function in NSA.py based on the Natively trainable Sparse Attention mechanism mentioned in the paper.", "retrieval_content": [], "retrieval_context": null, "unit_test_file": "unit_test/unit_test_1.py" } ]
20	Trusting Your Evidence: Hallucinate Less with Context-aware Decoding	20-CAD	https://arxiv.org/abs/2305.14739	2,023	https://github.com/xhan77/context-aware-decoding.git	context-aware-decoding-main	[ { "category": "Decoding & Search Strategies", "class_name": "", "goal_file": "unit_test/unit_test_1.py", "goal_function": "context_aware_sampling", "golden_file": "golden_files/unit_test_1_golden.py", "index": 1, "instruction": "Implement the function context_aware_sampling in the unit_test/unit_test_1.py and pass the test by running unit_test/unit_test_1.py file, you may ignore the following parameters: alpha, max_length, temperature", "retrieval_content": null, "retrieval_context": [], "unit_test_file": "unit_test/unit_test_1.py" } ]
22	Token-level Direct Preference Optimization	22-TokenDPO	https://arxiv.org/abs/2404.11999	2,024	https://github.com/Vance0124/Token-level-Direct-Preference-Optimization/tree/master	TokenDPO-main	[ { "category": "Training Objectives & Optimization Techniques", "class_name": "", "goal_file": "trainers.py", "goal_function": "tdpo_loss", "golden_file": "golden_files/trainers_golden.py", "index": 1, "instruction": "Implement the `tdpo_loss` function in trainers.py based on the Token-level DPO loss mentioned in paper.pdf and the code repository. You may ignore the following parameters: if_tdpo2.", "retrieval_content": null, "retrieval_context": [], "unit_test_file": "unit_test/unit_test_1.py" } ]
23	A Simple Framework for Contrastive Learning of Visual Representations	23-SimCLR	https://arxiv.org/pdf/2002.05709	2,020	https://github.com/skywalkerzhang/SimCLR/tree/master	SimCLR-main	[ { "category": "Training Objectives & Optimization Techniques", "class_name": "SimCLR", "goal_file": "simclr.py", "goal_function": "info_nce_loss", "golden_file": "golden_files/simclr_golden.py", "index": 1, "instruction": "Implement the `info_nce_loss` method of `SimCLR` class in simclr.py that computes the InfoNCE loss for a batch of features, based on the method described in the paper.pdf and the repository.", "retrieval_content": null, "retrieval_context": [], "unit_test_file": "unit_test/unit_test_1.py" } ]
25	AdaLoRA: Adaptive Budget Allocation for Parameter-Efficient Fine-Tuning	25-AdaLora	https://arxiv.org/pdf/2303.10512	2,023	https://github.com/QingruZhang/AdaLoRA/tree/main	AdaLoRA	[ { "category": "Neural Network Architectures & Modules", "class_name": "SVDLinear", "goal_file": "loralib/loralib/adalora.py", "goal_function": "forward", "golden_file": "golden_files/adalora_golden.py", "index": 1, "instruction": "Implement the forward function of SVD-based Adaptated linear layer in loralib/loralib/adalora.py mentioned in the section 3.1 of the paper.", "retrieval_content": [], "retrieval_context": null, "unit_test_file": "unit_test/unit_test_1.py" } ]
28	Self-Play Fine-Tuning Converts Weak Language Models to Strong Language Models	28-SPIN	https://arxiv.org/abs/2401.01335	2,024	https://github.com/uclaml/SPIN	SPIN	[ { "category": "loss function", "class_name": "SPINTrainer", "goal_file": "spin/alignment/trainer.py", "goal_function": "spin_loss", "golden_file": "golden_files/trainer_golden.py", "index": 1, "instruction": "Implement the `spin_loss` method of `SPINTrainer` class in `.spin/alignment/trainer.py` based on the loss function of SPIN algorithm mentioned in the paper.pdf.", "retrieval_content": [], "retrieval_context": null, "unit_test_file": "unit_test/unit_test_1.py" } ]
