id stringlengths 2 115 | lastModified stringlengths 24 24 | tags list | author stringlengths 2 42 ⌀ | description stringlengths 0 6.67k ⌀ | citation stringlengths 0 10.7k ⌀ | likes int64 0 3.66k | downloads int64 0 8.89M | created timestamp[us] | card stringlengths 11 977k | card_len int64 11 977k | embeddings list |
|---|---|---|---|---|---|---|---|---|---|---|---|
Maxx0/testingagain | 2023-10-12T08:56:49.000Z | [
"region:us"
] | Maxx0 | null | null | 0 | 0 | 2023-10-12T08:56:27 | Entry not found | 15 | [
[
-0.021392822265625,
-0.0149688720703125,
0.057220458984375,
0.0288238525390625,
-0.03509521484375,
0.046539306640625,
0.052520751953125,
0.005046844482421875,
0.0513916015625,
0.016998291015625,
-0.052093505859375,
-0.01495361328125,
-0.060333251953125,
0.03... |
tr416/client_base_model_dataset_20231012_092752 | 2023-10-12T09:27:53.000Z | [
"region:us"
] | tr416 | null | null | 0 | 0 | 2023-10-12T09:27:53 | Entry not found | 15 | [
[
-0.021392822265625,
-0.0149688720703125,
0.057220458984375,
0.0288238525390625,
-0.03509521484375,
0.046539306640625,
0.052520751953125,
0.005046844482421875,
0.0513916015625,
0.016998291015625,
-0.052093505859375,
-0.01495361328125,
-0.060333251953125,
0.03... |
tr416/client_base_dataset_20231012_092907 | 2023-10-12T09:29:07.000Z | [
"region:us"
] | tr416 | null | null | 0 | 0 | 2023-10-12T09:29:07 | Entry not found | 15 | [
[
-0.021392822265625,
-0.0149688720703125,
0.057220458984375,
0.0288238525390625,
-0.03509521484375,
0.046539306640625,
0.052520751953125,
0.005046844482421875,
0.0513916015625,
0.016998291015625,
-0.052093505859375,
-0.01495361328125,
-0.060333251953125,
0.03... |
tr416/base_model_client_dataset_20231012_093051 | 2023-10-12T09:30:53.000Z | [
"region:us"
] | tr416 | null | null | 0 | 0 | 2023-10-12T09:30:51 | ---
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
- split: test
path: data/test-*
dataset_info:
features:
- name: input_ids
sequence: int32
- name: attention_mask
sequence: int8
splits:
- name: train
num_bytes: 75203880.0
num_examples: 29285
- name: test
num_bytes: 760128.0
num_examples: 296
download_size: 12773623
dataset_size: 75964008.0
---
# Dataset Card for "base_model_client_dataset_20231012_093051"
[More Information needed](https://github.com/huggingface/datasets/blob/main/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards) | 627 | [
[
-0.035064697265625,
-0.0160064697265625,
0.0101776123046875,
0.027557373046875,
-0.002185821533203125,
-0.01560211181640625,
0.030975341796875,
-0.01001739501953125,
0.04595947265625,
0.05841064453125,
-0.057891845703125,
-0.044158935546875,
-0.0251922607421875,... |
open-llm-leaderboard/details_harborwater__open-llama-3b-everything-v2 | 2023-10-29T00:44:10.000Z | [
"region:us"
] | open-llm-leaderboard | null | null | 0 | 0 | 2023-10-12T09:37:29 | ---
pretty_name: Evaluation run of harborwater/open-llama-3b-everything-v2
dataset_summary: "Dataset automatically created during the evaluation run of model\
\ [harborwater/open-llama-3b-everything-v2](https://huggingface.co/harborwater/open-llama-3b-everything-v2)\
\ on the [Open LLM Leaderboard](https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard).\n\
\nThe dataset is composed of 64 configuration, each one coresponding to one of the\
\ evaluated task.\n\nThe 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.\n\
\nAn additional configuration \"results\" store all the aggregated results of the\
\ run (and is used to compute and display the agregated metrics on the [Open LLM\
\ Leaderboard](https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard)).\n\
\nTo load the details from a run, you can for instance do the following:\n```python\n\
from datasets import load_dataset\ndata = load_dataset(\"open-llm-leaderboard/details_harborwater__open-llama-3b-everything-v2\"\
,\n\t\"harness_winogrande_5\",\n\tsplit=\"train\")\n```\n\n## Latest results\n\n\
These are the [latest results from run 2023-10-29T00:43:57.732775](https://huggingface.co/datasets/open-llm-leaderboard/details_harborwater__open-llama-3b-everything-v2/blob/main/results_2023-10-29T00-43-57.732775.json)(note\
\ that their might be results for other tasks in the repos if successive evals didn't\
\ cover the same tasks. You find each in the results and the \"latest\" split for\
\ each eval):\n\n```python\n{\n \"all\": {\n \"em\": 0.0020973154362416107,\n\
\ \"em_stderr\": 0.0004685065030368325,\n \"f1\": 0.0560864093959733,\n\
\ \"f1_stderr\": 0.0013597729822813858,\n \"acc\": 0.341030820866541,\n\
\ \"acc_stderr\": 0.008350924483766176\n },\n \"harness|drop|3\": {\n\
\ \"em\": 0.0020973154362416107,\n \"em_stderr\": 0.0004685065030368325,\n\
\ \"f1\": 0.0560864093959733,\n \"f1_stderr\": 0.0013597729822813858\n\
\ },\n \"harness|gsm8k|5\": {\n \"acc\": 0.01592115238817286,\n \
\ \"acc_stderr\": 0.0034478192723889915\n },\n \"harness|winogrande|5\"\
: {\n \"acc\": 0.6661404893449092,\n \"acc_stderr\": 0.013254029695143358\n\
\ }\n}\n```"
repo_url: https://huggingface.co/harborwater/open-llama-3b-everything-v2
leaderboard_url: https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard
point_of_contact: clementine@hf.co
configs:
- config_name: harness_arc_challenge_25
data_files:
- split: 2023_10_12T09_37_10.252705
path:
- '**/details_harness|arc:challenge|25_2023-10-12T09-37-10.252705.parquet'
- split: latest
path:
- '**/details_harness|arc:challenge|25_2023-10-12T09-37-10.252705.parquet'
- config_name: harness_drop_3
data_files:
- split: 2023_10_29T00_43_57.732775
path:
- '**/details_harness|drop|3_2023-10-29T00-43-57.732775.parquet'
- split: latest
path:
- '**/details_harness|drop|3_2023-10-29T00-43-57.732775.parquet'
- config_name: harness_gsm8k_5
data_files:
- split: 2023_10_29T00_43_57.732775
path:
- '**/details_harness|gsm8k|5_2023-10-29T00-43-57.732775.parquet'
- split: latest
path:
- '**/details_harness|gsm8k|5_2023-10-29T00-43-57.732775.parquet'
- config_name: harness_hellaswag_10
data_files:
- split: 2023_10_12T09_37_10.252705
path:
- '**/details_harness|hellaswag|10_2023-10-12T09-37-10.252705.parquet'
- split: latest
path:
- '**/details_harness|hellaswag|10_2023-10-12T09-37-10.252705.parquet'
- config_name: harness_hendrycksTest_5
data_files:
- split: 2023_10_12T09_37_10.252705
path:
- '**/details_harness|hendrycksTest-abstract_algebra|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-anatomy|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-astronomy|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-business_ethics|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-clinical_knowledge|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-college_biology|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-college_chemistry|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-college_computer_science|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-college_mathematics|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-college_medicine|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-college_physics|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-computer_security|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-conceptual_physics|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-econometrics|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-electrical_engineering|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-elementary_mathematics|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-formal_logic|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-global_facts|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-high_school_biology|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-high_school_chemistry|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-high_school_computer_science|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-high_school_european_history|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-high_school_geography|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-high_school_government_and_politics|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-high_school_macroeconomics|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-high_school_mathematics|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-high_school_microeconomics|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-high_school_physics|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-high_school_psychology|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-high_school_statistics|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-high_school_us_history|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-high_school_world_history|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-human_aging|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-human_sexuality|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-international_law|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-jurisprudence|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-logical_fallacies|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-machine_learning|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-management|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-marketing|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-medical_genetics|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-miscellaneous|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-moral_disputes|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-moral_scenarios|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-nutrition|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-philosophy|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-prehistory|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-professional_accounting|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-professional_law|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-professional_medicine|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-professional_psychology|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-public_relations|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-security_studies|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-sociology|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-us_foreign_policy|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-virology|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-world_religions|5_2023-10-12T09-37-10.252705.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-abstract_algebra|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-anatomy|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-astronomy|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-business_ethics|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-clinical_knowledge|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-college_biology|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-college_chemistry|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-college_computer_science|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-college_mathematics|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-college_medicine|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-college_physics|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-computer_security|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-conceptual_physics|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-econometrics|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-electrical_engineering|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-elementary_mathematics|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-formal_logic|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-global_facts|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-high_school_biology|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-high_school_chemistry|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-high_school_computer_science|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-high_school_european_history|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-high_school_geography|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-high_school_government_and_politics|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-high_school_macroeconomics|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-high_school_mathematics|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-high_school_microeconomics|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-high_school_physics|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-high_school_psychology|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-high_school_statistics|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-high_school_us_history|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-high_school_world_history|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-human_aging|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-human_sexuality|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-international_law|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-jurisprudence|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-logical_fallacies|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-machine_learning|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-management|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-marketing|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-medical_genetics|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-miscellaneous|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-moral_disputes|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-moral_scenarios|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-nutrition|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-philosophy|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-prehistory|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-professional_accounting|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-professional_law|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-professional_medicine|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-professional_psychology|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-public_relations|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-security_studies|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-sociology|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-us_foreign_policy|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-virology|5_2023-10-12T09-37-10.252705.parquet'
- '**/details_harness|hendrycksTest-world_religions|5_2023-10-12T09-37-10.252705.parquet'
- config_name: harness_hendrycksTest_abstract_algebra_5
data_files:
- split: 2023_10_12T09_37_10.252705
path:
- '**/details_harness|hendrycksTest-abstract_algebra|5_2023-10-12T09-37-10.252705.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-abstract_algebra|5_2023-10-12T09-37-10.252705.parquet'
- config_name: harness_hendrycksTest_anatomy_5
data_files:
- split: 2023_10_12T09_37_10.252705
path:
- '**/details_harness|hendrycksTest-anatomy|5_2023-10-12T09-37-10.252705.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-anatomy|5_2023-10-12T09-37-10.252705.parquet'
- config_name: harness_hendrycksTest_astronomy_5
data_files:
- split: 2023_10_12T09_37_10.252705
path:
- '**/details_harness|hendrycksTest-astronomy|5_2023-10-12T09-37-10.252705.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-astronomy|5_2023-10-12T09-37-10.252705.parquet'
- config_name: harness_hendrycksTest_business_ethics_5
data_files:
- split: 2023_10_12T09_37_10.252705
path:
- '**/details_harness|hendrycksTest-business_ethics|5_2023-10-12T09-37-10.252705.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-business_ethics|5_2023-10-12T09-37-10.252705.parquet'
- config_name: harness_hendrycksTest_clinical_knowledge_5
data_files:
- split: 2023_10_12T09_37_10.252705
path:
- '**/details_harness|hendrycksTest-clinical_knowledge|5_2023-10-12T09-37-10.252705.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-clinical_knowledge|5_2023-10-12T09-37-10.252705.parquet'
- config_name: harness_hendrycksTest_college_biology_5
data_files:
- split: 2023_10_12T09_37_10.252705
path:
- '**/details_harness|hendrycksTest-college_biology|5_2023-10-12T09-37-10.252705.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-college_biology|5_2023-10-12T09-37-10.252705.parquet'
- config_name: harness_hendrycksTest_college_chemistry_5
data_files:
- split: 2023_10_12T09_37_10.252705
path:
- '**/details_harness|hendrycksTest-college_chemistry|5_2023-10-12T09-37-10.252705.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-college_chemistry|5_2023-10-12T09-37-10.252705.parquet'
- config_name: harness_hendrycksTest_college_computer_science_5
data_files:
- split: 2023_10_12T09_37_10.252705
path:
- '**/details_harness|hendrycksTest-college_computer_science|5_2023-10-12T09-37-10.252705.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-college_computer_science|5_2023-10-12T09-37-10.252705.parquet'
- config_name: harness_hendrycksTest_college_mathematics_5
data_files:
- split: 2023_10_12T09_37_10.252705
path:
- '**/details_harness|hendrycksTest-college_mathematics|5_2023-10-12T09-37-10.252705.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-college_mathematics|5_2023-10-12T09-37-10.252705.parquet'
- config_name: harness_hendrycksTest_college_medicine_5
data_files:
- split: 2023_10_12T09_37_10.252705
path:
- '**/details_harness|hendrycksTest-college_medicine|5_2023-10-12T09-37-10.252705.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-college_medicine|5_2023-10-12T09-37-10.252705.parquet'
- config_name: harness_hendrycksTest_college_physics_5
data_files:
- split: 2023_10_12T09_37_10.252705
path:
