datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
hle_math_category_phi4code-reasoning-phi4-templatePhi4-ensemble-teacher-forcing-record-logits-dataphi-4-eval-logs-and-scoresCaseHOLD_Phi4_Reasoning
CaseHOLD_Phi4_Reasoning
Overview
This repository provides reasoning annotations generated using the Phi-4 language model for the CaseHOLD dataset, a legal question answering benchmark based on U.S. case law.
The purpose of this dataset is to support research on legal reasoning, explainable legal QA, and reasoning-augmented language models. The reasoning annotations are intended for research and educational use, particularly in the context of Legal NLP and Legal AI.… See the full description on the dataset page: https://huggingface.co/datasets/nguyenkhanh87/CaseHOLD_Phi4_Reasoning.medical-reasoning-processed_phi4_sfthle_ja_phi4https://huggingface.co/datasets/cais/hle
Humanity's Last Examのquestionをphi4で日本語訳したものです。
どのような質問があるかの中身の確認用です。きれいに出力されているかはあまり確認していません。
中身の確認用ページ(初回ロード遅い)https://if001.github.io/hle_sample/
いくつか出力が途切れているものがあります。
Input IDs of length 7215 > the model's max sequence length of 4096.Input IDs of length 7348 > the model's max sequence length of 4096.Input IDs of length 9698 > the model's max sequence length of 4096.Input IDs of length 10144 > the model's max sequence length of 4096.Input IDs of… See the full description on the dataset page: https://huggingface.co/datasets/if001/hle_ja_phi4.arm_o0_phi4_multi_full_14_allbenhaotang__phi4-qwq-sky-t1-details
Dataset Card for Evaluation run of benhaotang/phi4-qwq-sky-t1
Dataset automatically created during the evaluation run of model benhaotang/phi4-qwq-sky-t1
The dataset is composed of 38 configuration(s), each one corresponding to one of the evaluated task.
The dataset has been created from 1 run(s). Each run can be found as a specific split in each configuration, the split being named using the timestamp of the run.The "train" split is always pointing to the latest results.
An… See the full description on the dataset page: https://huggingface.co/datasets/open-llm-leaderboard/benhaotang__phi4-qwq-sky-t1-details.EpistemeAI__DeepThinkers-Phi4-details
Dataset Card for Evaluation run of EpistemeAI/DeepThinkers-Phi4
Dataset automatically created during the evaluation run of model EpistemeAI/DeepThinkers-Phi4
The dataset is composed of 38 configuration(s), each one corresponding to one of the evaluated task.
The dataset has been created from 1 run(s). Each run can be found as a specific split in each configuration, the split being named using the timestamp of the run.The "train" split is always pointing to the latest results.
An… See the full description on the dataset page: https://huggingface.co/datasets/open-llm-leaderboard/EpistemeAI__DeepThinkers-Phi4-details.Triangle104__Phi4-RP-o1-Ablit-details
Dataset Card for Evaluation run of Triangle104/Phi4-RP-o1-Ablit
Dataset automatically created during the evaluation run of model Triangle104/Phi4-RP-o1-Ablit
The dataset is composed of 38 configuration(s), each one corresponding to one of the evaluated task.
The dataset has been created from 1 run(s). Each run can be found as a specific split in each configuration, the split being named using the timestamp of the run.The "train" split is always pointing to the latest results.
An… See the full description on the dataset page: https://huggingface.co/datasets/open-llm-leaderboard/Triangle104__Phi4-RP-o1-Ablit-details.Quazim0t0__Phi4Basis-14B-sce-details
Dataset Card for Evaluation run of Quazim0t0/Phi4Basis-14B-sce
Dataset automatically created during the evaluation run of model Quazim0t0/Phi4Basis-14B-sce
The dataset is composed of 38 configuration(s), each one corresponding to one of the evaluated task.
The dataset has been created from 2 run(s). Each run can be found as a specific split in each configuration, the split being named using the timestamp of the run.The "train" split is always pointing to the latest results.
An… See the full description on the dataset page: https://huggingface.co/datasets/open-llm-leaderboard/Quazim0t0__Phi4Basis-14B-sce-details.phi4-function-calling-dataset-v3chemistry-reasoning-phi4-vs-deepseekv3CaseHOLD_DeepSeek_R1_14B_Phi4_Reasoningpython_all_phi4_maintainmsmarco_hard_negatives_phi4-14bThe data was used in the paper Don't Retrieve, Generate: Prompting LLMs for Synthetic Training Data in Dense Retrieval.
Hard-Negatives generated by Phi4-14B using 10,000 passages from the MS-Marco dataset
If this dataset was useful consider citing us :)
@misc{sinha2025dontretrievegenerateprompting,
title={Don't Retrieve, Generate: Prompting LLMs for Synthetic Training Data in Dense Retrieval},
author={Aarush Sinha},
year={2025},
eprint={2504.21015}… See the full description on the dataset page: https://huggingface.co/datasets/chungimungi/msmarco_hard_negatives_phi4-14b.Phi4Millennium-SF-Mathematical-Reasoning
Phi4Millennium-SF
Dataset 2 of 5 in the Atem Training Pipeline
Phi4Millennium-SF is a supervised fine-tuning (SFT) dataset of 2,932
mathematical and structured reasoning examples generated by
microsoft/phi-4 with structured chain-of-thought prompting. It forms
the second component of the distillation pipeline used to train
Atem, a Qwen2.5-1.5B-based model targeting analytical reasoning
performance beyond its parameter class.
Why Phi-4 as a Teacher Model
Phi-4… See the full description on the dataset page: https://huggingface.co/datasets/EphAsad/Phi4Millennium-SF-Mathematical-Reasoning.phi4-conversationsRaw responses generated by Phi4 , questions from alamios/Mistral-Small-24B-Instruct-2501-Conversations
Made it to use on the QwenPhi 0.5B Draft model, but the finetune did not yield much improvement, still I have generated the dataset so here is the raw data hopefully it is useful for someone.
arm_o0_phi4_multi_smoketestPhi4SmolDScommit_all_phi4_maintainPhi4RS_11phi4-medical-preprocessedfinetome-phi4-format
FineTome-Phi4-Format
This dataset contains 15000 samples from the mlabonne/FineTome-100k dataset,
formatted specifically for Phi-4 models with the system prompt set to "detailed thinking off".
Format
Each example follows the Phi chat format:
<|im_start|>system<|im_sep|>detailed thinking off<|im_end|>
<|im_start|>user<|im_sep|>[User content]<|im_end|>
<|im_start|>assistant<|im_sep|>[Assistant response]<|im_end|>
Source
The original data comes from… See the full description on the dataset page: https://huggingface.co/datasets/ykarout/finetome-phi4-format.Phi-4-mini-reasoning-secalignGAPO_data_phi4Phi-4-reasoning-plus_eval_5554
mlfoundations-dev/Phi-4-reasoning-plus_eval_5554
Precomputed model outputs for evaluation.
Evaluation Results
Summary
Metric
AIME24
AMC23
MATH500
MMLUPro
JEEBench
GPQADiamond
LiveCodeBench
CodeElo
CodeForces
HLE
HMMT
AIME25
LiveCodeBenchv5
Accuracy
76.0
96.2
84.0
14.6
83.5
66.8
0.8
2.4
3.5
7.1
53.0
68.0
0.5
AIME24
Average Accuracy: 76.00% ± 1.23%
Number of Runs: 10
Run
Accuracy
Questions Solved
Total Questions
1… See the full description on the dataset page: https://huggingface.co/datasets/mlfoundations-dev/Phi-4-reasoning-plus_eval_5554.Phi4RS_7Phi4RS_23