30	ERGO: Event Relational Graph Transformer for Document-level Event Causality Identification	30-ERGO	https://aclanthology.org/2022.coling-1.185.pdf	2,022	https://github.com/chenmeiqii/ERGO	ERGO	[ { "category": "Training Objectives & Optimization Techniques", "class_name": "focal_loss", "goal_file": "model.py", "goal_function": "forward", "golden_file": "golden_files/model_golden.py", "index": 1, "instruction": "Implement the loss function in the focal_loss class in model.py based on the final adaptive focal loss mentioned in the paper and the code repository. Justify whether summing or averaging the loss based on the `self.size_average` value.", "retrieval_content": null, "retrieval_context": [], "unit_test_file": "unit_test/unit_test_1.py" } ]
31	Aspect-Based Sentiment Analysis with Syntax-Opinion-Sentiment Reasoning Chain	31-Syn-Chain	https://aclanthology.org/2025.coling-main.210.pdf	2,025	https://github.com/rf-x/Syn-Chain-ABSA	Syn-Chain-ABSA	[ { "category": "", "class_name": "", "goal_file": "conll_tree.py", "goal_function": "spacy_result_to_conll", "golden_file": "golden_files/conll_tree_golden.py", "index": 1, "instruction": "Implement the spacy_result_to_conll function in conll_tree.py to generate a dependency parse in the CoNLL-U format, as mentioned in the paper and the code repository.", "retrieval_content": [], "retrieval_context": null, "unit_test_file": "unit_test/unit_test_1.py" } ]
1	Direct Preference Optimization: Your Language Model is Secretly a Reward Model	01-DPO	https://arxiv.org/pdf/2305.18290	2,023	https://github.com/eric-mitchell/direct-preference-optimization	direct-preference-optimization	[ { "category": "Training Objectives & Optimization Techniques", "class_name": "", "goal_file": "trainers.py", "goal_function": "preference_loss", "golden_file": "golden_files/trainers_golden.py", "index": 1, "instruction": "Implement the preference_loss function in trainers.py based on the DPO loss mentioned in the paper and the code repository. You may ignore the following parameters: ipo, reference_free and label_smoothing.", "retrieval_content": null, "retrieval_context": [], "unit_test_file": "unit_test/unit_test_1.py" } ]
2	Language Models as Hierarchy Encoders	02-HITs	https://arxiv.org/pdf/2401.11374	2,024	https://github.com/KRR-Oxford/HierarchyTransformers	HierarchyTransformers	[ { "category": "Training Objectives & Optimization Techniques", "class_name": "HierarchyTransformerLoss", "goal_file": "src/hierarchy_transformers/losses/hit_loss.py", "goal_function": "forward", "golden_file": "golden_files/hit_loss_golden.py", "index": 1, "instruction": "Implement the forward function of calulating the Hyperbolic Loss in src/hierarchy_transformers/losses/hit_loss.py based on paper.pdf.", "retrieval_content": null, "retrieval_context": [ "src/hierarchy_transformers/models/hierarchy_transformer/hit.py", "src/hierarchy_transformers/models/hierarchy_transformer/hyperbolic.py" ], "unit_test_file": "unit_test/unit_test_1.py" } ]
3	DoLa: Decoding by Contrasting Layers Improves Factuality in Large Language Models	03-DoLa	https://arxiv.org/abs/2309.03883	2,023	https://github.com/voidism/DoLa.git	DoLa-main	[ { "category": "Decoding & Search Strategies", "class_name": "", "goal_file": "unit_test/unit_test_1.py", "goal_function": "dola_greedy_decode_agent", "golden_file": "golden_files/unit_test_1_golden.py", "index": 1, "instruction": "Implement the function dola_greedy_decode_agent in the unit_test/unit_test_1.py and pass the test by running unit_test/unit_test_1.py file", "retrieval_content": null, "retrieval_context": [], "unit_test_file": "unit_test/unit_test_1.py" } ]
4	Exploring Concept Depth: How Large Language Models Acquire Knowledge and Concept at Different Layers?	04-CD	https://arxiv.org/abs/2404.07066	2,024	https://github.com/Luckfort/CD	CD-main	[ { "category": "Interpretability & Explainability", "class_name": "", "goal_file": "main.py", "goal_function": "probing", "golden_file": "golden_files/main_golden.py", "index": 1, "instruction": "Implement `probing` function in main.py based on paerp.pdf and the repository. Generate Probing function to evaluate whether the model can effectively distinguish between different data based on hidden states from different layers, using logistic regression classifier.", "retrieval_content": [], "retrieval_context": null, "unit_test_file": "unit_test/unit_test_1.py" } ]