- '**/details_harness|hendrycksTest-college_physics|5_2023-10-12T09-37-10.252705.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-college_physics|5_2023-10-12T09-37-10.252705.parquet'
- config_name: harness_hendrycksTest_computer_security_5
data_files:
- split: 2023_10_12T09_37_10.252705
path:
- '**/details_harness|hendrycksTest-computer_security|5_2023-10-12T09-37-10.252705.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-computer_security|5_2023-10-12T09-37-10.252705.parquet'
- config_name: harness_hendrycksTest_conceptual_physics_5
data_files:
- split: 2023_10_12T09_37_10.252705
path:
- '**/details_harness|hendrycksTest-conceptual_physics|5_2023-10-12T09-37-10.252705.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-conceptual_physics|5_2023-10-12T09-37-10.252705.parquet'
- config_name: harness_hendrycksTest_econometrics_5
data_files:
- split: 2023_10_12T09_37_10.252705
path:
- '**/details_harness|hendrycksTest-econometrics|5_2023-10-12T09-37-10.252705.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-econometrics|5_2023-10-12T09-37-10.252705.parquet'
- config_name: harness_hendrycksTest_electrical_engineering_5
data_files:
- split: 2023_10_12T09_37_10.252705
path:
- '**/details_harness|hendrycksTest-electrical_engineering|5_2023-10-12T09-37-10.252705.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-electrical_engineering|5_2023-10-12T09-37-10.252705.parquet'
- config_name: harness_hendrycksTest_elementary_mathematics_5
data_files:
- split: 2023_10_12T09_37_10.252705
path:
- '**/details_harness|hendrycksTest-elementary_mathematics|5_2023-10-12T09-37-10.252705.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-elementary_mathematics|5_2023-10-12T09-37-10.252705.parquet'
- config_name: harness_hendrycksTest_formal_logic_5
data_files:
- split: 2023_10_12T09_37_10.252705
path:
- '**/details_harness|hendrycksTest-formal_logic|5_2023-10-12T09-37-10.252705.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-formal_logic|5_2023-10-12T09-37-10.252705.parquet'
- config_name: harness_hendrycksTest_global_facts_5
data_files:
- split: 2023_10_12T09_37_10.252705
path:
- '**/details_harness|hendrycksTest-global_facts|5_2023-10-12T09-37-10.252705.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-global_facts|5_2023-10-12T09-37-10.252705.parquet'
- config_name: harness_hendrycksTest_high_school_biology_5
data_files:
- split: 2023_10_12T09_37_10.252705
path:
- '**/details_harness|hendrycksTest-high_school_biology|5_2023-10-12T09-37-10.252705.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-high_school_biology|5_2023-10-12T09-37-10.252705.parquet'
- config_name: harness_hendrycksTest_high_school_chemistry_5
data_files:
- split: 2023_10_12T09_37_10.252705
path:
- '**/details_harness|hendrycksTest-high_school_chemistry|5_2023-10-12T09-37-10.252705.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-high_school_chemistry|5_2023-10-12T09-37-10.252705.parquet'
- config_name: harness_hendrycksTest_high_school_computer_science_5
data_files:
- split: 2023_10_12T09_37_10.252705
path:
- '**/details_harness|hendrycksTest-high_school_computer_science|5_2023-10-12T09-37-10.252705.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-high_school_computer_science|5_2023-10-12T09-37-10.252705.parquet'
- config_name: harness_hendrycksTest_high_school_european_history_5
data_files:
- split: 2023_10_12T09_37_10.252705
path:
- '**/details_harness|hendrycksTest-high_school_european_history|5_2023-10-12T09-37-10.252705.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-high_school_european_history|5_2023-10-12T09-37-10.252705.parquet'
- config_name: harness_hendrycksTest_high_school_geography_5
data_files:
- split: 2023_10_12T09_37_10.252705
path:
- '**/details_harness|hendrycksTest-high_school_geography|5_2023-10-12T09-37-10.252705.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-high_school_geography|5_2023-10-12T09-37-10.252705.parquet'
- config_name: harness_hendrycksTest_high_school_government_and_politics_5
data_files:
- split: 2023_10_12T09_37_10.252705
path:
- '**/details_harness|hendrycksTest-high_school_government_and_politics|5_2023-10-12T09-37-10.252705.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-high_school_government_and_politics|5_2023-10-12T09-37-10.252705.parquet'
- config_name: harness_hendrycksTest_high_school_macroeconomics_5
data_files:
- split: 2023_10_12T09_37_10.252705
path:
- '**/details_harness|hendrycksTest-high_school_macroeconomics|5_2023-10-12T09-37-10.252705.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-high_school_macroeconomics|5_2023-10-12T09-37-10.252705.parquet'
- config_name: harness_hendrycksTest_high_school_mathematics_5
data_files:
- split: 2023_10_12T09_37_10.252705
path:
- '**/details_harness|hendrycksTest-high_school_mathematics|5_2023-10-12T09-37-10.252705.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-high_school_mathematics|5_2023-10-12T09-37-10.252705.parquet'
- config_name: harness_hendrycksTest_high_school_microeconomics_5
data_files:
- split: 2023_10_12T09_37_10.252705
path:
- '**/details_harness|hendrycksTest-high_school_microeconomics|5_2023-10-12T09-37-10.252705.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-high_school_microeconomics|5_2023-10-12T09-37-10.252705.parquet'
- config_name: harness_hendrycksTest_high_school_physics_5
data_files:
- split: 2023_10_12T09_37_10.252705
path:
- '**/details_harness|hendrycksTest-high_school_physics|5_2023-10-12T09-37-10.252705.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-high_school_physics|5_2023-10-12T09-37-10.252705.parquet'
- config_name: harness_hendrycksTest_high_school_psychology_5
data_files:
- split: 2023_10_12T09_37_10.252705
path:
- '**/details_harness|hendrycksTest-high_school_psychology|5_2023-10-12T09-37-10.252705.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-high_school_psychology|5_2023-10-12T09-37-10.252705.parquet'
- config_name: harness_hendrycksTest_high_school_statistics_5
data_files:
- split: 2023_10_12T09_37_10.252705
path:
- '**/details_harness|hendrycksTest-high_school_statistics|5_2023-10-12T09-37-10.252705.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-high_school_statistics|5_2023-10-12T09-37-10.252705.parquet'
- config_name: harness_hendrycksTest_high_school_us_history_5
data_files:
- split: 2023_10_12T09_37_10.252705
path:
- '**/details_harness|hendrycksTest-high_school_us_history|5_2023-10-12T09-37-10.252705.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-high_school_us_history|5_2023-10-12T09-37-10.252705.parquet'
- config_name: harness_hendrycksTest_high_school_world_history_5
data_files:
- split: 2023_10_12T09_37_10.252705
path:
- '**/details_harness|hendrycksTest-high_school_world_history|5_2023-10-12T09-37-10.252705.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-high_school_world_history|5_2023-10-12T09-37-10.252705.parquet'
- config_name: harness_hendrycksTest_human_aging_5
data_files:
- split: 2023_10_12T09_37_10.252705
path:
- '**/details_harness|hendrycksTest-human_aging|5_2023-10-12T09-37-10.252705.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-human_aging|5_2023-10-12T09-37-10.252705.parquet'
- config_name: harness_hendrycksTest_human_sexuality_5
data_files:
- split: 2023_10_12T09_37_10.252705
path:
- '**/details_harness|hendrycksTest-human_sexuality|5_2023-10-12T09-37-10.252705.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-human_sexuality|5_2023-10-12T09-37-10.252705.parquet'
- config_name: harness_hendrycksTest_international_law_5
data_files:
- split: 2023_10_12T09_37_10.252705
path:
- '**/details_harness|hendrycksTest-international_law|5_2023-10-12T09-37-10.252705.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-international_law|5_2023-10-12T09-37-10.252705.parquet'
- config_name: harness_hendrycksTest_jurisprudence_5
data_files:
- split: 2023_10_12T09_37_10.252705
path:
- '**/details_harness|hendrycksTest-jurisprudence|5_2023-10-12T09-37-10.252705.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-jurisprudence|5_2023-10-12T09-37-10.252705.parquet'
- config_name: harness_hendrycksTest_logical_fallacies_5
data_files:
- split: 2023_10_12T09_37_10.252705
path:
- '**/details_harness|hendrycksTest-logical_fallacies|5_2023-10-12T09-37-10.252705.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-logical_fallacies|5_2023-10-12T09-37-10.252705.parquet'
- config_name: harness_hendrycksTest_machine_learning_5
data_files:
- split: 2023_10_12T09_37_10.252705
path:
- '**/details_harness|hendrycksTest-machine_learning|5_2023-10-12T09-37-10.252705.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-machine_learning|5_2023-10-12T09-37-10.252705.parquet'
- config_name: harness_hendrycksTest_management_5
data_files:
- split: 2023_10_12T09_37_10.252705
path:
- '**/details_harness|hendrycksTest-management|5_2023-10-12T09-37-10.252705.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-management|5_2023-10-12T09-37-10.252705.parquet'
- config_name: harness_hendrycksTest_marketing_5
data_files:
- split: 2023_10_12T09_37_10.252705
path:
- '**/details_harness|hendrycksTest-marketing|5_2023-10-12T09-37-10.252705.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-marketing|5_2023-10-12T09-37-10.252705.parquet'
- config_name: harness_hendrycksTest_medical_genetics_5
data_files:
- split: 2023_10_12T09_37_10.252705
path:
- '**/details_harness|hendrycksTest-medical_genetics|5_2023-10-12T09-37-10.252705.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-medical_genetics|5_2023-10-12T09-37-10.252705.parquet'
- config_name: harness_hendrycksTest_miscellaneous_5
data_files:
- split: 2023_10_12T09_37_10.252705
path:
- '**/details_harness|hendrycksTest-miscellaneous|5_2023-10-12T09-37-10.252705.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-miscellaneous|5_2023-10-12T09-37-10.252705.parquet'
- config_name: harness_hendrycksTest_moral_disputes_5
data_files:
- split: 2023_10_12T09_37_10.252705
path:
- '**/details_harness|hendrycksTest-moral_disputes|5_2023-10-12T09-37-10.252705.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-moral_disputes|5_2023-10-12T09-37-10.252705.parquet'
- config_name: harness_hendrycksTest_moral_scenarios_5
data_files:
- split: 2023_10_12T09_37_10.252705
path:
- '**/details_harness|hendrycksTest-moral_scenarios|5_2023-10-12T09-37-10.252705.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-moral_scenarios|5_2023-10-12T09-37-10.252705.parquet'
- config_name: harness_hendrycksTest_nutrition_5
data_files:
- split: 2023_10_12T09_37_10.252705
path:
- '**/details_harness|hendrycksTest-nutrition|5_2023-10-12T09-37-10.252705.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-nutrition|5_2023-10-12T09-37-10.252705.parquet'
- config_name: harness_hendrycksTest_philosophy_5
data_files:
- split: 2023_10_12T09_37_10.252705
path:
- '**/details_harness|hendrycksTest-philosophy|5_2023-10-12T09-37-10.252705.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-philosophy|5_2023-10-12T09-37-10.252705.parquet'
- config_name: harness_hendrycksTest_prehistory_5
data_files:
- split: 2023_10_12T09_37_10.252705
path:
- '**/details_harness|hendrycksTest-prehistory|5_2023-10-12T09-37-10.252705.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-prehistory|5_2023-10-12T09-37-10.252705.parquet'
- config_name: harness_hendrycksTest_professional_accounting_5
data_files:
- split: 2023_10_12T09_37_10.252705
path:
- '**/details_harness|hendrycksTest-professional_accounting|5_2023-10-12T09-37-10.252705.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-professional_accounting|5_2023-10-12T09-37-10.252705.parquet'
- config_name: harness_hendrycksTest_professional_law_5
data_files:
- split: 2023_10_12T09_37_10.252705
path:
- '**/details_harness|hendrycksTest-professional_law|5_2023-10-12T09-37-10.252705.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-professional_law|5_2023-10-12T09-37-10.252705.parquet'
- config_name: harness_hendrycksTest_professional_medicine_5
data_files:
- split: 2023_10_12T09_37_10.252705
path:
- '**/details_harness|hendrycksTest-professional_medicine|5_2023-10-12T09-37-10.252705.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-professional_medicine|5_2023-10-12T09-37-10.252705.parquet'
- config_name: harness_hendrycksTest_professional_psychology_5
data_files:
- split: 2023_10_12T09_37_10.252705
path:
- '**/details_harness|hendrycksTest-professional_psychology|5_2023-10-12T09-37-10.252705.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-professional_psychology|5_2023-10-12T09-37-10.252705.parquet'
- config_name: harness_hendrycksTest_public_relations_5
data_files:
- split: 2023_10_12T09_37_10.252705
path:
- '**/details_harness|hendrycksTest-public_relations|5_2023-10-12T09-37-10.252705.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-public_relations|5_2023-10-12T09-37-10.252705.parquet'
- config_name: harness_hendrycksTest_security_studies_5
data_files:
- split: 2023_10_12T09_37_10.252705
path:
- '**/details_harness|hendrycksTest-security_studies|5_2023-10-12T09-37-10.252705.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-security_studies|5_2023-10-12T09-37-10.252705.parquet'
- config_name: harness_hendrycksTest_sociology_5
data_files:
- split: 2023_10_12T09_37_10.252705
path:
- '**/details_harness|hendrycksTest-sociology|5_2023-10-12T09-37-10.252705.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-sociology|5_2023-10-12T09-37-10.252705.parquet'
- config_name: harness_hendrycksTest_us_foreign_policy_5
data_files:
- split: 2023_10_12T09_37_10.252705
path:
- '**/details_harness|hendrycksTest-us_foreign_policy|5_2023-10-12T09-37-10.252705.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-us_foreign_policy|5_2023-10-12T09-37-10.252705.parquet'
- config_name: harness_hendrycksTest_virology_5
data_files:
- split: 2023_10_12T09_37_10.252705
path:
- '**/details_harness|hendrycksTest-virology|5_2023-10-12T09-37-10.252705.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-virology|5_2023-10-12T09-37-10.252705.parquet'
- config_name: harness_hendrycksTest_world_religions_5
data_files:
- split: 2023_10_12T09_37_10.252705
path:
- '**/details_harness|hendrycksTest-world_religions|5_2023-10-12T09-37-10.252705.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-world_religions|5_2023-10-12T09-37-10.252705.parquet'
- config_name: harness_truthfulqa_mc_0
data_files:
- split: 2023_10_12T09_37_10.252705
path:
- '**/details_harness|truthfulqa:mc|0_2023-10-12T09-37-10.252705.parquet'
- split: latest
path:
- '**/details_harness|truthfulqa:mc|0_2023-10-12T09-37-10.252705.parquet'
- config_name: harness_winogrande_5
data_files:
- split: 2023_10_29T00_43_57.732775
path:
- '**/details_harness|winogrande|5_2023-10-29T00-43-57.732775.parquet'
- split: latest
path:
- '**/details_harness|winogrande|5_2023-10-29T00-43-57.732775.parquet'
- config_name: results
data_files:
- split: 2023_10_12T09_37_10.252705
path:
- results_2023-10-12T09-37-10.252705.parquet
- split: 2023_10_29T00_43_57.732775
path:
- results_2023-10-29T00-43-57.732775.parquet
- split: latest
path:
- results_2023-10-29T00-43-57.732775.parquet
---
# Dataset Card for Evaluation run of harborwater/open-llama-3b-everything-v2
## Dataset Description
- **Homepage:**
- **Repository:** https://huggingface.co/harborwater/open-llama-3b-everything-v2
- **Paper:**
- **Leaderboard:** https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard
- **Point of Contact:** clementine@hf.co
### Dataset Summary
Dataset automatically created during the evaluation run of model [harborwater/open-llama-3b-everything-v2](https://huggingface.co/harborwater/open-llama-3b-everything-v2) on the [Open LLM Leaderboard](https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard).
The dataset is composed of 64 configuration, each one coresponding 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 additional configuration "results" store all the aggregated results of the run (and is used to compute and display the agregated metrics on the [Open LLM Leaderboard](https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard)).
To load the details from a run, you can for instance do the following:
```python
from datasets import load_dataset
data = load_dataset("open-llm-leaderboard/details_harborwater__open-llama-3b-everything-v2",
"harness_winogrande_5",
split="train")
```
## Latest results
These are the [latest results from run 2023-10-29T00:43:57.732775](https://huggingface.co/datasets/open-llm-leaderboard/details_harborwater__open-llama-3b-everything-v2/blob/main/results_2023-10-29T00-43-57.732775.json)(note that their might be results for other tasks in the repos if successive evals didn't cover the same tasks. You find each in the results and the "latest" split for each eval):
```python
{
"all": {
"em": 0.0020973154362416107,
"em_stderr": 0.0004685065030368325,
"f1": 0.0560864093959733,
"f1_stderr": 0.0013597729822813858,
"acc": 0.341030820866541,
"acc_stderr": 0.008350924483766176
},
"harness|drop|3": {
"em": 0.0020973154362416107,
"em_stderr": 0.0004685065030368325,
"f1": 0.0560864093959733,
"f1_stderr": 0.0013597729822813858
},
"harness|gsm8k|5": {
"acc": 0.01592115238817286,
"acc_stderr": 0.0034478192723889915
},
"harness|winogrande|5": {
"acc": 0.6661404893449092,
"acc_stderr": 0.013254029695143358
}
}
```
### Supported Tasks and Leaderboards
[More Information Needed]
### Languages
[More Information Needed]
## Dataset Structure
### Data Instances
[More Information Needed]
### Data Fields
[More Information Needed]
### Data Splits
[More Information Needed]
## Dataset Creation
### Curation Rationale
[More Information Needed]
### Source Data
#### Initial Data Collection and Normalization
[More Information Needed]
#### Who are the source language producers?
[More Information Needed]
### Annotations
#### Annotation process
[More Information Needed]
#### Who are the annotators?
[More Information Needed]
### Personal and Sensitive Information
[More Information Needed]
## Considerations for Using the Data
### Social Impact of Dataset
[More Information Needed]
### Discussion of Biases
[More Information Needed]
### Other Known Limitations
[More Information Needed]
## Additional Information
### Dataset Curators
[More Information Needed]
### Licensing Information
[More Information Needed]
### Citation Information
[More Information Needed]
### Contributions
[More Information Needed] | 38,792 | [
[
-0.0247344970703125,
-0.049407958984375,
0.0254364013671875,
0.0217742919921875,
-0.0181732177734375,
0.01145172119140625,
-0.0161895751953125,
-0.018646240234375,
0.02764892578125,
0.042816162109375,
-0.045806884765625,
-0.0723876953125,
-0.051971435546875,
... |
CCRs/chatgpt-paraphrases-kz | 2023-10-12T11:10:51.000Z | [
"region:us"
] | CCRs | null | null | 0 | 0 | 2023-10-12T11:10:51 | Entry not found | 15 | [
[
-0.02142333984375,
-0.01495361328125,
0.05718994140625,
0.0288238525390625,
-0.035064697265625,
0.046539306640625,
0.052520751953125,
0.005062103271484375,
0.0513916015625,
0.016998291015625,
-0.052093505859375,
-0.014984130859375,
-0.060394287109375,
0.0379... |
Thanmay/indic-xnli | 2023-10-13T13:46:57.000Z | [
"region:us"
] | Thanmay | null | null | 0 | 0 | 2023-10-12T11:55:45 | ---
configs:
- config_name: default
data_files:
- split: test
path: data/test-*
- split: validation
path: data/validation-*
dataset_info:
features:
- name: premise
dtype: string
- name: hypothesis
dtype: string
- name: label
dtype:
class_label:
names:
'0': entailment
'1': neutral
'2': contradiction
- name: itv2 hi premise
dtype: string
- name: itv2 hi hypothesis
dtype: string
- name: itv2 gu premise
dtype: string
- name: itv2 gu hypothesis
dtype: string
- name: itv2 kn premise
dtype: string
- name: itv2 kn hypothesis
dtype: string
- name: itv2 ml premise
dtype: string
- name: itv2 ml hypothesis
dtype: string
- name: itv2 mr premise
dtype: string
- name: itv2 mr hypothesis
dtype: string
- name: itv2 or premise
dtype: string
- name: itv2 or hypothesis
dtype: string
- name: itv2 pa premise
dtype: string
- name: itv2 pa hypothesis
dtype: string
- name: itv2 bn premise
dtype: string
- name: itv2 bn hypothesis
dtype: string
splits:
- name: test
num_bytes: 8389920
num_examples: 5010
- name: validation
num_bytes: 4161518
num_examples: 2490
download_size: 4269813
dataset_size: 12551438
---
# Dataset Card for "indic-xnli"
[More Information needed](https://github.com/huggingface/datasets/blob/main/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards) | 1,462 | [
[
-0.04107666015625,
0.010894775390625,
-0.0012874603271484375,
0.01457977294921875,
-0.0130767822265625,
0.016510009765625,
0.0077972412109375,
-0.03155517578125,
0.0767822265625,
0.027496337890625,
-0.06427001953125,
-0.0513916015625,
-0.0208892822265625,
-0... |
Asaad101/DT_data_0 | 2023-10-12T12:13:36.000Z | [
"region:us"
] | Asaad101 | null | null | 0 | 0 | 2023-10-12T12:13:23 | ---
dataset_info:
features:
- name: images
sequence: image
- name: actions
sequence: string
splits:
- name: train
num_bytes: 14340602.0
num_examples: 23
download_size: 4232
dataset_size: 14340602.0
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
---
# Dataset Card for "DT_data_0"
[More Information needed](https://github.com/huggingface/datasets/blob/main/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards) | 481 | [
[
-0.0408935546875,
-0.03045654296875,
0.019805908203125,
0.01067352294921875,
-0.0207977294921875,
0.00319671630859375,
0.035247802734375,
-0.0023746490478515625,
0.06915283203125,
0.025909423828125,
-0.05950927734375,
-0.060943603515625,
-0.045074462890625,
... |
Asaad101/DT_data_1 | 2023-10-12T12:13:37.000Z | [
"region:us"
] | Asaad101 | null | null | 0 | 0 | 2023-10-12T12:13:37 | Entry not found | 15 | [
[
-0.02142333984375,
-0.01495361328125,
0.05718994140625,
0.0288238525390625,
-0.035064697265625,
0.046539306640625,
0.052520751953125,
0.005062103271484375,
0.0513916015625,
0.016998291015625,
-0.052093505859375,
-0.014984130859375,
-0.060394287109375,
0.0379... |
ostapeno/platy_4iter_SUB_10_icl5_mD1000_prmp00 | 2023-10-12T12:31:53.000Z | [
"region:us"
] | ostapeno | null | null | 0 | 0 | 2023-10-12T12:28:37 | Dataset generated with the setting:
- python projects/wiki_experts/cli_qa_creator.py e2e --model_setting=platy --n_icl=5 --sub_names=SUB_10 --num_iterations=4 --max_documents_per_subject=1000 --upload_to_hub=1
Standard prompts for response and instructions (0,0). Model setting platy. | 287 | [
[
-0.039215087890625,
-0.038726806640625,
0.0244598388671875,
0.001659393310546875,
0.00804901123046875,
-0.031829833984375,
0.0037994384765625,
0.017669677734375,
0.00240325927734375,
0.0589599609375,
-0.064697265625,
-0.0211639404296875,
-0.01168060302734375,
... |
reidlai/ohlcv.csv | 2023-10-12T12:52:19.000Z | [
"region:us"
] | reidlai | null | null | 0 | 0 | 2023-10-12T12:51:29 | Entry not found | 15 | [
[
-0.02142333984375,
-0.01495361328125,
0.05718994140625,
0.0288238525390625,
-0.035064697265625,
0.046539306640625,
0.052520751953125,
0.005062103271484375,
0.0513916015625,
0.016998291015625,
-0.052093505859375,
-0.014984130859375,
-0.060394287109375,
0.0379... |
erictsai/Prompt_Mask | 2023-10-12T12:59:51.000Z | [
"region:us"
] | erictsai | null | null | 1 | 0 | 2023-10-12T12:59:51 | Entry not found | 15 | [
[
-0.02142333984375,
-0.01495361328125,
0.05718994140625,
0.0288238525390625,
-0.035064697265625,
0.046539306640625,
0.052520751953125,
0.005062103271484375,
0.0513916015625,
0.016998291015625,
-0.052093505859375,
-0.014984130859375,
-0.060394287109375,
0.0379... |
SerjaoDaNasa/nando1 | 2023-10-22T17:03:28.000Z | [
"region:us"
] | SerjaoDaNasa | null | null | 0 | 0 | 2023-10-12T13:02:58 | Entry not found | 15 | [
[
-0.02142333984375,
-0.01495361328125,
0.05718994140625,
0.0288238525390625,
-0.035064697265625,
0.046539306640625,
0.052520751953125,
0.005062103271484375,
0.0513916015625,
0.016998291015625,
-0.052093505859375,
-0.014984130859375,
-0.060394287109375,
0.0379... |
ovi054/video-data | 2023-10-12T13:23:17.000Z | [
"region:us"
] | ovi054 | null | null | 0 | 0 | 2023-10-12T13:08:02 | ---
configs:
- config_name: default
data_files:
- split: train
path: data.csv
---
# Dataset Card for Dataset Name
<!-- Provide a quick summary of the dataset. -->
## Dataset Details
### Dataset Description
<!-- Provide a longer summary of what this dataset is. -->
- **Curated by:** [More Information Needed]
- **Funded by [optional]:** [More Information Needed]
- **Shared by [optional]:** [More Information Needed]
- **Language(s) (NLP):** [More Information Needed]
- **License:** [More Information Needed]
### Dataset Sources [optional]
<!-- Provide the basic links for the dataset. -->
- **Repository:** [More Information Needed]
- **Paper [optional]:** [More Information Needed]
- **Demo [optional]:** [More Information Needed]
## Uses
<!-- Address questions around how the dataset is intended to be used. -->
### Direct Use
<!-- This section describes suitable use cases for the dataset. -->
[More Information Needed]
### Out-of-Scope Use
<!-- This section addresses misuse, malicious use, and uses that the dataset will not work well for. -->
[More Information Needed]
## Dataset Structure
<!-- This section provides a description of the dataset fields, and additional information about the dataset structure such as criteria used to create the splits, relationships between data points, etc. -->
[More Information Needed]
## Dataset Creation
### Curation Rationale
<!-- Motivation for the creation of this dataset. -->
[More Information Needed]
### Source Data
<!-- This section describes the source data (e.g. news text and headlines, social media posts, translated sentences, ...). -->
#### Data Collection and Processing
<!-- This section describes the data collection and processing process such as data selection criteria, filtering and normalization methods, tools and libraries used, etc. -->
[More Information Needed]
#### Who are the source data producers?
<!-- This section describes the people or systems who originally created the data. It should also include self-reported demographic or identity information for the source data creators if this information is available. -->
[More Information Needed]
### Annotations [optional]
<!-- If the dataset contains annotations which are not part of the initial data collection, use this section to describe them. -->
#### Annotation process
<!-- This section describes the annotation process such as annotation tools used in the process, the amount of data annotated, annotation guidelines provided to the annotators, interannotator statistics, annotation validation, etc. -->
[More Information Needed]
#### Who are the annotators?
<!-- This section describes the people or systems who created the annotations. -->
[More Information Needed]
#### Personal and Sensitive Information
<!-- State whether the dataset contains data that might be considered personal, sensitive, or private (e.g., data that reveals addresses, uniquely identifiable names or aliases, racial or ethnic origins, sexual orientations, religious beliefs, political opinions, financial or health data, etc.). If efforts were made to anonymize the data, describe the anonymization process. -->
[More Information Needed]
## Bias, Risks, and Limitations
<!-- This section is meant to convey both technical and sociotechnical limitations. -->
[More Information Needed]
### Recommendations
<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
Users should be made aware of the risks, biases and limitations of the dataset. More information needed for further recommendations.
## Citation [optional]
<!-- If there is a paper or blog post introducing the dataset, the APA and Bibtex information for that should go in this section. -->
**BibTeX:**
[More Information Needed]
**APA:**
[More Information Needed]
## Glossary [optional]
<!-- If relevant, include terms and calculations in this section that can help readers understand the dataset or dataset card. -->
[More Information Needed]
## More Information [optional]
[More Information Needed]
## Dataset Card Authors [optional]
[More Information Needed]
## Dataset Card Contact
[More Information Needed] | 4,220 | [
[
-0.037261962890625,
-0.037994384765625,
0.013671875,
0.01546478271484375,
-0.0257110595703125,
-0.004154205322265625,
-0.0016651153564453125,
-0.04644775390625,
0.033935546875,
0.059967041015625,
-0.05645751953125,
-0.0716552734375,
-0.039520263671875,
0.009... |
ostapeno/insruction_evolution | 2023-10-12T13:09:07.000Z | [
"region:us"
] | ostapeno | null | null | 0 | 0 | 2023-10-12T13:09:05 | ---
dataset_info:
features:
- name: instruction_0
dtype: string
- name: id
dtype: string
- name: instruction_1
dtype: string
- name: instruction_2
dtype: string
- name: instruction_3
dtype: string
splits:
- name: train
num_bytes: 11982727
num_examples: 23121
download_size: 5433213
dataset_size: 11982727
---
# Dataset Card for "insruction_evolution"
[More Information needed](https://github.com/huggingface/datasets/blob/main/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards) | 531 | [
[
-0.037567138671875,
-0.025726318359375,
0.008758544921875,
0.009735107421875,
-0.017486572265625,
0.01285552978515625,
0.024810791015625,
-0.01560211181640625,
0.05401611328125,
0.02679443359375,
-0.072021484375,
-0.03277587890625,
-0.0266876220703125,
-0.00... |
sanchit-gandhi/rev16_csv | 2023-10-12T13:33:07.000Z | [
"region:us"
] | sanchit-gandhi | null | null | 0 | 0 | 2023-10-12T13:09:35 | Entry not found | 15 | [
[
-0.02142333984375,
-0.01495361328125,
0.05718994140625,
0.0288238525390625,
-0.035064697265625,
0.046539306640625,
0.052520751953125,
0.005062103271484375,
0.0513916015625,
0.016998291015625,
-0.052093505859375,
-0.014984130859375,
-0.060394287109375,
0.0379... |
ostapeno/platy_icl5_maxD10_maxC1000000_prmt00_3 | 2023-10-12T13:15:40.000Z | [
"region:us"
] | ostapeno | null | null | 0 | 0 | 2023-10-12T13:15:29 | ## model_setting_name: platy
## max_context_length: 512
## icl_examples: 5
## icl_dataset_name: lukaemon/mmlu
## max_documents_per_subject: 10
## max_contexts_per_subject: 1000000
## icl_use_out_options: True
## seed_dataset: sordonia/my-wiki-latex_mmlu_from_valid_all
## subjects: SUB_10
## response_template: 0
## inverse_template: 0
| 336 | [
[
-0.03680419921875,
-0.024566650390625,
0.0261383056640625,
0.034759521484375,
-0.02874755859375,
-0.0195770263671875,
-0.00327301025390625,
0.01727294921875,
-0.0081787109375,
0.032470703125,
-0.06512451171875,
-0.039581298828125,
-0.0271148681640625,
0.0190... |
ostapeno/platy_icl5_maxD10_maxC1000000_prmt11_3 | 2023-10-12T13:16:32.000Z | [
"region:us"
] | ostapeno | null | null | 0 | 0 | 2023-10-12T13:16:25 | Entry not found | 15 | [
[
-0.02142333984375,
-0.01495361328125,
0.05718994140625,
0.0288238525390625,
-0.035064697265625,
0.046539306640625,
0.052520751953125,
0.005062103271484375,
0.0513916015625,
0.016998291015625,
-0.052093505859375,
-0.014984130859375,
-0.060394287109375,
0.0379... |
ostapeno/platy_icl5_maxD10_maxC1000000_prmt01_3 | 2023-10-12T13:16:32.000Z | [
"region:us"
] | ostapeno | null | null | 0 | 0 | 2023-10-12T13:16:31 | Entry not found | 15 | [
[
-0.02142333984375,
-0.01495361328125,
0.05718994140625,
0.0288238525390625,
-0.035064697265625,
0.046539306640625,
0.052520751953125,
0.005062103271484375,
0.0513916015625,
0.016998291015625,
-0.052093505859375,
-0.014984130859375,
-0.060394287109375,
0.0379... |
sordonia/platy_icl0_maxD-1_maxC1000_0 | 2023-10-12T13:22:34.000Z | [
"region:us"
] | sordonia | null | null | 0 | 0 | 2023-10-12T13:22:15 | ---
configs:
- config_name: default
data_files:
- split: formal_logic
path: data/formal_logic-*
- split: machine_learning
path: data/machine_learning-*
- split: global_facts
path: data/global_facts-*
- split: abstract_algebra
path: data/abstract_algebra-*
- split: high_school_physics
path: data/high_school_physics-*
- split: college_biology
path: data/college_biology-*
- split: high_school_government_and_politics
path: data/high_school_government_and_politics-*
- split: prehistory
path: data/prehistory-*
- split: security_studies
path: data/security_studies-*
- split: sociology
path: data/sociology-*
dataset_info:
features:
- name: id
dtype: string
- name: context
dtype: string
- name: docno
dtype: string
- name: subject
dtype: string
- name: icl_examples
dtype: 'null'
- name: instruction
dtype: string
- name: author_instr
dtype: string
- name: response
dtype: string
- name: author_response
dtype: string
- name: normalized_cumul_logprob_response
dtype: float64
splits:
- name: formal_logic
num_bytes: 3003267.417225118
num_examples: 863
- name: machine_learning
num_bytes: 2784025.415735915
num_examples: 800
- name: global_facts
num_bytes: 2822305.7652022834
num_examples: 811
- name: abstract_algebra
num_bytes: 2603063.76371308
num_examples: 748
- name: high_school_physics
num_bytes: 2867546.178207992
num_examples: 824
- name: college_biology
num_bytes: 2964987.0677587492
num_examples: 852
- name: high_school_government_and_politics
num_bytes: 2710944.748572847
num_examples: 779
- name: prehistory
num_bytes: 2641344.113179449
num_examples: 759
- name: security_studies
num_bytes: 2860586.1146686524
num_examples: 822
- name: sociology
num_bytes: 2784025.415735915
num_examples: 800
download_size: 15087112
dataset_size: 28042096.000000004
---
# Dataset Card for "platy_icl0_maxD-1_maxC1000_0"
[More Information needed](https://github.com/huggingface/datasets/blob/main/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards) | 2,178 | [
[
-0.04913330078125,
-0.00859832763671875,
0.004024505615234375,
0.031585693359375,
-0.01641845703125,
-0.001247406005859375,
0.01934814453125,
0.009429931640625,
0.049957275390625,
0.045745849609375,
-0.04827880859375,
-0.062286376953125,
-0.047576904296875,
... |
ccmusic-database/CNPM | 2023-10-15T03:32:20.000Z | [
"task_categories:audio-classification",
"size_categories:n<1K",
"language:zh",
"language:en",
"license:mit",
"music",
"art",
"region:us"
] | ccmusic-database | Based on the working idea of combining manual labeling with computer in the construction of World Music Database, this database collects and labels the audio of five modes (including five tones, six tones and seven tones) of "Gong, Shang, Jue, Zhi and Yu". At the same time, it makes a detailed analysis of the judgment of Chinese national pentatonic modes, and finds application scenarios and technical models, which can provide raw data for the analysis and retrieval of Chinese national music characteristics. | @dataset{zhaorui_liu_2021_5676893,
author = {Zhaorui Liu, Monan Zhou, Shenyang Xu, Wei Li and Zijin Li},
title = {CCMUSIC DATABASE: A Music Data Sharing Platform for Computational Musicology Research},
month = {nov},
year = {2021},
publisher = {Zenodo},
version = {1.1},
doi = {10.5281/zenodo.5676893},
url = {https://doi.org/10.5281/zenodo.5676893}
} | 1 | 0 | 2023-10-12T13:22:17 | ---
license: mit
task_categories:
- audio-classification
language:
- zh
- en
tags:
- music
- art
pretty_name: Chinese National Pentatonic Mode Dataset
size_categories:
- n<1K
---
# Dataset Card for Chinese National Pentatonic Mode Dataset
## Dataset Description
- **Homepage:** <https://ccmusic-database.github.io>
- **Repository:** <https://huggingface.co/datasets/ccmusic-database/CNPM>
- **Paper:** <https://doi.org/10.5281/zenodo.5676893>
- **Leaderboard:** <https://ccmusic-database.github.io/team.html>
- **Point of Contact:** Chinese Ethnic Pentatonic Scale; Database; Music Information Retrieval; Pentatonic Therapy
### Dataset Summary
Based on the working idea of combining manual labeling with computer in the construction of World Music Database, this database collects and labels the audio of five modes (including five tones, six tones and seven tones) of "Gong, Shang, Jue, Zhi and Yu". At the same time, it makes a detailed analysis of the judgment of Chinese national pentatonic modes, and finds application scenarios and technical models, which can provide raw data for the analysis and retrieval of Chinese national music characteristics.