5	SimPO: Simple Preference Optimization with a Reference-Free Reward	05-SimPO	https://arxiv.org/abs/2405.14734	2,024	https://github.com/princeton-nlp/SimPO	SimPO	[ { "category": "loss function", "class_name": "SimPOTrainer", "goal_file": "scripts/simpo_trainer.py", "goal_function": "simpo_loss", "golden_file": "golden_files/simpo_trainer_golden.py", "index": 1, "instruction": "Implement the `simpo_loss` function of `SimPOTrainer` class in './scripts/simpo_trainer.py' based on the SimPO loss descripted in the paper.", "retrieval_content": [], "retrieval_context": null, "unit_test_file": "unit_test/unit_test_1.py" } ]
6	Representation Engineering: A Top-Down Approach to AI Transparency	06-RepE	https://arxiv.org/abs/2310.01405	2,023	https://github.com/andyzoujm/representation-engineering	representation-engineering-main	[ { "category": "Feature Learning & Representation", "class_name": "", "goal_file": "unit_test/unit_test_1.py", "goal_function": "get_rep_directions_agent", "golden_file": "golden_files/unit_test_1_golden.py", "index": 1, "instruction": "Implement the function get_rep_directions_agent in the unit_test/unit_test_1.py and pass the test by running unit_test/unit_test_1.py file", "retrieval_content": null, "retrieval_context": [], "unit_test_file": "unit_test/unit_test_1.py" } ]
7	BERTScore: Evaluating Text Generation with BERT	07-bertscore	https://openreview.net/pdf?id=SkeHuCVFDr	2,020	https://github.com/Tiiiger/bert_score/tree/master	bert_score	[ { "category": "Evaluation Metrics", "class_name": "", "goal_file": "bert_score_goal.py", "goal_function": "greedy_cos_idf", "golden_file": "golden_files/bert_score_golden.py", "index": 1, "instruction": "You need to implement the greedy_cos_idf function in bert_score_goal.py based on paper.pdf, which calculates a similarity score between two sentences by using BERT embeddings and performing a greedy matching between all reference and candidate words, with cosine similarity between vector representations as the scoring", "retrieval_content": [], "retrieval_context": null, "unit_test_file": "unit_test/unit_test_1.py" } ]
8	StepLength	08-StepLength	https://arxiv.org/abs/2401.04925	2,024	https://github.com/MingyuJ666/The-Impact-of-Reasoning-Step-Length-on-Large-Language-Models	stepLength-main	[ { "category": "Prompt Engineering & Instruction Tuning", "class_name": "", "goal_file": "run_inference.py", "goal_function": "get_sentence", "golden_file": "golden_files/run_inference_golden.py", "index": 1, "instruction": "Implement `get_sentence` function in run_inference.py based on paper.pdf and the repository. This code constructs the input k for an agent based on the specified prompting method args_method. If the method is 'zero_shot', it appends a direct answer trigger to the question x. If it is 'zero_shot_cot', it appends a chain-of-thought trigger to x. For 'few_shot' and 'few_shot_cot', it prepends the demonstration demo1 to x. If the method is 'auto_cot', it combines demo1, x, and the CoT trigger. The final string k is used as the input for the agent to simulate how the agent would behave under different prompting strtegies.", "retrieval_content": [], "retrieval_context": null, "unit_test_file": "unit_test/unit_test_1.py" } ]
9	Prompt-Based Monte-Carlo Tree Search for Goal-oriented Dialogue Policy Planning	09-GDPZero	https://aclanthology.org/2023.emnlp-main.439.pdf	2,023	https://github.com/jasonyux/GDPZero	GDPZero	[ { "category": "Decoding & Search Strategies", "class_name": "OpenLoopMCTS", "goal_file": "core/mcts.py", "goal_function": "find_best_action", "golden_file": "golden_files/mcts_golden.py", "index": 1, "instruction": "Implement the self.find_best_action function in the OpenLoopMCTS class in core/mcts.py based on the PUCT formula mentioned in the paper and the code repository.", "retrieval_content": [], "retrieval_context": null, "unit_test_file": "unit_test/unit_test_1.py" } ]