### Supported Tasks and Leaderboards
MIR, audio classification
### Languages
Chinese, English
## Dataset Structure
### Data Instances
.zip(.wav), .csv
### Data Fields
Mode Type, Name, Performer, Album Name, National Mode Name, Tonggong System, Audio Links
### Data Splits
train
## Usage
```
from datasets import load_dataset
dataset = load_dataset("ccmusic-dabase/CNPM", split='train')
for data in dataset:
print(data)
```
## Dataset Creation
### Curation Rationale
Lack of a dataset for Chinese National Pentatonic Mode
### Source Data
#### Initial Data Collection and Normalization
Weixin Ren, Mingjin Che, Zhaowen Wang, Qinyu Li, Jiaye Hu, Fan Xia, Wei Li, Monan Zhou
#### Who are the source language producers?
Teachers & students from FD-LAMT, CCOM, SCCM
### Annotations
#### Annotation process
Based on the working idea of combining manual labeling with computer in the construction of World Music Database, this database collects and labels the audio of five modes (including five tones, six tones and seven tones) of "Gong, Shang, Jue, Zhi and Yu". At the same time, it makes a detailed analysis of the judgment of Chinese national pentatonic modes, and finds application scenarios and technical models, which can provide raw data for the analysis and retrieval of Chinese national music characteristics.
#### Who are the annotators?
Teachers & students from FD-LAMT, CCOM, SCCM
### Personal and Sensitive Information
Due to copyright reasons, only some of the audio can be released directly. This part of the audio is the Shang mode and Jue mode tracks performed by professional performers. The rest of the audio needs to be searched and downloaded by the dataset user from music platforms such as Kugou Music, NetEase Cloud Music and QQ Music, based on song titles, artists and album names.
## Considerations for Using the Data
### Social Impact of Dataset
Promoting the development of music AI industry
### Discussion of Biases
Only for Traditional Chinese Instruments
### Other Known Limitations
Only for Pentatonic Mode
## Additional Information
### Dataset Curators
Weixin Ren, Mingjin Che, Zhaowen Wang, Qinyu Li, Jiaye Hu, Fan Xia, Wei Li
### Evalution
[任伟鑫,车明锦,汪照文,孟文武,李沁雨,胡佳弋,夏凡,李伟.CNPM Database:一个用于计算音乐学的中国民族五声调式数据库[J].复旦学报(自然科学版),2022,61(05):555-563.DOI:10.15943/j.cnki.fdxb-jns.20221017.008.](https://kns.cnki.net/kcms2/article/abstract?v=lD5CuVSaeOtw0E2oWliKSMrLiLDt9iwvkwoTgSclPspwUECyt4uNZ6T7DCLlfwMqohXCQXkFzf_XjAUOQ3CAkhPqNj20H8eG9UfUVuHEey0x7Kqp32fMlJiM9xuPtdVMvC1PB2qW0qI=&uniplatform=NZKPT&src=copy)
### Licensing Information
```
MIT License
Copyright (c) FD-LAMT
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.
```
### Citation Information
```
@dataset{zhaorui_liu_2021_5676893,
author = {Zhaorui Liu, Monan Zhou, Shenyang Xu, Wei Li and Zijin Li},
title = {CCMUSIC DATABASE: A Music Data Sharing Platform for Computational Musicology Research},
month = {nov},
year = {2021},
publisher = {Zenodo},
version = {1.1},
doi = {10.5281/zenodo.5676893},
url = {https://doi.org/10.5281/zenodo.5676893}
}
```
### Contributions
Provide a dataset for Chinese National Pentatonic Mode | 5,334 | [
[
-0.03369140625,
-0.030487060546875,
0.0107879638671875,
0.036346435546875,
-0.046142578125,
-0.0179901123046875,
-0.046905517578125,
-0.023162841796875,
0.00971221923828125,
0.041961669921875,
-0.057891845703125,
-0.0732421875,
0.00170135498046875,
0.0097885... |
renumics/spotlight-sayakpaul-nyu_depth_v2-enrichment | 2023-10-12T21:32:53.000Z | [
"region:us"
] | renumics | null | null | 0 | 0 | 2023-10-12T13:23:39 | ---
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
- split: validation
path: data/validation-*
dataset_info:
features:
- name: image.embedding
sequence: float32
length: 2
- name: depth_map.embedding
sequence: float32
length: 2
splits:
- name: train
num_bytes: 761344
num_examples: 47584
- name: validation
num_bytes: 10464
num_examples: 654
download_size: 1073092
dataset_size: 771808
---
# Dataset Card for "spotlight-sayakpaul-nyu_depth_v2-enrichment"
[More Information needed](https://github.com/huggingface/datasets/blob/main/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards) | 679 | [
[
-0.034454345703125,
-0.0216064453125,
0.019775390625,
0.0182647705078125,
-0.0145263671875,
0.00455474853515625,
0.0172119140625,
-0.029937744140625,
0.06298828125,
0.0335693359375,
-0.0482177734375,
-0.039520263671875,
-0.03387451171875,
-0.0164794921875,
... |
ccmusic-database/GZ_IsoTech | 2023-11-01T12:35:02.000Z | [
"task_categories:audio-classification",
"size_categories:n<1K",
"language:zh",
"language:en",
"license:mit",
"music",
"art",
"arxiv:2209.08774",
"region:us"
] | ccmusic-database | null | null | 1 | 0 | 2023-10-12T13:23:57 | ---
license: mit
task_categories:
- audio-classification
language:
- zh
- en
tags:
- music
- art
pretty_name: GZ_IsoTech Dataset
size_categories:
- n<1K
---
# Dataset Card for GZ_IsoTech Dataset
## Dataset Description
- **Homepage:** <https://ccmusic-database.github.io>
- **Repository:** <https://huggingface.co/datasets/ccmusic-database/Guzheng_Tech99>
- **Paper:** <https://doi.org/10.5281/zenodo.5676893>
- **Leaderboard:** <https://ccmusic-database.github.io/team.html>
- **Point of Contact:** <https://arxiv.org/abs/2209.08774>
### Dataset Summary
The Guzheng is a kind of traditional Chinese instruments with diverse playing techniques. Instrument playing techniques (IPT) play an important role in musical performance. However, most of the existing works for IPT detection show low efficiency for variable-length audio and provide no assurance in the generalization as they rely on a single sound bank for training and testing. In this study, we propose an end-to-end Guzheng playing technique detection system using Fully Convolutional Networks that can be applied to variable-length audio. Because each Guzheng playing technique is applied to a note, a dedicated onset detector is trained to divide an audio into several notes and its predictions are fused with frame-wise IPT predictions. During fusion, we add the IPT predictions frame by frame inside each note and get the IPT with the highest probability within each note as the final output of that note. We create a new dataset named GZ_IsoTech from multiple sound banks and real-world recordings for Guzheng performance analysis. Our approach achieves 87.97% in frame-level accuracy and 80.76% in note-level F1-score, outperforming existing works by a large margin, which indicates the effectiveness of our proposed method in IPT detection.
This database contains 2824 audio clips of guzheng playing techniques. Among them, 2328 pieces were collected from virtual sound banks, and 496 pieces were played and recorded by a professional guzheng performer. These clips cover almost all the tones in the range of guzheng and the most commonly used playing techniques in guzheng performance. According to the different playing techniques of guzheng, the clips are divided into 8 categories: Vibrato(chanyin), Upward Portamento(shanghuayin), Downward Portamento(xiahuayin), Returning Portamento(huihuayin), Glissando (guazou, huazhi), Tremolo(yaozhi), Harmonic(fanyin), Plucks(gou,da,mo,tuo…).
### Supported Tasks and Leaderboards
MIR, audio classification
### Languages
Chinese, English
## Dataset Structure
### Data Instances
.zip(.flac, .csv)
### Data Fields
This database contains 2824 audio clips of guzheng playing techniques. Among them, 2328 pieces were collected from virtual sound banks, and 496 pieces were played and recorded by a professional guzheng performer. These clips cover almost all the tones in the range of guzheng and the most commonly used playing techniques in guzheng performance. According to the different playing techniques of guzheng, the clips are divided into 8 categories: Vibrato(chanyin), Upward Portamento(shanghuayin), Downward Portamento(xiahuayin), Returning Portamento(huihuayin), Glissando (guazou, huazhi), Tremolo(yaozhi), Harmonic(fanyin), Plucks(gou,da,mo,tuo…).
### Data Splits
train, valid, test
## Dataset Creation
### Curation Rationale
The Guzheng is a kind of traditional Chinese instruments with diverse playing techniques. Instrument playing techniques (IPT) play an important role in musical performance. However, most of the existing works for IPT detection show low efficiency for variable-length audio and provide no assurance in the generalization as they rely on a single sound bank for training and testing. In this study, we propose an end-to-end Guzheng playing technique detection system using Fully Convolutional Networks that can be applied to variable-length audio. Because each Guzheng playing technique is applied to a note, a dedicated onset detector is trained to divide an audio into several notes and its predictions are fused with frame-wise IPT predictions. During fusion, we add the IPT predictions frame by frame inside each note and get the IPT with the highest probability within each note as the final output of that note. We create a new dataset named GZ_IsoTech from multiple sound banks and real-world recordings for Guzheng performance analysis. Our approach achieves 87.97% in frame-level accuracy and 80.76% in note-level F1-score, outperforming existing works by a large margin, which indicates the effectiveness of our proposed method in IPT detection.
### Source Data
#### Initial Data Collection and Normalization
Dichucheng Li, Monan Zhou
#### Who are the source language producers?
Students from FD-LAMT
### Annotations
#### Annotation process
This database contains 2824 audio clips of guzheng playing techniques. Among them, 2328 pieces were collected from virtual sound banks, and 496 pieces were played and recorded by a professional guzheng performer. These clips cover almost all the tones in the range of guzheng and the most commonly used playing techniques in guzheng performance. According to the different playing techniques of guzheng, the clips are divided into 8 categories: Vibrato(chanyin), Upward Portamento(shanghuayin), Downward Portamento(xiahuayin), Returning Portamento(huihuayin), Glissando (guazou, huazhi), Tremolo(yaozhi), Harmonic(fanyin), Plucks(gou,da,mo,tuo…).
#### Who are the annotators?
Students from FD-LAMT
### Personal and Sensitive Information
None
## Considerations for Using the Data
### Social Impact of Dataset
Promoting the development of music AI industry
### Discussion of Biases
Only for Traditional Chinese Instruments
### Other Known Limitations
Insufficient sample
## Additional Information
### Dataset Curators
Dichucheng Li
### Evaluation
[Li, Dichucheng, Yulun Wu, Qinyu Li, Jiahao Zhao, Yi Yu, Fan Xia and Wei Li. “Playing Technique Detection by Fusing Note Onset Information in Guzheng Performance.” International Society for Music Information Retrieval Conference (2022).](https://archives.ismir.net/ismir2022/paper/000037.pdf)
### Licensing Information
```
MIT License
Copyright (c) FD-LAMT
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.
```
### Citation Information
```
@dataset{zhaorui_liu_2021_5676893,
author = {Zhaorui Liu, Monan Zhou, Shenyang Xu, Wei Li, Zhaowen Wang and Zijin Li},
title = {CCMUSIC DATABASE: A Music Data Sharing Platform for Computational Musicology Research},
month = {nov},
year = {2021},
publisher = {Zenodo},
version = {1.1},
doi = {10.5281/zenodo.5676893},
url = {https://doi.org/10.5281/zenodo.5676893}
}
```
### Contributions
Promoting the development of music AI industry | 7,783 | [
[
-0.03436279296875,
-0.05108642578125,
0.00228118896484375,
0.046905517578125,
-0.0266265869140625,
-0.0129852294921875,
-0.034149169921875,
-0.03350830078125,
0.001895904541015625,
0.0260009765625,
-0.040557861328125,
-0.06683349609375,
-0.017303466796875,
-... |
sordonia/platy_icl0_maxD-1_maxC1000_2 | 2023-10-12T13:28:00.000Z | [
"region:us"
] | sordonia | null | null | 0 | 0 | 2023-10-12T13:27:42 | ---
configs:
- config_name: default
data_files:
- split: formal_logic
path: data/formal_logic-*
- split: machine_learning
path: data/machine_learning-*
- split: global_facts
path: data/global_facts-*
- split: abstract_algebra
path: data/abstract_algebra-*
- split: high_school_physics
path: data/high_school_physics-*
- split: college_biology
path: data/college_biology-*
- split: high_school_government_and_politics
path: data/high_school_government_and_politics-*
- split: prehistory
path: data/prehistory-*
- split: security_studies
path: data/security_studies-*
- split: sociology
path: data/sociology-*
dataset_info:
features:
- name: id
dtype: string
- name: context
dtype: string
- name: docno
dtype: string
- name: subject
dtype: string
- name: icl_examples
dtype: 'null'
- name: instruction
dtype: string
- name: author_instr
dtype: string
- name: response
dtype: string
- name: author_response
dtype: string
- name: normalized_cumul_logprob_response
dtype: float64
splits:
- name: formal_logic
num_bytes: 2112433.034561337
num_examples: 621
- name: machine_learning
num_bytes: 1857308.271933156
num_examples: 546
- name: global_facts
num_bytes: 1721241.731864793
num_examples: 506
- name: abstract_algebra
num_bytes: 1510338.5947588303
num_examples: 444
- name: high_school_physics
num_bytes: 1918538.2149639195
num_examples: 564
- name: college_biology
num_bytes: 2027391.4470186098
num_examples: 596
- name: high_school_government_and_politics
num_bytes: 1714438.4048613748
num_examples: 504
- name: prehistory
num_bytes: 1357263.7371819217
num_examples: 399
- name: security_studies
num_bytes: 1823291.6369160654
num_examples: 536
- name: sociology
num_bytes: 1870914.9259399923
num_examples: 550
download_size: 9799083
dataset_size: 17913160.0
---
# Dataset Card for "platy_icl0_maxD-1_maxC1000_2"
[More Information needed](https://github.com/huggingface/datasets/blob/main/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards) | 2,174 | [
[
-0.04656982421875,
-0.00946807861328125,
0.0032062530517578125,
0.035491943359375,
-0.0179290771484375,
-0.0020503997802734375,
0.0198516845703125,
0.006145477294921875,
0.042144775390625,
0.044403076171875,
-0.047637939453125,
-0.052520751953125,
-0.05078125,
... |
Vanwise/sddepends-1.6.0 | 2023-10-12T14:54:13.000Z | [
"region:us"
] | Vanwise | null | null | 0 | 0 | 2023-10-12T13:35:51 | Entry not found | 15 | [
[
-0.02142333984375,
-0.01495361328125,
0.05718994140625,
0.0288238525390625,
-0.035064697265625,
0.046539306640625,
0.052520751953125,
0.005062103271484375,
0.0513916015625,
0.016998291015625,
-0.052093505859375,
-0.014984130859375,
-0.060394287109375,
0.0379... |
meet-dagur4/Dog-Breed-Image-Data | 2023-10-12T13:44:17.000Z | [
"region:us"
] | meet-dagur4 | null | null | 0 | 0 | 2023-10-12T13:44:17 | Entry not found | 15 | [
[
-0.021392822265625,
-0.01494598388671875,
0.05718994140625,
0.028839111328125,
-0.0350341796875,
0.046539306640625,
0.052490234375,
0.00507354736328125,
0.051361083984375,
0.01702880859375,
-0.052093505859375,
-0.01494598388671875,
-0.06036376953125,
0.03790... |
SouravModak/name_of_your_dataset | 2023-10-12T14:00:27.000Z | [
"region:us"
] | SouravModak | null | null | 0 | 0 | 2023-10-12T14:00:22 | ---
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
dataset_info:
features:
- name: image
dtype: image
splits:
- name: train
num_bytes: 22631043.6
num_examples: 1300
download_size: 22594225
dataset_size: 22631043.6
---
# Dataset Card for "name_of_your_dataset"
[More Information needed](https://github.com/huggingface/datasets/blob/main/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards) | 455 | [
[
-0.0367431640625,
-0.0241851806640625,
0.00804901123046875,
0.01458740234375,
-0.0123443603515625,
0.011444091796875,
0.024566650390625,
0.004039764404296875,
0.08087158203125,
0.033111572265625,
-0.054656982421875,
-0.0543212890625,
-0.053131103515625,
-0.0... |
secret07/pic | 2023-10-12T16:44:13.000Z | [
"region:us"
] | secret07 | null | null | 0 | 0 | 2023-10-12T14:06:28 | Entry not found | 15 | [
[
-0.021392822265625,
-0.01494598388671875,
0.05718994140625,
0.028839111328125,
-0.0350341796875,
0.046539306640625,
0.052490234375,
0.00507354736328125,
0.051361083984375,
0.01702880859375,
-0.052093505859375,
-0.01494598388671875,
-0.06036376953125,
0.03790... |
weiiiii0622/HW1-Chinese_Extractive_QA | 2023-10-12T14:27:20.000Z | [
"region:us"
] | weiiiii0622 | null | null | 0 | 0 | 2023-10-12T14:17:05 | Entry not found | 15 | [
[
-0.021392822265625,
-0.01494598388671875,
0.05718994140625,
0.028839111328125,
-0.0350341796875,
0.046539306640625,
0.052490234375,
0.00507354736328125,
0.051361083984375,
0.01702880859375,
-0.052093505859375,
-0.01494598388671875,
-0.06036376953125,
0.03790... |
Wanfq/Explore_Instruct_Rewriting_10k | 2023-10-16T02:16:43.000Z | [
"language:en",
"license:cc-by-nc-4.0",
"arxiv:2310.09168",
"region:us"
] | Wanfq | null | null | 1 | 0 | 2023-10-12T14:22:20 | ---
license: cc-by-nc-4.0
language:
- en
---
<p align="center" width="100%">
</p>
<div id="top" align="center">
**Explore-Instruct: Enhancing Domain-Specific Instruction Coverage through Active Exploration**
<h4> |<a href="https://arxiv.org/abs/2310.09168"> 📑 Paper </a> |
<a href="https://huggingface.co/datasets?sort=trending&search=Explore_Instruct"> 🤗 Data </a> |
<a href="https://huggingface.co/models?sort=trending&search=Explore-LM"> 🤗 Model </a> |
<a href="https://github.com/fanqiwan/Explore-Instruct"> 🐱 Github Repo </a> |
</h4>
<!-- **Authors:** -->
_**Fanqi Wan<sup>†</sup>, Xinting Huang<sup>‡</sup>, Tao Yang<sup>†</sup>, Xiaojun Quan<sup>†</sup>, Wei Bi<sup>‡</sup>, Shuming Shi<sup>‡</sup>**_
<!-- **Affiliations:** -->
_<sup>†</sup> Sun Yat-sen University,
<sup>‡</sup> Tencent AI Lab_
</div>
## News
- **Oct 16, 2023:** 🔥 We're excited to announce that the Explore-Instruct datasets in brainstorming, rewriting, and math domains are now available on 🤗 [Huggingface Datasets](https://huggingface.co/datasets?sort=trending&search=Explore_Instruct)! Additionally, we've released Explore-LM models that have been initialized with LLaMA-7B and fine-tuned with the Explore-Instruct data in each domain. You can find these models on 🤗 [Huggingface Models](https://huggingface.co/models?sort=trending&search=Explore-LM). Happy exploring and instructing!
## Contents
- [Overview](#overview)
- [Data Release](#data-release)
- [Model Release](#model-release)
- [Data Generation Process](#data-generation-process)
- [Fine-tuning](#fine-tuning)
- [Evaluation](#evaluation)
- [Limitations](#limitations)
- [License](#license)
- [Citation](#citation)
- [Acknowledgements](#acknowledgments)
## Overview
We propose Explore-Instruct, a novel approach to enhancing domain-specific instruction coverage. We posit that the domain space is inherently structured akin to a tree, reminiscent of cognitive science ontologies. Drawing from the essence of classical search algorithms and incorporating the power of LLMs, Explore-Instruct is conceived to actively traverse the domain space and generate instruction-tuning data, **not** necessitating a predefined tree structure. Specifically, Explore-Instruct employs two strategic operations: lookahead and backtracking exploration:
- **Lookahead** delves into a multitude of potential fine-grained sub-tasks, thereby mapping out a complex network of tasks
- **Backtracking** seeks alternative branches to widen the search boundary, hence extending the domain spectrum.
<p align="center">
<img src="https://github.com/fanqiwan/Explore-Instruct/blob/main/assets/fig2.png?raw=true" width="95%"> <br>
</p>
## Data Release
We release the Explore-Instruct data in brainstorming, rewriting, and math domains on 🤗 [Huggingface Datasets](https://huggingface.co/datasets?sort=trending&search=Explore_Instruct). Each domain includes two versions of datasets: the basic and extended version. The base version contains 10k instruction-tuning data and the extended version contains 16k, 32k, and 64k instruction-tuning data for each domain respectively. Each dataset is a structured data file in the JSON format. It consists of a list of dictionaries, with each dictionary containing the following fields:
- `instruction`: `str`, describes the task the model should perform.
- `input`: `str`, optional context or input for the task.
- `output`: `str`, ground-truth output text for the task and input text.
The results of data-centric analysis are shown as follows:
<p align="left">
<img src="https://github.com/fanqiwan/Explore-Instruct/blob/main/assets/fig1.png?raw=true" width="50%"> <br>
</p>
| Method | Brainstorming Unique<br/>V-N pairs | Rewriting Unique<br/>V-N pairs | Math Unique<br/>V-N pairs |
|:--------------------------------|:----------------------------------:|:------------------------------:|:-------------------------:|
| _Domain-Specific Human-Curated_ | 2 | 8 | 3 |
| _Domain-Aware Self-Instruct_ | 781 | 1715 | 451 |
| Explore-Instruct | **790** | **2015** | **917** |
## Model Release
We release the Explore-LM models in brainstorming, rewriting, and math domains on 🤗 [Huggingface Models](https://huggingface.co/models?sort=trending&search=Explore-LM). Each domain includes two versions of models: the basic and extended version trained with the corresponding version of dataset.
The results of automatic and human evaluation in three domains are shown as follows:
- Automatic evaluation:
| Automatic Comparison in the Brainstorming Domain | Win:Tie:Lose | Beat Rate |
|:-------------------------------------------------|:------------:|:---------:|
| Explore-LM vs Domain-Curated-LM | 194:1:13 | 93.72 |
| Explore-LM-Ext vs Domain-Curated-LM | 196:1:11 | 94.69 |
| Explore-LM vs Domain-Instruct-LM | 114:56:38 | 75.00 |
| Explore-LM-Ext vs Domain-Instruct-LM | 122:55:31 | 79.74 |
| Explore-LM vs ChatGPT | 52:71:85 | 37.96 |
| Explore-LM-Ext vs ChatGPT | 83:69:56 | 59.71 |
| Automatic Comparison in the Rewriting Domain | Win:Tie:Lose | Beat Rate |
|:---------------------------------------------|:------------:|:---------:|
| Explore-LM vs Domain-Curated-LM | 50:38:6 | 89.29 |
| Explore-LM-Ext vs Domain-Curated-LM | 53:37:4 | 92.98 |
| Explore-LM vs Domain-Instruct-LM | 34:49:11 | 75.56 |
| Explore-LM-Ext vs Domain-Instruct-LM | 35:53:6 | 85.37 |
| Explore-LM vs ChatGPT | 11:59:24 | 31.43 |
| Explore-LM-Ext vs ChatGPT | 12:56:26 | 31.58 |
| Automatic Comparison in the Math Domain | Accuracy Rate |
|:----------------------------------------|:-------------:|
| Domain-Curated-LM | 3.4 |
| Domain-Instruct-LM | 4.0 |
| Explore-LM | 6.8 |
| Explore-LM-Ext | 8.4 |
| ChatGPT | 34.8 |
- Human evaluation:
<p align="left">
<img src="https://github.com/fanqiwan/Explore-Instruct/blob/main/assets/fig5.png?raw=true" width="95%"> <br>
</p>
## Data Generation Process
To generate the domain-specific instruction-tuning data, please follow the following commands step by step:
### Domain Space Exploration
```
python3 generate_instruction.py \
--action extend \
--save_dir ./en_data/demo_domain \ # input dir include current domain tree for exploration
--out_dir ./en_data/demo_domain_exploration \ # output dir of the explored new domain tree
--lang <LANGUAGE> \ # currently support 'en'
--domain demo_domain \ # domain for exploration
--extend_nums <TASK_NUMBER_DEPTH_0>,...,<TASK_NUMBER_DEPTH_MAX_DEPTH-1> \ # exploration breadth at each depth
--max_depth <MAX_DEPTH> \ # exploration depth
--assistant_name <ASSISTANT_NAME> # currently support openai and claude
```
### Instruction-Tuning Data Generation
```
python3 generate_instruction.py \
--action enrich \
--save_dir ./en_data/demo_domain_exploration \ # input dir include current domain tree for data generation
--out_dir ./en_data/demo_domain_generation \ # output dir of the domain tree with generated data
--lang <LANGUAGE> \ # currently support 'en'
--domain demo_domain \ # domain for exploration
--enrich_nums <DATA_NUMBER_DEPTH_0>,...,<DATA_NUMBER_DEPTH_MAX_DEPTH> \ # data number for task at each depth
--enrich_batch_size <BATCH_SIZE> \ # batch size for data generation
--assistant_name <ASSISTANT_NAME> # currently support openai and claude
```
### Task Pruning
```
python3 generate_instruction.py \
--action prune \
--save_dir ./en_data/demo_domain_generation \ # input dir include current domain tree for task pruning
--out_dir ./en_data/demo_domain_pruning \ # output dir of the domain tree with 'pruned_subtasks_name.json' file
--lang <LANGUAGE> \ # currently support 'en'
--domain demo_domain \ # domain for exploration
--pruned_file ./en_data/demo_domain_pruning/pruned_subtasks_name.json \ # file of pruned tasks
--prune_threshold <PRUNE_THRESHOLD> \ # threshold of rouge-l overlap between task names
--assistant_name <ASSISTANT_NAME> # currently support openai and claude
```
### Data Filtering
```
python3 generate_instruction.py \
--action filter \
--save_dir ./en_data/demo_domain_pruning \ # input dir include current domain tree for data filtering
--out_dir ./en_data/demo_domain_filtering \ # output dir of the domain tree with fitered data
--lang <LANGUAGE> \ # currently support 'en'
--domain demo_domain \ # domain for exploration
--pruned_file ./en_data/demo_domain_pruning/pruned_subtasks_name.json \ # file of pruned tasks
--filter_threshold <FILTER_THRESHOLD> \ # threshold of rouge-l overlap between instructions
--assistant_name <ASSISTANT_NAME> # currently support openai and claude
```
### Data Sampling
```
python3 generate_instruction.py \
--action sample \
--save_dir ./en_data/demo_domain_filtering \ # input dir include current domain tree for data sampling
--out_dir ./en_data/demo_domain_sampling \ # output dir of the domain tree with sampled data
--lang <LANGUAGE> \ # currently support 'en'
--domain demo_domain \ # domain for exploration
--pruned_file ./en_data/demo_domain_filtering/pruned_subtasks_name.json \ # file of pruned tasks
--sample_example_num <SAMPLE_EXAMPLES_NUM> \ # number of sampled examples
--sample_max_depth <SAMPLE_MAX_DEPTH> \ # max depth for data sampling
--sample_use_pruned \ # do not sample from pruned tasks
--assistant_name <ASSISTANT_NAME> # currently support openai and claude
```
## Fine-tuning
We fine-tune LLaMA-7B with the following hyperparameters:
| Hyperparameter | Global Batch Size | Learning rate | Epochs | Max length | Weight decay |
|:----------------|-------------------:|---------------:|--------:|------------:|--------------:|
| LLaMA 7B | 128 | 2e-5 | 3 | 512| 0 |
To reproduce the training procedure, please use the following command:
```
deepspeed --num_gpus=8 ./train/train.py \
--deepspeed ./deepspeed_config/deepspeed_zero3_offload_config.json \
--model_name_or_path decapoda-research/llama-7b-hf \
--data_path ./en_data/demo_domain_sampling \
--fp16 True \
--output_dir ./training_results/explore-lm-7b-demo-domain \
--num_train_epochs 3 \
--per_device_train_batch_size 2 \
--per_device_eval_batch_size 2 \
--gradient_accumulation_steps 8 \
--evaluation_strategy "no" \
--model_max_length 512 \
--save_strategy "steps" \
--save_steps 2000 \
--save_total_limit 1 \
--learning_rate 2e-5 \
--weight_decay 0. \
--warmup_ratio 0.03 \
--lr_scheduler_type "cosine" \
--logging_steps 1 \
--prompt_type alpaca \
2>&1 | tee ./training_logs/explore-lm-7b-demo-domain.log
python3 ./train/zero_to_fp32.py \
--checkpoint_dir ./training_results/explore-lm-7b-demo-domain \
--output_file ./training_results/explore-lm-7b-demo-domain/pytorch_model.bin
```
## Evaluation
The evaluation datasets for different domains are as follows:
- Brainstorming and Rewriting: From the corresponding categories in the translated test set of BELLE. ([en_eval_set.jsonl](./eval/question/en_eval_set.jsonl))
- Math: From randomly selected 500 questions from the test set of MATH. ([MATH_eval_set_sample.jsonl](./eval/question/MATH_eval_set_sample.jsonl))
The evaluation metrics for different domains are as follows:
- Brainstorming and Rewriting: Both automatic and human evaluations following Vicuna.
- Math: Accuracy Rate metric in solving math problems.
The automatic evaluation commands for different domains are as follows:
```
# Brainstorming and Rewriting Domain
# 1. Inference
python3 ./eval/generate.py \
--model_id <MODEL_ID> \
--model_path <MODEL_PATH> \
--question_file ./eval/question/en_eval_set.jsonl \
--answer_file ./eval/answer/<MODEL_ID>.jsonl \
--num_gpus 8 \
--num_beams 1 \
--temperature 0.7 \
--max_new_tokens 512 \
--prompt_type alpaca \
--do_sample
# 2. Evaluation
python3 ./eval/chatgpt_score.py \
--baseline_file ./eval/answer/<MODEL_1>.jsonl \ # answer of baseline model to compare with
--answer_file ./eval/answer/<MODEL_2>.jsonl \ # answer of evaluation model
--review_file ./eval/review/<MODEL_1>_cp_<MODEL_2>_<DOMAIN>.jsonl \ # review from chatgpt
--prompt_file ./eval/prompt/en_review_prompt_compare.jsonl \ # evaluation prompt for chatgpt
--target_classes <DOMAIN> \ # evaluation domain
--batch_size <BATCH_SIZE> \
--review_model "gpt-3.5-turbo-0301"
```
```
# Math Domain
# 1. Inference
python3 ./eval/generate.py \
--model_id <MODEL_ID> \
--model_path <MODEL_PATH> \
--question_file ./eval/question/MATH_eval_set_sample.jsonl \
--answer_file ./eval/answer/<MODEL_ID>.jsonl \
--num_gpus 8 \
--num_beams 10 \
--temperature 1.0 \
--max_new_tokens 512 \
--prompt_type alpaca
# 2. Evaluation
python3 ./eval/auto_eval.py \
--question_file ./eval/question/MATH_eval_set_sample.jsonl \
--answer_file ./eval/answer/<MODEL_ID>.jsonl # answer of evaluation model
```
## Limitations
Explore-Instruct is still under development and needs a lot of improvements. We acknowledge that our work focuses on the enhancement of domain-specific instruction coverage and does not address other aspects of instruction-tuning, such as the generation of complex and challenging instructions or the mitigation of toxic and harmful instructions. Future work is needed to explore the potential of our approach in these areas.
## License
Explore-Instruct is intended and licensed for research use only. The dataset is CC BY NC 4.0 (allowing only non-commercial use) and models trained using the dataset should not be used outside of research purposes. The weights of Explore-LM models are also CC BY NC 4.0 (allowing only non-commercial use).
## Citation
If you find this work is relevant with your research or applications, please feel free to cite our work!