11

MoverScore: Text Generation Evaluating with Contextualized Embeddings and Earth Mover Distance

11-moverscore

https://aclanthology.org/D19-1053.pdf

2,019

https://github.com/AIPHES/emnlp19-moverscore

moverscore

[ { "category": "Evaluation Metrics & Benchmarking", "class_name": "", "goal_file": "movescore.py", "goal_function": "word_mover_score", "golden_file": "golden_files/movescore_golden.py", "index": 1, "instruction": "You need to implement the word_mover_score function in movescore.py based on paper.pdf, which calculates a similarity score between two sentences by using BERT embeddings and by calculating calculating the value of 1-Word Mover's Distance metric", "retrieval_content": [], "retrieval_context": null, "unit_test_file": "unit_test/unit_test_1.py" } ]

13

Probing the Decision Boundaries of In-context Learning in Large Language Models

13-DecBound

https://arxiv.org/pdf/2406.11233

2,024

https://github.com/siyan-zhao/ICL_decision_boundary

DecBound-main

[ { "category": "Data Augmentation & Generation", "class_name": "", "goal_file": "data_utils.py", "goal_function": "generate_tasks", "golden_file": "golden_files/data_utils_golden.py", "index": 1, "instruction": "Implement `generate_tasks` function in data_utils.py based on paper.pdf and the repository, which generates multiple machine learning tasks, each represented as a separate dataset.", "retrieval_content": [], "retrieval_context": null, "unit_test_file": "unit_test/unit_test.py" } ]

14

EmojiPrompt: Generative Prompt Obfuscation for Privacy-Preserving Communication with Cloud-based LLMs

14-EmojiCrypt

https://arxiv.org/pdf/2402.05868

2,025

https://github.com/agiresearch/EmojiCrypt

EmojiCrypt-main

[ { "category": "Evaluation Metrics", "class_name": "", "goal_file": "Tabular.py", "goal_function": "compute_mean_cosine_similarity", "golden_file": "golden_files/Tabular_golden.py", "index": 1, "instruction": "Implement `compute_mean_cosine_similarity` function in Tabular.py based on the paper.pdf and the repository. Computes the mean cosine similarity between pairs of encrypted and decrypted feature levels.", "retrieval_content": [], "retrieval_context": null, "unit_test_file": "unit_test/unit_test_1.py" } ]

15

Language Models Predict Empathy Gaps Between Social In-groups and Out-groups

15-EmpathyBias

https://arxiv.org/pdf/2503.01030

2,025

https://github.com/houyu0930/intergroup-empathy-bias

EmpathyBias-main

[ { "category": "Evaluation Metrics", "class_name": "", "goal_file": "analysis/analysis_after_processing.py", "goal_function": "get_delta", "golden_file": "golden_files/analysis_after_processing_golden.py", "index": 1, "instruction": "Implement `get_delta` function analysis/analysis_after_processing.py based on paper.pdf and the repository. Compute the difference in average intensity between in-group and out-group interactions.", "retrieval_content": [], "retrieval_context": null, "unit_test_file": "unit_test/unit_test_1.py" }, { "category": "Evaluation Metrics", "class_name": "", "goal_file": "analysis/analysis_after_processing.py", "goal_function": "get_filtered_matrix", "golden_file": "golden_files/analysis_after_processing_golden.py", "index": 2, "instruction": "Implement `get_filtered_matrix` function analysis/analysis_after_processing.py based on paper.pdf and the repository. Compute the standardized emotion intensity matrix for a specific group, model, and prompt variation, excluding specified data points (by ID).", "retrieval_content": [], "retrieval_context": null, "unit_test_file": "unit_test/unit_test_2.py" } ]