```
@misc{wan2023explore,
title={Explore-Instruct: Enhancing Domain-Specific Instruction Coverage through Active Exploration},
author={Fanqi, Wan and Xinting, Huang and Tao, Yang and Xiaojun, Quan and Wei, Bi and Shuming, Shi},
year={2023},
eprint={2310.09168},
archivePrefix={arXiv},
primaryClass={cs.CL}
}
```
## Acknowledgments
This repo benefits from [Stanford-Alpaca](https://github.com/tatsu-lab/stanford_alpaca) and [Vicuna](https://github.com/lm-sys/FastChat). Thanks for their wonderful works!
| 15,179 | [
[
-0.042694091796875,
-0.0736083984375,
0.025787353515625,
0.01262664794921875,
0.0161285400390625,
0.006938934326171875,
-0.0211639404296875,
-0.02081298828125,
0.00553131103515625,
0.0247955322265625,
-0.0706787109375,
-0.059661865234375,
-0.0380859375,
-0.0... |
Wanfq/Explore_Instruct_Rewriting_32k | 2023-10-16T02:17:26.000Z | [
"language:en",
"license:cc-by-nc-4.0",
"arxiv:2310.09168",
"region:us"
] | Wanfq | null | null | 3 | 0 | 2023-10-12T14:23:24 | ---
license: cc-by-nc-4.0
language:
- en
---
<p align="center" width="100%">
</p>
<div id="top" align="center">
**Explore-Instruct: Enhancing Domain-Specific Instruction Coverage through Active Exploration**
<h4> |<a href="https://arxiv.org/abs/2310.09168"> 📑 Paper </a> |
<a href="https://huggingface.co/datasets?sort=trending&search=Explore_Instruct"> 🤗 Data </a> |
<a href="https://huggingface.co/models?sort=trending&search=Explore-LM"> 🤗 Model </a> |
<a href="https://github.com/fanqiwan/Explore-Instruct"> 🐱 Github Repo </a> |
</h4>
<!-- **Authors:** -->
_**Fanqi Wan<sup>†</sup>, Xinting Huang<sup>‡</sup>, Tao Yang<sup>†</sup>, Xiaojun Quan<sup>†</sup>, Wei Bi<sup>‡</sup>, Shuming Shi<sup>‡</sup>**_
<!-- **Affiliations:** -->
_<sup>†</sup> Sun Yat-sen University,
<sup>‡</sup> Tencent AI Lab_
</div>
## News
- **Oct 16, 2023:** 🔥 We're excited to announce that the Explore-Instruct datasets in brainstorming, rewriting, and math domains are now available on 🤗 [Huggingface Datasets](https://huggingface.co/datasets?sort=trending&search=Explore_Instruct)! Additionally, we've released Explore-LM models that have been initialized with LLaMA-7B and fine-tuned with the Explore-Instruct data in each domain. You can find these models on 🤗 [Huggingface Models](https://huggingface.co/models?sort=trending&search=Explore-LM). Happy exploring and instructing!
## Contents
- [Overview](#overview)
- [Data Release](#data-release)
- [Model Release](#model-release)
- [Data Generation Process](#data-generation-process)
- [Fine-tuning](#fine-tuning)
- [Evaluation](#evaluation)
- [Limitations](#limitations)
- [License](#license)
- [Citation](#citation)
- [Acknowledgements](#acknowledgments)
## Overview
We propose Explore-Instruct, a novel approach to enhancing domain-specific instruction coverage. We posit that the domain space is inherently structured akin to a tree, reminiscent of cognitive science ontologies. Drawing from the essence of classical search algorithms and incorporating the power of LLMs, Explore-Instruct is conceived to actively traverse the domain space and generate instruction-tuning data, **not** necessitating a predefined tree structure. Specifically, Explore-Instruct employs two strategic operations: lookahead and backtracking exploration:
- **Lookahead** delves into a multitude of potential fine-grained sub-tasks, thereby mapping out a complex network of tasks
- **Backtracking** seeks alternative branches to widen the search boundary, hence extending the domain spectrum.
<p align="center">
<img src="https://github.com/fanqiwan/Explore-Instruct/blob/main/assets/fig2.png?raw=true" width="95%"> <br>
</p>
## Data Release
We release the Explore-Instruct data in brainstorming, rewriting, and math domains on 🤗 [Huggingface Datasets](https://huggingface.co/datasets?sort=trending&search=Explore_Instruct). Each domain includes two versions of datasets: the basic and extended version. The base version contains 10k instruction-tuning data and the extended version contains 16k, 32k, and 64k instruction-tuning data for each domain respectively. Each dataset is a structured data file in the JSON format. It consists of a list of dictionaries, with each dictionary containing the following fields:
- `instruction`: `str`, describes the task the model should perform.
- `input`: `str`, optional context or input for the task.
- `output`: `str`, ground-truth output text for the task and input text.
The results of data-centric analysis are shown as follows:
<p align="left">
<img src="https://github.com/fanqiwan/Explore-Instruct/blob/main/assets/fig1.png?raw=true" width="50%"> <br>
</p>
| Method | Brainstorming Unique<br/>V-N pairs | Rewriting Unique<br/>V-N pairs | Math Unique<br/>V-N pairs |
|:--------------------------------|:----------------------------------:|:------------------------------:|:-------------------------:|
| _Domain-Specific Human-Curated_ | 2 | 8 | 3 |
| _Domain-Aware Self-Instruct_ | 781 | 1715 | 451 |
| Explore-Instruct | **790** | **2015** | **917** |
## Model Release
We release the Explore-LM models in brainstorming, rewriting, and math domains on 🤗 [Huggingface Models](https://huggingface.co/models?sort=trending&search=Explore-LM). Each domain includes two versions of models: the basic and extended version trained with the corresponding version of dataset.
The results of automatic and human evaluation in three domains are shown as follows:
- Automatic evaluation:
| Automatic Comparison in the Brainstorming Domain | Win:Tie:Lose | Beat Rate |
|:-------------------------------------------------|:------------:|:---------:|
| Explore-LM vs Domain-Curated-LM | 194:1:13 | 93.72 |
| Explore-LM-Ext vs Domain-Curated-LM | 196:1:11 | 94.69 |
| Explore-LM vs Domain-Instruct-LM | 114:56:38 | 75.00 |
| Explore-LM-Ext vs Domain-Instruct-LM | 122:55:31 | 79.74 |
| Explore-LM vs ChatGPT | 52:71:85 | 37.96 |
| Explore-LM-Ext vs ChatGPT | 83:69:56 | 59.71 |
| Automatic Comparison in the Rewriting Domain | Win:Tie:Lose | Beat Rate |
|:---------------------------------------------|:------------:|:---------:|
| Explore-LM vs Domain-Curated-LM | 50:38:6 | 89.29 |
| Explore-LM-Ext vs Domain-Curated-LM | 53:37:4 | 92.98 |
| Explore-LM vs Domain-Instruct-LM | 34:49:11 | 75.56 |
| Explore-LM-Ext vs Domain-Instruct-LM | 35:53:6 | 85.37 |
| Explore-LM vs ChatGPT | 11:59:24 | 31.43 |
| Explore-LM-Ext vs ChatGPT | 12:56:26 | 31.58 |
| Automatic Comparison in the Math Domain | Accuracy Rate |
|:----------------------------------------|:-------------:|
| Domain-Curated-LM | 3.4 |
| Domain-Instruct-LM | 4.0 |
| Explore-LM | 6.8 |
| Explore-LM-Ext | 8.4 |
| ChatGPT | 34.8 |
- Human evaluation:
<p align="left">
<img src="https://github.com/fanqiwan/Explore-Instruct/blob/main/assets/fig5.png?raw=true" width="95%"> <br>
</p>
## Data Generation Process
To generate the domain-specific instruction-tuning data, please follow the following commands step by step:
### Domain Space Exploration
```
python3 generate_instruction.py \
--action extend \
--save_dir ./en_data/demo_domain \ # input dir include current domain tree for exploration
--out_dir ./en_data/demo_domain_exploration \ # output dir of the explored new domain tree
--lang <LANGUAGE> \ # currently support 'en'
--domain demo_domain \ # domain for exploration
--extend_nums <TASK_NUMBER_DEPTH_0>,...,<TASK_NUMBER_DEPTH_MAX_DEPTH-1> \ # exploration breadth at each depth
--max_depth <MAX_DEPTH> \ # exploration depth
--assistant_name <ASSISTANT_NAME> # currently support openai and claude
```
### Instruction-Tuning Data Generation
```
python3 generate_instruction.py \
--action enrich \
--save_dir ./en_data/demo_domain_exploration \ # input dir include current domain tree for data generation
--out_dir ./en_data/demo_domain_generation \ # output dir of the domain tree with generated data
--lang <LANGUAGE> \ # currently support 'en'
--domain demo_domain \ # domain for exploration
--enrich_nums <DATA_NUMBER_DEPTH_0>,...,<DATA_NUMBER_DEPTH_MAX_DEPTH> \ # data number for task at each depth
--enrich_batch_size <BATCH_SIZE> \ # batch size for data generation
--assistant_name <ASSISTANT_NAME> # currently support openai and claude
```
### Task Pruning
```
python3 generate_instruction.py \
--action prune \
--save_dir ./en_data/demo_domain_generation \ # input dir include current domain tree for task pruning
--out_dir ./en_data/demo_domain_pruning \ # output dir of the domain tree with 'pruned_subtasks_name.json' file
--lang <LANGUAGE> \ # currently support 'en'
--domain demo_domain \ # domain for exploration
--pruned_file ./en_data/demo_domain_pruning/pruned_subtasks_name.json \ # file of pruned tasks
--prune_threshold <PRUNE_THRESHOLD> \ # threshold of rouge-l overlap between task names
--assistant_name <ASSISTANT_NAME> # currently support openai and claude
```
### Data Filtering
```
python3 generate_instruction.py \
--action filter \
--save_dir ./en_data/demo_domain_pruning \ # input dir include current domain tree for data filtering
--out_dir ./en_data/demo_domain_filtering \ # output dir of the domain tree with fitered data
--lang <LANGUAGE> \ # currently support 'en'
--domain demo_domain \ # domain for exploration
--pruned_file ./en_data/demo_domain_pruning/pruned_subtasks_name.json \ # file of pruned tasks
--filter_threshold <FILTER_THRESHOLD> \ # threshold of rouge-l overlap between instructions
--assistant_name <ASSISTANT_NAME> # currently support openai and claude
```
### Data Sampling
```
python3 generate_instruction.py \
--action sample \
--save_dir ./en_data/demo_domain_filtering \ # input dir include current domain tree for data sampling
--out_dir ./en_data/demo_domain_sampling \ # output dir of the domain tree with sampled data
--lang <LANGUAGE> \ # currently support 'en'
--domain demo_domain \ # domain for exploration
--pruned_file ./en_data/demo_domain_filtering/pruned_subtasks_name.json \ # file of pruned tasks
--sample_example_num <SAMPLE_EXAMPLES_NUM> \ # number of sampled examples
--sample_max_depth <SAMPLE_MAX_DEPTH> \ # max depth for data sampling
--sample_use_pruned \ # do not sample from pruned tasks
--assistant_name <ASSISTANT_NAME> # currently support openai and claude
```
## Fine-tuning
We fine-tune LLaMA-7B with the following hyperparameters:
| Hyperparameter | Global Batch Size | Learning rate | Epochs | Max length | Weight decay |
|:----------------|-------------------:|---------------:|--------:|------------:|--------------:|
| LLaMA 7B | 128 | 2e-5 | 3 | 512| 0 |
To reproduce the training procedure, please use the following command:
```
deepspeed --num_gpus=8 ./train/train.py \
--deepspeed ./deepspeed_config/deepspeed_zero3_offload_config.json \
--model_name_or_path decapoda-research/llama-7b-hf \
--data_path ./en_data/demo_domain_sampling \
--fp16 True \
--output_dir ./training_results/explore-lm-7b-demo-domain \
--num_train_epochs 3 \
--per_device_train_batch_size 2 \
--per_device_eval_batch_size 2 \
--gradient_accumulation_steps 8 \
--evaluation_strategy "no" \
--model_max_length 512 \
--save_strategy "steps" \
--save_steps 2000 \
--save_total_limit 1 \
--learning_rate 2e-5 \
--weight_decay 0. \
--warmup_ratio 0.03 \
--lr_scheduler_type "cosine" \
--logging_steps 1 \
--prompt_type alpaca \
2>&1 | tee ./training_logs/explore-lm-7b-demo-domain.log
python3 ./train/zero_to_fp32.py \
--checkpoint_dir ./training_results/explore-lm-7b-demo-domain \
--output_file ./training_results/explore-lm-7b-demo-domain/pytorch_model.bin
```
## Evaluation
The evaluation datasets for different domains are as follows:
- Brainstorming and Rewriting: From the corresponding categories in the translated test set of BELLE. ([en_eval_set.jsonl](./eval/question/en_eval_set.jsonl))
- Math: From randomly selected 500 questions from the test set of MATH. ([MATH_eval_set_sample.jsonl](./eval/question/MATH_eval_set_sample.jsonl))
The evaluation metrics for different domains are as follows:
- Brainstorming and Rewriting: Both automatic and human evaluations following Vicuna.
- Math: Accuracy Rate metric in solving math problems.
The automatic evaluation commands for different domains are as follows:
```
# Brainstorming and Rewriting Domain
# 1. Inference
python3 ./eval/generate.py \
--model_id <MODEL_ID> \
--model_path <MODEL_PATH> \
--question_file ./eval/question/en_eval_set.jsonl \
--answer_file ./eval/answer/<MODEL_ID>.jsonl \
--num_gpus 8 \
--num_beams 1 \
--temperature 0.7 \
--max_new_tokens 512 \
--prompt_type alpaca \
--do_sample
# 2. Evaluation
python3 ./eval/chatgpt_score.py \
--baseline_file ./eval/answer/<MODEL_1>.jsonl \ # answer of baseline model to compare with
--answer_file ./eval/answer/<MODEL_2>.jsonl \ # answer of evaluation model
--review_file ./eval/review/<MODEL_1>_cp_<MODEL_2>_<DOMAIN>.jsonl \ # review from chatgpt
--prompt_file ./eval/prompt/en_review_prompt_compare.jsonl \ # evaluation prompt for chatgpt
--target_classes <DOMAIN> \ # evaluation domain
--batch_size <BATCH_SIZE> \
--review_model "gpt-3.5-turbo-0301"
```
```
# Math Domain
# 1. Inference
python3 ./eval/generate.py \
--model_id <MODEL_ID> \
--model_path <MODEL_PATH> \
--question_file ./eval/question/MATH_eval_set_sample.jsonl \
--answer_file ./eval/answer/<MODEL_ID>.jsonl \
--num_gpus 8 \
--num_beams 10 \
--temperature 1.0 \
--max_new_tokens 512 \
--prompt_type alpaca
# 2. Evaluation
python3 ./eval/auto_eval.py \
--question_file ./eval/question/MATH_eval_set_sample.jsonl \
--answer_file ./eval/answer/<MODEL_ID>.jsonl # answer of evaluation model
```
## Limitations
Explore-Instruct is still under development and needs a lot of improvements. We acknowledge that our work focuses on the enhancement of domain-specific instruction coverage and does not address other aspects of instruction-tuning, such as the generation of complex and challenging instructions or the mitigation of toxic and harmful instructions. Future work is needed to explore the potential of our approach in these areas.
## License
Explore-Instruct is intended and licensed for research use only. The dataset is CC BY NC 4.0 (allowing only non-commercial use) and models trained using the dataset should not be used outside of research purposes. The weights of Explore-LM models are also CC BY NC 4.0 (allowing only non-commercial use).
## Citation
If you find this work is relevant with your research or applications, please feel free to cite our work!
```
@misc{wan2023explore,
title={Explore-Instruct: Enhancing Domain-Specific Instruction Coverage through Active Exploration},
author={Fanqi, Wan and Xinting, Huang and Tao, Yang and Xiaojun, Quan and Wei, Bi and Shuming, Shi},
year={2023},
eprint={2310.09168},
archivePrefix={arXiv},
primaryClass={cs.CL}
}
```
## Acknowledgments
This repo benefits from [Stanford-Alpaca](https://github.com/tatsu-lab/stanford_alpaca) and [Vicuna](https://github.com/lm-sys/FastChat). Thanks for their wonderful works!
| 15,179 | [
[
-0.042694091796875,
-0.0736083984375,
0.025787353515625,
0.01262664794921875,
0.0161285400390625,
0.006938934326171875,
-0.0211639404296875,
-0.02081298828125,
0.00553131103515625,
0.0247955322265625,
-0.0706787109375,
-0.059661865234375,
-0.0380859375,
-0.0... |
Wanfq/Explore_Instruct_Brainstorming_10k | 2023-10-16T02:18:01.000Z | [
"language:en",
"license:cc-by-nc-4.0",
"arxiv:2310.09168",
"region:us"
] | Wanfq | null | null | 0 | 0 | 2023-10-12T14:27:21 | ---
license: cc-by-nc-4.0
language:
- en
---
<p align="center" width="100%">
</p>
<div id="top" align="center">
**Explore-Instruct: Enhancing Domain-Specific Instruction Coverage through Active Exploration**
<h4> |<a href="https://arxiv.org/abs/2310.09168"> 📑 Paper </a> |
<a href="https://huggingface.co/datasets?sort=trending&search=Explore_Instruct"> 🤗 Data </a> |
<a href="https://huggingface.co/models?sort=trending&search=Explore-LM"> 🤗 Model </a> |
<a href="https://github.com/fanqiwan/Explore-Instruct"> 🐱 Github Repo </a> |
</h4>
<!-- **Authors:** -->
_**Fanqi Wan<sup>†</sup>, Xinting Huang<sup>‡</sup>, Tao Yang<sup>†</sup>, Xiaojun Quan<sup>†</sup>, Wei Bi<sup>‡</sup>, Shuming Shi<sup>‡</sup>**_
<!-- **Affiliations:** -->
_<sup>†</sup> Sun Yat-sen University,
<sup>‡</sup> Tencent AI Lab_
</div>
## News
- **Oct 16, 2023:** 🔥 We're excited to announce that the Explore-Instruct datasets in brainstorming, rewriting, and math domains are now available on 🤗 [Huggingface Datasets](https://huggingface.co/datasets?sort=trending&search=Explore_Instruct)! Additionally, we've released Explore-LM models that have been initialized with LLaMA-7B and fine-tuned with the Explore-Instruct data in each domain. You can find these models on 🤗 [Huggingface Models](https://huggingface.co/models?sort=trending&search=Explore-LM). Happy exploring and instructing!
## Contents
- [Overview](#overview)
- [Data Release](#data-release)
- [Model Release](#model-release)
- [Data Generation Process](#data-generation-process)
- [Fine-tuning](#fine-tuning)
- [Evaluation](#evaluation)
- [Limitations](#limitations)
- [License](#license)
- [Citation](#citation)
- [Acknowledgements](#acknowledgments)
## Overview
We propose Explore-Instruct, a novel approach to enhancing domain-specific instruction coverage. We posit that the domain space is inherently structured akin to a tree, reminiscent of cognitive science ontologies. Drawing from the essence of classical search algorithms and incorporating the power of LLMs, Explore-Instruct is conceived to actively traverse the domain space and generate instruction-tuning data, **not** necessitating a predefined tree structure. Specifically, Explore-Instruct employs two strategic operations: lookahead and backtracking exploration:
- **Lookahead** delves into a multitude of potential fine-grained sub-tasks, thereby mapping out a complex network of tasks
- **Backtracking** seeks alternative branches to widen the search boundary, hence extending the domain spectrum.
<p align="center">
<img src="https://github.com/fanqiwan/Explore-Instruct/blob/main/assets/fig2.png?raw=true" width="95%"> <br>
</p>
## Data Release
We release the Explore-Instruct data in brainstorming, rewriting, and math domains on 🤗 [Huggingface Datasets](https://huggingface.co/datasets?sort=trending&search=Explore_Instruct). Each domain includes two versions of datasets: the basic and extended version. The base version contains 10k instruction-tuning data and the extended version contains 16k, 32k, and 64k instruction-tuning data for each domain respectively. Each dataset is a structured data file in the JSON format. It consists of a list of dictionaries, with each dictionary containing the following fields:
- `instruction`: `str`, describes the task the model should perform.
- `input`: `str`, optional context or input for the task.
- `output`: `str`, ground-truth output text for the task and input text.
The results of data-centric analysis are shown as follows:
<p align="left">
<img src="https://github.com/fanqiwan/Explore-Instruct/blob/main/assets/fig1.png?raw=true" width="50%"> <br>
</p>
| Method | Brainstorming Unique<br/>V-N pairs | Rewriting Unique<br/>V-N pairs | Math Unique<br/>V-N pairs |
|:--------------------------------|:----------------------------------:|:------------------------------:|:-------------------------:|
| _Domain-Specific Human-Curated_ | 2 | 8 | 3 |
| _Domain-Aware Self-Instruct_ | 781 | 1715 | 451 |
| Explore-Instruct | **790** | **2015** | **917** |
## Model Release
We release the Explore-LM models in brainstorming, rewriting, and math domains on 🤗 [Huggingface Models](https://huggingface.co/models?sort=trending&search=Explore-LM). Each domain includes two versions of models: the basic and extended version trained with the corresponding version of dataset.
The results of automatic and human evaluation in three domains are shown as follows:
- Automatic evaluation:
| Automatic Comparison in the Brainstorming Domain | Win:Tie:Lose | Beat Rate |
|:-------------------------------------------------|:------------:|:---------:|
| Explore-LM vs Domain-Curated-LM | 194:1:13 | 93.72 |
| Explore-LM-Ext vs Domain-Curated-LM | 196:1:11 | 94.69 |
| Explore-LM vs Domain-Instruct-LM | 114:56:38 | 75.00 |
| Explore-LM-Ext vs Domain-Instruct-LM | 122:55:31 | 79.74 |
| Explore-LM vs ChatGPT | 52:71:85 | 37.96 |
| Explore-LM-Ext vs ChatGPT | 83:69:56 | 59.71 |
| Automatic Comparison in the Rewriting Domain | Win:Tie:Lose | Beat Rate |
|:---------------------------------------------|:------------:|:---------:|
| Explore-LM vs Domain-Curated-LM | 50:38:6 | 89.29 |
| Explore-LM-Ext vs Domain-Curated-LM | 53:37:4 | 92.98 |
| Explore-LM vs Domain-Instruct-LM | 34:49:11 | 75.56 |
| Explore-LM-Ext vs Domain-Instruct-LM | 35:53:6 | 85.37 |
| Explore-LM vs ChatGPT | 11:59:24 | 31.43 |
| Explore-LM-Ext vs ChatGPT | 12:56:26 | 31.58 |
| Automatic Comparison in the Math Domain | Accuracy Rate |
|:----------------------------------------|:-------------:|
| Domain-Curated-LM | 3.4 |
| Domain-Instruct-LM | 4.0 |
| Explore-LM | 6.8 |
| Explore-LM-Ext | 8.4 |
| ChatGPT | 34.8 |
- Human evaluation:
<p align="left">
<img src="https://github.com/fanqiwan/Explore-Instruct/blob/main/assets/fig5.png?raw=true" width="95%"> <br>
</p>
## Data Generation Process
To generate the domain-specific instruction-tuning data, please follow the following commands step by step:
### Domain Space Exploration
```
python3 generate_instruction.py \
--action extend \
--save_dir ./en_data/demo_domain \ # input dir include current domain tree for exploration
--out_dir ./en_data/demo_domain_exploration \ # output dir of the explored new domain tree
--lang <LANGUAGE> \ # currently support 'en'
--domain demo_domain \ # domain for exploration
--extend_nums <TASK_NUMBER_DEPTH_0>,...,<TASK_NUMBER_DEPTH_MAX_DEPTH-1> \ # exploration breadth at each depth
--max_depth <MAX_DEPTH> \ # exploration depth
--assistant_name <ASSISTANT_NAME> # currently support openai and claude
```
### Instruction-Tuning Data Generation
```
python3 generate_instruction.py \
--action enrich \
--save_dir ./en_data/demo_domain_exploration \ # input dir include current domain tree for data generation
--out_dir ./en_data/demo_domain_generation \ # output dir of the domain tree with generated data
--lang <LANGUAGE> \ # currently support 'en'
--domain demo_domain \ # domain for exploration
--enrich_nums <DATA_NUMBER_DEPTH_0>,...,<DATA_NUMBER_DEPTH_MAX_DEPTH> \ # data number for task at each depth
--enrich_batch_size <BATCH_SIZE> \ # batch size for data generation
--assistant_name <ASSISTANT_NAME> # currently support openai and claude
```
### Task Pruning
```
python3 generate_instruction.py \
--action prune \
--save_dir ./en_data/demo_domain_generation \ # input dir include current domain tree for task pruning
--out_dir ./en_data/demo_domain_pruning \ # output dir of the domain tree with 'pruned_subtasks_name.json' file
--lang <LANGUAGE> \ # currently support 'en'
--domain demo_domain \ # domain for exploration
--pruned_file ./en_data/demo_domain_pruning/pruned_subtasks_name.json \ # file of pruned tasks
--prune_threshold <PRUNE_THRESHOLD> \ # threshold of rouge-l overlap between task names
--assistant_name <ASSISTANT_NAME> # currently support openai and claude
```
### Data Filtering
```
python3 generate_instruction.py \
--action filter \
--save_dir ./en_data/demo_domain_pruning \ # input dir include current domain tree for data filtering
--out_dir ./en_data/demo_domain_filtering \ # output dir of the domain tree with fitered data
--lang <LANGUAGE> \ # currently support 'en'
--domain demo_domain \ # domain for exploration
--pruned_file ./en_data/demo_domain_pruning/pruned_subtasks_name.json \ # file of pruned tasks
--filter_threshold <FILTER_THRESHOLD> \ # threshold of rouge-l overlap between instructions
--assistant_name <ASSISTANT_NAME> # currently support openai and claude
```
### Data Sampling
```
python3 generate_instruction.py \
--action sample \
--save_dir ./en_data/demo_domain_filtering \ # input dir include current domain tree for data sampling
--out_dir ./en_data/demo_domain_sampling \ # output dir of the domain tree with sampled data
--lang <LANGUAGE> \ # currently support 'en'
--domain demo_domain \ # domain for exploration
--pruned_file ./en_data/demo_domain_filtering/pruned_subtasks_name.json \ # file of pruned tasks
--sample_example_num <SAMPLE_EXAMPLES_NUM> \ # number of sampled examples
--sample_max_depth <SAMPLE_MAX_DEPTH> \ # max depth for data sampling
--sample_use_pruned \ # do not sample from pruned tasks
--assistant_name <ASSISTANT_NAME> # currently support openai and claude
```
## Fine-tuning
We fine-tune LLaMA-7B with the following hyperparameters:
| Hyperparameter | Global Batch Size | Learning rate | Epochs | Max length | Weight decay |
|:----------------|-------------------:|---------------:|--------:|------------:|--------------:|
| LLaMA 7B | 128 | 2e-5 | 3 | 512| 0 |
To reproduce the training procedure, please use the following command:
```
deepspeed --num_gpus=8 ./train/train.py \
--deepspeed ./deepspeed_config/deepspeed_zero3_offload_config.json \
--model_name_or_path decapoda-research/llama-7b-hf \
--data_path ./en_data/demo_domain_sampling \
--fp16 True \
--output_dir ./training_results/explore-lm-7b-demo-domain \
--num_train_epochs 3 \
--per_device_train_batch_size 2 \
--per_device_eval_batch_size 2 \
--gradient_accumulation_steps 8 \
--evaluation_strategy "no" \
--model_max_length 512 \
--save_strategy "steps" \
--save_steps 2000 \
--save_total_limit 1 \
--learning_rate 2e-5 \
--weight_decay 0. \
--warmup_ratio 0.03 \
--lr_scheduler_type "cosine" \
--logging_steps 1 \
--prompt_type alpaca \
2>&1 | tee ./training_logs/explore-lm-7b-demo-domain.log
python3 ./train/zero_to_fp32.py \
--checkpoint_dir ./training_results/explore-lm-7b-demo-domain \
--output_file ./training_results/explore-lm-7b-demo-domain/pytorch_model.bin
```
## Evaluation
The evaluation datasets for different domains are as follows:
- Brainstorming and Rewriting: From the corresponding categories in the translated test set of BELLE. ([en_eval_set.jsonl](./eval/question/en_eval_set.jsonl))
- Math: From randomly selected 500 questions from the test set of MATH. ([MATH_eval_set_sample.jsonl](./eval/question/MATH_eval_set_sample.jsonl))
The evaluation metrics for different domains are as follows:
- Brainstorming and Rewriting: Both automatic and human evaluations following Vicuna.
- Math: Accuracy Rate metric in solving math problems.
The automatic evaluation commands for different domains are as follows:
```
# Brainstorming and Rewriting Domain
# 1. Inference
python3 ./eval/generate.py \
--model_id <MODEL_ID> \
--model_path <MODEL_PATH> \
--question_file ./eval/question/en_eval_set.jsonl \
--answer_file ./eval/answer/<MODEL_ID>.jsonl \
--num_gpus 8 \
--num_beams 1 \
--temperature 0.7 \
--max_new_tokens 512 \
--prompt_type alpaca \
--do_sample
# 2. Evaluation
python3 ./eval/chatgpt_score.py \
--baseline_file ./eval/answer/<MODEL_1>.jsonl \ # answer of baseline model to compare with
--answer_file ./eval/answer/<MODEL_2>.jsonl \ # answer of evaluation model
--review_file ./eval/review/<MODEL_1>_cp_<MODEL_2>_<DOMAIN>.jsonl \ # review from chatgpt
--prompt_file ./eval/prompt/en_review_prompt_compare.jsonl \ # evaluation prompt for chatgpt
--target_classes <DOMAIN> \ # evaluation domain
--batch_size <BATCH_SIZE> \
--review_model "gpt-3.5-turbo-0301"
```
```
# Math Domain
# 1. Inference
python3 ./eval/generate.py \
--model_id <MODEL_ID> \
--model_path <MODEL_PATH> \
--question_file ./eval/question/MATH_eval_set_sample.jsonl \
--answer_file ./eval/answer/<MODEL_ID>.jsonl \
--num_gpus 8 \
--num_beams 10 \
--temperature 1.0 \
--max_new_tokens 512 \
--prompt_type alpaca
# 2. Evaluation
python3 ./eval/auto_eval.py \
--question_file ./eval/question/MATH_eval_set_sample.jsonl \
--answer_file ./eval/answer/<MODEL_ID>.jsonl # answer of evaluation model
```
## Limitations
Explore-Instruct is still under development and needs a lot of improvements. We acknowledge that our work focuses on the enhancement of domain-specific instruction coverage and does not address other aspects of instruction-tuning, such as the generation of complex and challenging instructions or the mitigation of toxic and harmful instructions. Future work is needed to explore the potential of our approach in these areas.
## License
Explore-Instruct is intended and licensed for research use only. The dataset is CC BY NC 4.0 (allowing only non-commercial use) and models trained using the dataset should not be used outside of research purposes. The weights of Explore-LM models are also CC BY NC 4.0 (allowing only non-commercial use).
## Citation
If you find this work is relevant with your research or applications, please feel free to cite our work!
```
@misc{wan2023explore,
title={Explore-Instruct: Enhancing Domain-Specific Instruction Coverage through Active Exploration},
author={Fanqi, Wan and Xinting, Huang and Tao, Yang and Xiaojun, Quan and Wei, Bi and Shuming, Shi},
year={2023},
eprint={2310.09168},
archivePrefix={arXiv},
primaryClass={cs.CL}
}
```
## Acknowledgments
This repo benefits from [Stanford-Alpaca](https://github.com/tatsu-lab/stanford_alpaca) and [Vicuna](https://github.com/lm-sys/FastChat). Thanks for their wonderful works!
| 15,179 | [
[
-0.042694091796875,
-0.0736083984375,
0.025787353515625,
0.01262664794921875,
0.0161285400390625,
0.006938934326171875,
-0.0211639404296875,
-0.02081298828125,
0.00553131103515625,
0.0247955322265625,
-0.0706787109375,
-0.059661865234375,
-0.0380859375,
-0.0... |
Wanfq/Explore_Instruct_Brainstorming_16k | 2023-10-16T02:18:38.000Z | [
"language:en",
"license:cc-by-nc-4.0",
"arxiv:2310.09168",
"region:us"
] | Wanfq | null | null | 4 | 0 | 2023-10-12T14:28:06 | ---
license: cc-by-nc-4.0
language:
- en
---
<p align="center" width="100%">
</p>
<div id="top" align="center">
**Explore-Instruct: Enhancing Domain-Specific Instruction Coverage through Active Exploration**
<h4> |<a href="https://arxiv.org/abs/2310.09168"> 📑 Paper </a> |
<a href="https://huggingface.co/datasets?sort=trending&search=Explore_Instruct"> 🤗 Data </a> |
<a href="https://huggingface.co/models?sort=trending&search=Explore-LM"> 🤗 Model </a> |
<a href="https://github.com/fanqiwan/Explore-Instruct"> 🐱 Github Repo </a> |
</h4>
<!-- **Authors:** -->
_**Fanqi Wan<sup>†</sup>, Xinting Huang<sup>‡</sup>, Tao Yang<sup>†</sup>, Xiaojun Quan<sup>†</sup>, Wei Bi<sup>‡</sup>, Shuming Shi<sup>‡</sup>**_
<!-- **Affiliations:** -->
_<sup>†</sup> Sun Yat-sen University,
<sup>‡</sup> Tencent AI Lab_
</div>
## News
- **Oct 16, 2023:** 🔥 We're excited to announce that the Explore-Instruct datasets in brainstorming, rewriting, and math domains are now available on 🤗 [Huggingface Datasets](https://huggingface.co/datasets?sort=trending&search=Explore_Instruct)! Additionally, we've released Explore-LM models that have been initialized with LLaMA-7B and fine-tuned with the Explore-Instruct data in each domain. You can find these models on 🤗 [Huggingface Models](https://huggingface.co/models?sort=trending&search=Explore-LM). Happy exploring and instructing!