16

Probing Language Models on Their Knowledge Source

16-KnowProb

https://aclanthology.org/2024.blackboxnlp-1.35.pdf

2,024

https://github.com/Zineddine-Tighidet/knowledge-probing-framework

KnowProb-main

[ { "category": "Interpretability & Explainability", "class_name": "", "goal_file": "src/classification/classifier.py", "goal_function": "perform_classification_by_relation_group", "golden_file": "golden_files/classifier_golden.py", "index": 1, "instruction": "Implement `perform_classification_by_relation_group` function in src/classification/classifier.py based on paper.pdf and the repository. This function performs classification over the saved activations, the classification can be module-wise and token-wise (i.e. vertical).", "retrieval_content": [], "retrieval_context": null, "unit_test_file": "unit_test/unit_test_1.py" }, { "category": "Data Augmentation & Generation", "class_name": "", "goal_file": "src/data.py", "goal_function": "remove_object_subject_overlap", "golden_file": "golden_files/data_golden.py", "index": 2, "instruction": "Implement `remove_object_subject_overlap` function in src/data.py based on paper.pdf and the repository. This function deletes the examples where the subject and the parametric object are similar using a Jaro-Winkler string similarity.", "retrieval_content": [], "retrieval_context": null, "unit_test_file": "unit_test/unit_test_2.py" }, { "category": "Information Extraction & Knowledge Integration", "class_name": "", "goal_file": "src/parametric_knowledge.py", "goal_function": "is_parametric_object_not_in_the_prompt", "golden_file": "golden_files/parametric_knowledge_golden.py", "index": 3, "instruction": "Implement `is_parametric_object_not_in_the_prompt` function in src/parametric_knowledge.py based on paper.pdf and the repository. This function checks whether the parametric_object is included in the one-shot examples that were used to guide the LLM during parametric knowledge building as it would mean that it's biased by the prompt.", "retrieval_content": [], "retrieval_context": null, "unit_test_file": "unit_test/unit_test_3.py" }, { "category": "Information Extraction & Knowledge Integration", "class_name": "", "goal_file": "src/data.py", "goal_function": "return_entity_overlap_between_relation_groups", "golden_file": "golden_files/data_golden.py", "index": 4, "instruction": "Implement `return_entity_overlap_between_relation_groups` function in src/data.py based on paper.pdf and the repository. This function returns a dict of dicts representing a kind of matrix of shape (nb_relation_groups, nb_relation_groups), where each value is the number of examples in a relation group that have the same subject or object in another relation group.", "retrieval_content": [], "retrieval_context": null, "unit_test_file": "unit_test/unit_test_4.py" }, { "category": "Data Augmentation & Generation", "class_name": "", "goal_file": "src/data.py", "goal_function": "generate_counter_parametric_knowledge_dataset", "golden_file": "golden_files/data_golden.py", "index": 5, "instruction": "Implement `generate_counter_parametric_knowledge_dataset` function in src/data.py based on paper.pdf and the repository. This function generates the counter-parametric knowledge using the parametric knowledge.", "retrieval_content": [], "retrieval_context": null, "unit_test_file": "unit_test/unit_test_5.py" } ]

18

Massive Activations in Large Language Models

18-MassActiv

https://arxiv.org/abs/2402.17762

2,024

https://github.com/locuslab/massive-activations

MassActiv-main

[ { "category": "Interpretability & Explainability", "class_name": "", "goal_file": "main_vit.py", "goal_function": "run_exp1", "golden_file": "golden_files/main_vit_golden.py", "index": 1, "instruction": "Implement run_exp1 function in main_vit.py. Run 3D feature visualization for a specific layer in a Vision Transformer (ViT) model. This function extracts the absolute feature activations from a specified transformer layer, visualizes them in 3D, and saves both the input parameters and feature activations as serialized pickle files for reproducibility and analysis.", "retrieval_content": [], "retrieval_context": null, "unit_test_file": "unit_test/unit_test.py" } ]

19

Native Sparse Attention: Hardware-Aligned and Natively Trainable Sparse Attention

19-Native-Sparse-Attention

https://arxiv.org/abs/2502.11089

2,025

https://github.com/fla-org/native-sparse-attention

NSA

[ { "category": "Neural Network Architectures & Modules", "class_name": "", "goal_file": "NSA.py", "goal_function": "nsa", "golden_file": "golden_files/NSA_golden.py", "index": 1, "instruction": "Please implement the `nsa` function in NSA.py based on the Natively trainable Sparse Attention mechanism mentioned in the paper.", "retrieval_content": [], "retrieval_context": null, "unit_test_file": "unit_test/unit_test_1.py" } ]

20

Trusting Your Evidence: Hallucinate Less with Context-aware Decoding

20-CAD

https://arxiv.org/abs/2305.14739

2,023

https://github.com/xhan77/context-aware-decoding.git

context-aware-decoding-main

[ { "category": "Decoding & Search Strategies", "class_name": "", "goal_file": "unit_test/unit_test_1.py", "goal_function": "context_aware_sampling", "golden_file": "golden_files/unit_test_1_golden.py", "index": 1, "instruction": "Implement the function context_aware_sampling in the unit_test/unit_test_1.py and pass the test by running unit_test/unit_test_1.py file, you may ignore the following parameters: alpha, max_length, temperature", "retrieval_content": null, "retrieval_context": [], "unit_test_file": "unit_test/unit_test_1.py" } ]

22

Token-level Direct Preference Optimization

22-TokenDPO

https://arxiv.org/abs/2404.11999

2,024

https://github.com/Vance0124/Token-level-Direct-Preference-Optimization/tree/master

TokenDPO-main

[ { "category": "Training Objectives & Optimization Techniques", "class_name": "", "goal_file": "trainers.py", "goal_function": "tdpo_loss", "golden_file": "golden_files/trainers_golden.py", "index": 1, "instruction": "Implement the `tdpo_loss` function in trainers.py based on the Token-level DPO loss mentioned in paper.pdf and the code repository. You may ignore the following parameters: if_tdpo2.", "retrieval_content": null, "retrieval_context": [], "unit_test_file": "unit_test/unit_test_1.py" } ]

23

A Simple Framework for Contrastive Learning of Visual Representations

23-SimCLR

https://arxiv.org/pdf/2002.05709

2,020

https://github.com/skywalkerzhang/SimCLR/tree/master

SimCLR-main

[ { "category": "Training Objectives & Optimization Techniques", "class_name": "SimCLR", "goal_file": "simclr.py", "goal_function": "info_nce_loss", "golden_file": "golden_files/simclr_golden.py", "index": 1, "instruction": "Implement the `info_nce_loss` method of `SimCLR` class in simclr.py that computes the InfoNCE loss for a batch of features, based on the method described in the paper.pdf and the repository.", "retrieval_content": null, "retrieval_context": [], "unit_test_file": "unit_test/unit_test_1.py" } ]

25

AdaLoRA: Adaptive Budget Allocation for Parameter-Efficient Fine-Tuning

25-AdaLora

https://arxiv.org/pdf/2303.10512

2,023

https://github.com/QingruZhang/AdaLoRA/tree/main

AdaLoRA

[ { "category": "Neural Network Architectures & Modules", "class_name": "SVDLinear", "goal_file": "loralib/loralib/adalora.py", "goal_function": "forward", "golden_file": "golden_files/adalora_golden.py", "index": 1, "instruction": "Implement the forward function of SVD-based Adaptated linear layer in loralib/loralib/adalora.py mentioned in the section 3.1 of the paper.", "retrieval_content": [], "retrieval_context": null, "unit_test_file": "unit_test/unit_test_1.py" } ]

28

Self-Play Fine-Tuning Converts Weak Language Models to Strong Language Models

28-SPIN

https://arxiv.org/abs/2401.01335

2,024

https://github.com/uclaml/SPIN

SPIN

[ { "category": "loss function", "class_name": "SPINTrainer", "goal_file": "spin/alignment/trainer.py", "goal_function": "spin_loss", "golden_file": "golden_files/trainer_golden.py", "index": 1, "instruction": "Implement the `spin_loss` method of `SPINTrainer` class in `.spin/alignment/trainer.py` based on the loss function of SPIN algorithm mentioned in the paper.pdf.", "retrieval_content": [], "retrieval_context": null, "unit_test_file": "unit_test/unit_test_1.py" } ]

30

ERGO: Event Relational Graph Transformer for Document-level Event Causality Identification

30-ERGO

https://aclanthology.org/2022.coling-1.185.pdf

2,022

https://github.com/chenmeiqii/ERGO

ERGO

[ { "category": "Training Objectives & Optimization Techniques", "class_name": "focal_loss", "goal_file": "model.py", "goal_function": "forward", "golden_file": "golden_files/model_golden.py", "index": 1, "instruction": "Implement the loss function in the focal_loss class in model.py based on the final adaptive focal loss mentioned in the paper and the code repository. Justify whether summing or averaging the loss based on the `self.size_average` value.", "retrieval_content": null, "retrieval_context": [], "unit_test_file": "unit_test/unit_test_1.py" } ]

31

Aspect-Based Sentiment Analysis with Syntax-Opinion-Sentiment Reasoning Chain

31-Syn-Chain

https://aclanthology.org/2025.coling-main.210.pdf

2,025

https://github.com/rf-x/Syn-Chain-ABSA

Syn-Chain-ABSA

[ { "category": "", "class_name": "", "goal_file": "conll_tree.py", "goal_function": "spacy_result_to_conll", "golden_file": "golden_files/conll_tree_golden.py", "index": 1, "instruction": "Implement the spacy_result_to_conll function in conll_tree.py to generate a dependency parse in the CoNLL-U format, as mentioned in the paper and the code repository.", "retrieval_content": [], "retrieval_context": null, "unit_test_file": "unit_test/unit_test_1.py" } ]

1

Direct Preference Optimization: Your Language Model is Secretly a Reward Model

01-DPO

https://arxiv.org/pdf/2305.18290

2,023

https://github.com/eric-mitchell/direct-preference-optimization

direct-preference-optimization

[ { "category": "Training Objectives & Optimization Techniques", "class_name": "", "goal_file": "trainers.py", "goal_function": "preference_loss", "golden_file": "golden_files/trainers_golden.py", "index": 1, "instruction": "Implement the preference_loss function in trainers.py based on the DPO loss mentioned in the paper and the code repository. You may ignore the following parameters: ipo, reference_free and label_smoothing.", "retrieval_content": null, "retrieval_context": [], "unit_test_file": "unit_test/unit_test_1.py" } ]

2

Language Models as Hierarchy Encoders

02-HITs

https://arxiv.org/pdf/2401.11374

2,024

https://github.com/KRR-Oxford/HierarchyTransformers

HierarchyTransformers

[ { "category": "Training Objectives & Optimization Techniques", "class_name": "HierarchyTransformerLoss", "goal_file": "src/hierarchy_transformers/losses/hit_loss.py", "goal_function": "forward", "golden_file": "golden_files/hit_loss_golden.py", "index": 1, "instruction": "Implement the forward function of calulating the Hyperbolic Loss in src/hierarchy_transformers/losses/hit_loss.py based on paper.pdf.", "retrieval_content": null, "retrieval_context": [ "src/hierarchy_transformers/models/hierarchy_transformer/hit.py", "src/hierarchy_transformers/models/hierarchy_transformer/hyperbolic.py" ], "unit_test_file": "unit_test/unit_test_1.py" } ]

3

DoLa: Decoding by Contrasting Layers Improves Factuality in Large Language Models

03-DoLa

https://arxiv.org/abs/2309.03883

2,023

https://github.com/voidism/DoLa.git

DoLa-main

[ { "category": "Decoding & Search Strategies", "class_name": "", "goal_file": "unit_test/unit_test_1.py", "goal_function": "dola_greedy_decode_agent", "golden_file": "golden_files/unit_test_1_golden.py", "index": 1, "instruction": "Implement the function dola_greedy_decode_agent in the unit_test/unit_test_1.py and pass the test by running unit_test/unit_test_1.py file", "retrieval_content": null, "retrieval_context": [], "unit_test_file": "unit_test/unit_test_1.py" } ]

4

Exploring Concept Depth: How Large Language Models Acquire Knowledge and Concept at Different Layers?

04-CD

https://arxiv.org/abs/2404.07066

2,024

https://github.com/Luckfort/CD

CD-main

[ { "category": "Interpretability & Explainability", "class_name": "", "goal_file": "main.py", "goal_function": "probing", "golden_file": "golden_files/main_golden.py", "index": 1, "instruction": "Implement `probing` function in main.py based on paerp.pdf and the repository. Generate Probing function to evaluate whether the model can effectively distinguish between different data based on hidden states from different layers, using logistic regression classifier.", "retrieval_content": [], "retrieval_context": null, "unit_test_file": "unit_test/unit_test_1.py" } ]

5

SimPO: Simple Preference Optimization with a Reference-Free Reward

05-SimPO

https://arxiv.org/abs/2405.14734

2,024

https://github.com/princeton-nlp/SimPO

SimPO

[ { "category": "loss function", "class_name": "SimPOTrainer", "goal_file": "scripts/simpo_trainer.py", "goal_function": "simpo_loss", "golden_file": "golden_files/simpo_trainer_golden.py", "index": 1, "instruction": "Implement the `simpo_loss` function of `SimPOTrainer` class in './scripts/simpo_trainer.py' based on the SimPO loss descripted in the paper.", "retrieval_content": [], "retrieval_context": null, "unit_test_file": "unit_test/unit_test_1.py" } ]

6

Representation Engineering: A Top-Down Approach to AI Transparency

06-RepE

https://arxiv.org/abs/2310.01405

2,023

https://github.com/andyzoujm/representation-engineering

representation-engineering-main

[ { "category": "Feature Learning & Representation", "class_name": "", "goal_file": "unit_test/unit_test_1.py", "goal_function": "get_rep_directions_agent", "golden_file": "golden_files/unit_test_1_golden.py", "index": 1, "instruction": "Implement the function get_rep_directions_agent in the unit_test/unit_test_1.py and pass the test by running unit_test/unit_test_1.py file", "retrieval_content": null, "retrieval_context": [], "unit_test_file": "unit_test/unit_test_1.py" } ]

7

BERTScore: Evaluating Text Generation with BERT

07-bertscore

https://openreview.net/pdf?id=SkeHuCVFDr

2,020

https://github.com/Tiiiger/bert_score/tree/master

bert_score

[ { "category": "Evaluation Metrics", "class_name": "", "goal_file": "bert_score_goal.py", "goal_function": "greedy_cos_idf", "golden_file": "golden_files/bert_score_golden.py", "index": 1, "instruction": "You need to implement the greedy_cos_idf function in bert_score_goal.py based on paper.pdf, which calculates a similarity score between two sentences by using BERT embeddings and performing a greedy matching between all reference and candidate words, with cosine similarity between vector representations as the scoring", "retrieval_content": [], "retrieval_context": null, "unit_test_file": "unit_test/unit_test_1.py" } ]

8

StepLength

08-StepLength

https://arxiv.org/abs/2401.04925

2,024

https://github.com/MingyuJ666/The-Impact-of-Reasoning-Step-Length-on-Large-Language-Models

stepLength-main

[ { "category": "Prompt Engineering & Instruction Tuning", "class_name": "", "goal_file": "run_inference.py", "goal_function": "get_sentence", "golden_file": "golden_files/run_inference_golden.py", "index": 1, "instruction": "Implement `get_sentence` function in run_inference.py based on paper.pdf and the repository. This code constructs the input k for an agent based on the specified prompting method args_method. If the method is 'zero_shot', it appends a direct answer trigger to the question x. If it is 'zero_shot_cot', it appends a chain-of-thought trigger to x. For 'few_shot' and 'few_shot_cot', it prepends the demonstration demo1 to x. If the method is 'auto_cot', it combines demo1, x, and the CoT trigger. The final string k is used as the input for the agent to simulate how the agent would behave under different prompting strtegies.", "retrieval_content": [], "retrieval_context": null, "unit_test_file": "unit_test/unit_test_1.py" } ]

9

Prompt-Based Monte-Carlo Tree Search for Goal-oriented Dialogue Policy Planning

09-GDPZero

https://aclanthology.org/2023.emnlp-main.439.pdf

2,023

https://github.com/jasonyux/GDPZero

GDPZero

[ { "category": "Decoding & Search Strategies", "class_name": "OpenLoopMCTS", "goal_file": "core/mcts.py", "goal_function": "find_best_action", "golden_file": "golden_files/mcts_golden.py", "index": 1, "instruction": "Implement the self.find_best_action function in the OpenLoopMCTS class in core/mcts.py based on the PUCT formula mentioned in the paper and the code repository.", "retrieval_content": [], "retrieval_context": null, "unit_test_file": "unit_test/unit_test_1.py" } ]