## Contents
- [Overview](#overview)
- [Data Release](#data-release)
- [Model Release](#model-release)
- [Data Generation Process](#data-generation-process)
- [Fine-tuning](#fine-tuning)
- [Evaluation](#evaluation)
- [Limitations](#limitations)
- [License](#license)
- [Citation](#citation)
- [Acknowledgements](#acknowledgments)
## Overview
We propose Explore-Instruct, a novel approach to enhancing domain-specific instruction coverage. We posit that the domain space is inherently structured akin to a tree, reminiscent of cognitive science ontologies. Drawing from the essence of classical search algorithms and incorporating the power of LLMs, Explore-Instruct is conceived to actively traverse the domain space and generate instruction-tuning data, **not** necessitating a predefined tree structure. Specifically, Explore-Instruct employs two strategic operations: lookahead and backtracking exploration:
- **Lookahead** delves into a multitude of potential fine-grained sub-tasks, thereby mapping out a complex network of tasks
- **Backtracking** seeks alternative branches to widen the search boundary, hence extending the domain spectrum.
<p align="center">
<img src="https://github.com/fanqiwan/Explore-Instruct/blob/main/assets/fig2.png?raw=true" width="95%"> <br>
</p>
## Data Release
We release the Explore-Instruct data in brainstorming, rewriting, and math domains on 🤗 [Huggingface Datasets](https://huggingface.co/datasets?sort=trending&search=Explore_Instruct). Each domain includes two versions of datasets: the basic and extended version. The base version contains 10k instruction-tuning data and the extended version contains 16k, 32k, and 64k instruction-tuning data for each domain respectively. Each dataset is a structured data file in the JSON format. It consists of a list of dictionaries, with each dictionary containing the following fields:
- `instruction`: `str`, describes the task the model should perform.
- `input`: `str`, optional context or input for the task.
- `output`: `str`, ground-truth output text for the task and input text.
The results of data-centric analysis are shown as follows:
<p align="left">
<img src="https://github.com/fanqiwan/Explore-Instruct/blob/main/assets/fig1.png?raw=true" width="50%"> <br>
</p>
| Method | Brainstorming Unique<br/>V-N pairs | Rewriting Unique<br/>V-N pairs | Math Unique<br/>V-N pairs |
|:--------------------------------|:----------------------------------:|:------------------------------:|:-------------------------:|
| _Domain-Specific Human-Curated_ | 2 | 8 | 3 |
| _Domain-Aware Self-Instruct_ | 781 | 1715 | 451 |
| Explore-Instruct | **790** | **2015** | **917** |
## Model Release
We release the Explore-LM models in brainstorming, rewriting, and math domains on 🤗 [Huggingface Models](https://huggingface.co/models?sort=trending&search=Explore-LM). Each domain includes two versions of models: the basic and extended version trained with the corresponding version of dataset.
The results of automatic and human evaluation in three domains are shown as follows:
- Automatic evaluation:
| Automatic Comparison in the Brainstorming Domain | Win:Tie:Lose | Beat Rate |
|:-------------------------------------------------|:------------:|:---------:|
| Explore-LM vs Domain-Curated-LM | 194:1:13 | 93.72 |
| Explore-LM-Ext vs Domain-Curated-LM | 196:1:11 | 94.69 |
| Explore-LM vs Domain-Instruct-LM | 114:56:38 | 75.00 |
| Explore-LM-Ext vs Domain-Instruct-LM | 122:55:31 | 79.74 |
| Explore-LM vs ChatGPT | 52:71:85 | 37.96 |
| Explore-LM-Ext vs ChatGPT | 83:69:56 | 59.71 |
| Automatic Comparison in the Rewriting Domain | Win:Tie:Lose | Beat Rate |
|:---------------------------------------------|:------------:|:---------:|
| Explore-LM vs Domain-Curated-LM | 50:38:6 | 89.29 |
| Explore-LM-Ext vs Domain-Curated-LM | 53:37:4 | 92.98 |
| Explore-LM vs Domain-Instruct-LM | 34:49:11 | 75.56 |
| Explore-LM-Ext vs Domain-Instruct-LM | 35:53:6 | 85.37 |
| Explore-LM vs ChatGPT | 11:59:24 | 31.43 |
| Explore-LM-Ext vs ChatGPT | 12:56:26 | 31.58 |
| Automatic Comparison in the Math Domain | Accuracy Rate |
|:----------------------------------------|:-------------:|
| Domain-Curated-LM | 3.4 |
| Domain-Instruct-LM | 4.0 |
| Explore-LM | 6.8 |
| Explore-LM-Ext | 8.4 |
| ChatGPT | 34.8 |
- Human evaluation:
<p align="left">
<img src="https://github.com/fanqiwan/Explore-Instruct/blob/main/assets/fig5.png?raw=true" width="95%"> <br>
</p>
## Data Generation Process
To generate the domain-specific instruction-tuning data, please follow the following commands step by step:
### Domain Space Exploration
```
python3 generate_instruction.py \
--action extend \
--save_dir ./en_data/demo_domain \ # input dir include current domain tree for exploration
--out_dir ./en_data/demo_domain_exploration \ # output dir of the explored new domain tree
--lang <LANGUAGE> \ # currently support 'en'
--domain demo_domain \ # domain for exploration
--extend_nums <TASK_NUMBER_DEPTH_0>,...,<TASK_NUMBER_DEPTH_MAX_DEPTH-1> \ # exploration breadth at each depth
--max_depth <MAX_DEPTH> \ # exploration depth
--assistant_name <ASSISTANT_NAME> # currently support openai and claude
```
### Instruction-Tuning Data Generation
```
python3 generate_instruction.py \
--action enrich \
--save_dir ./en_data/demo_domain_exploration \ # input dir include current domain tree for data generation
--out_dir ./en_data/demo_domain_generation \ # output dir of the domain tree with generated data
--lang <LANGUAGE> \ # currently support 'en'
--domain demo_domain \ # domain for exploration
--enrich_nums <DATA_NUMBER_DEPTH_0>,...,<DATA_NUMBER_DEPTH_MAX_DEPTH> \ # data number for task at each depth
--enrich_batch_size <BATCH_SIZE> \ # batch size for data generation
--assistant_name <ASSISTANT_NAME> # currently support openai and claude
```
### Task Pruning
```
python3 generate_instruction.py \
--action prune \
--save_dir ./en_data/demo_domain_generation \ # input dir include current domain tree for task pruning
--out_dir ./en_data/demo_domain_pruning \ # output dir of the domain tree with 'pruned_subtasks_name.json' file
--lang <LANGUAGE> \ # currently support 'en'
--domain demo_domain \ # domain for exploration
--pruned_file ./en_data/demo_domain_pruning/pruned_subtasks_name.json \ # file of pruned tasks
--prune_threshold <PRUNE_THRESHOLD> \ # threshold of rouge-l overlap between task names
--assistant_name <ASSISTANT_NAME> # currently support openai and claude
```
### Data Filtering
```
python3 generate_instruction.py \
--action filter \
--save_dir ./en_data/demo_domain_pruning \ # input dir include current domain tree for data filtering
--out_dir ./en_data/demo_domain_filtering \ # output dir of the domain tree with fitered data
--lang <LANGUAGE> \ # currently support 'en'
--domain demo_domain \ # domain for exploration
--pruned_file ./en_data/demo_domain_pruning/pruned_subtasks_name.json \ # file of pruned tasks
--filter_threshold <FILTER_THRESHOLD> \ # threshold of rouge-l overlap between instructions
--assistant_name <ASSISTANT_NAME> # currently support openai and claude
```
### Data Sampling
```
python3 generate_instruction.py \
--action sample \
--save_dir ./en_data/demo_domain_filtering \ # input dir include current domain tree for data sampling
--out_dir ./en_data/demo_domain_sampling \ # output dir of the domain tree with sampled data
--lang <LANGUAGE> \ # currently support 'en'
--domain demo_domain \ # domain for exploration
--pruned_file ./en_data/demo_domain_filtering/pruned_subtasks_name.json \ # file of pruned tasks
--sample_example_num <SAMPLE_EXAMPLES_NUM> \ # number of sampled examples
--sample_max_depth <SAMPLE_MAX_DEPTH> \ # max depth for data sampling
--sample_use_pruned \ # do not sample from pruned tasks
--assistant_name <ASSISTANT_NAME> # currently support openai and claude
```
## Fine-tuning
We fine-tune LLaMA-7B with the following hyperparameters:
| Hyperparameter | Global Batch Size | Learning rate | Epochs | Max length | Weight decay |
|:----------------|-------------------:|---------------:|--------:|------------:|--------------:|
| LLaMA 7B | 128 | 2e-5 | 3 | 512| 0 |
To reproduce the training procedure, please use the following command:
```
deepspeed --num_gpus=8 ./train/train.py \
--deepspeed ./deepspeed_config/deepspeed_zero3_offload_config.json \
--model_name_or_path decapoda-research/llama-7b-hf \
--data_path ./en_data/demo_domain_sampling \
--fp16 True \
--output_dir ./training_results/explore-lm-7b-demo-domain \
--num_train_epochs 3 \
--per_device_train_batch_size 2 \
--per_device_eval_batch_size 2 \
--gradient_accumulation_steps 8 \
--evaluation_strategy "no" \
--model_max_length 512 \
--save_strategy "steps" \
--save_steps 2000 \
--save_total_limit 1 \
--learning_rate 2e-5 \
--weight_decay 0. \
--warmup_ratio 0.03 \
--lr_scheduler_type "cosine" \
--logging_steps 1 \
--prompt_type alpaca \
2>&1 | tee ./training_logs/explore-lm-7b-demo-domain.log
python3 ./train/zero_to_fp32.py \
--checkpoint_dir ./training_results/explore-lm-7b-demo-domain \
--output_file ./training_results/explore-lm-7b-demo-domain/pytorch_model.bin
```
## Evaluation
The evaluation datasets for different domains are as follows:
- Brainstorming and Rewriting: From the corresponding categories in the translated test set of BELLE. ([en_eval_set.jsonl](./eval/question/en_eval_set.jsonl))
- Math: From randomly selected 500 questions from the test set of MATH. ([MATH_eval_set_sample.jsonl](./eval/question/MATH_eval_set_sample.jsonl))
The evaluation metrics for different domains are as follows:
- Brainstorming and Rewriting: Both automatic and human evaluations following Vicuna.
- Math: Accuracy Rate metric in solving math problems.
The automatic evaluation commands for different domains are as follows:
```
# Brainstorming and Rewriting Domain
# 1. Inference
python3 ./eval/generate.py \
--model_id <MODEL_ID> \
--model_path <MODEL_PATH> \
--question_file ./eval/question/en_eval_set.jsonl \
--answer_file ./eval/answer/<MODEL_ID>.jsonl \
--num_gpus 8 \
--num_beams 1 \
--temperature 0.7 \
--max_new_tokens 512 \
--prompt_type alpaca \
--do_sample
# 2. Evaluation
python3 ./eval/chatgpt_score.py \
--baseline_file ./eval/answer/<MODEL_1>.jsonl \ # answer of baseline model to compare with
--answer_file ./eval/answer/<MODEL_2>.jsonl \ # answer of evaluation model
--review_file ./eval/review/<MODEL_1>_cp_<MODEL_2>_<DOMAIN>.jsonl \ # review from chatgpt
--prompt_file ./eval/prompt/en_review_prompt_compare.jsonl \ # evaluation prompt for chatgpt
--target_classes <DOMAIN> \ # evaluation domain
--batch_size <BATCH_SIZE> \
--review_model "gpt-3.5-turbo-0301"
```
```
# Math Domain
# 1. Inference
python3 ./eval/generate.py \
--model_id <MODEL_ID> \
--model_path <MODEL_PATH> \
--question_file ./eval/question/MATH_eval_set_sample.jsonl \
--answer_file ./eval/answer/<MODEL_ID>.jsonl \
--num_gpus 8 \
--num_beams 10 \
--temperature 1.0 \
--max_new_tokens 512 \
--prompt_type alpaca
# 2. Evaluation
python3 ./eval/auto_eval.py \
--question_file ./eval/question/MATH_eval_set_sample.jsonl \
--answer_file ./eval/answer/<MODEL_ID>.jsonl # answer of evaluation model
```
## Limitations
Explore-Instruct is still under development and needs a lot of improvements. We acknowledge that our work focuses on the enhancement of domain-specific instruction coverage and does not address other aspects of instruction-tuning, such as the generation of complex and challenging instructions or the mitigation of toxic and harmful instructions. Future work is needed to explore the potential of our approach in these areas.
## License
Explore-Instruct is intended and licensed for research use only. The dataset is CC BY NC 4.0 (allowing only non-commercial use) and models trained using the dataset should not be used outside of research purposes. The weights of Explore-LM models are also CC BY NC 4.0 (allowing only non-commercial use).
## Citation
If you find this work is relevant with your research or applications, please feel free to cite our work!
```
@misc{wan2023explore,
title={Explore-Instruct: Enhancing Domain-Specific Instruction Coverage through Active Exploration},
author={Fanqi, Wan and Xinting, Huang and Tao, Yang and Xiaojun, Quan and Wei, Bi and Shuming, Shi},
year={2023},
eprint={2310.09168},
archivePrefix={arXiv},
primaryClass={cs.CL}
}
```
## Acknowledgments
This repo benefits from [Stanford-Alpaca](https://github.com/tatsu-lab/stanford_alpaca) and [Vicuna](https://github.com/lm-sys/FastChat). Thanks for their wonderful works!
| 15,179 | [
[
-0.042694091796875,
-0.0736083984375,
0.025787353515625,
0.01262664794921875,
0.0161285400390625,
0.006938934326171875,
-0.0211639404296875,
-0.02081298828125,
0.00553131103515625,
0.0247955322265625,
-0.0706787109375,
-0.059661865234375,
-0.0380859375,
-0.0... |
Wanfq/Explore_Instruct_Math_10k | 2023-10-16T02:19:13.000Z | [
"language:en",
"license:cc-by-nc-4.0",
"arxiv:2310.09168",
"region:us"
] | Wanfq | null | null | 1 | 0 | 2023-10-12T14:29:28 | ---
license: cc-by-nc-4.0
language:
- en
---
<p align="center" width="100%">
</p>
<div id="top" align="center">
**Explore-Instruct: Enhancing Domain-Specific Instruction Coverage through Active Exploration**
<h4> |<a href="https://arxiv.org/abs/2310.09168"> 📑 Paper </a> |
<a href="https://huggingface.co/datasets?sort=trending&search=Explore_Instruct"> 🤗 Data </a> |
<a href="https://huggingface.co/models?sort=trending&search=Explore-LM"> 🤗 Model </a> |
<a href="https://github.com/fanqiwan/Explore-Instruct"> 🐱 Github Repo </a> |
</h4>
<!-- **Authors:** -->
_**Fanqi Wan<sup>†</sup>, Xinting Huang<sup>‡</sup>, Tao Yang<sup>†</sup>, Xiaojun Quan<sup>†</sup>, Wei Bi<sup>‡</sup>, Shuming Shi<sup>‡</sup>**_
<!-- **Affiliations:** -->
_<sup>†</sup> Sun Yat-sen University,
<sup>‡</sup> Tencent AI Lab_
</div>
## News
- **Oct 16, 2023:** 🔥 We're excited to announce that the Explore-Instruct datasets in brainstorming, rewriting, and math domains are now available on 🤗 [Huggingface Datasets](https://huggingface.co/datasets?sort=trending&search=Explore_Instruct)! Additionally, we've released Explore-LM models that have been initialized with LLaMA-7B and fine-tuned with the Explore-Instruct data in each domain. You can find these models on 🤗 [Huggingface Models](https://huggingface.co/models?sort=trending&search=Explore-LM). Happy exploring and instructing!
## Contents
- [Overview](#overview)
- [Data Release](#data-release)
- [Model Release](#model-release)
- [Data Generation Process](#data-generation-process)
- [Fine-tuning](#fine-tuning)
- [Evaluation](#evaluation)
- [Limitations](#limitations)
- [License](#license)
- [Citation](#citation)
- [Acknowledgements](#acknowledgments)
## Overview
We propose Explore-Instruct, a novel approach to enhancing domain-specific instruction coverage. We posit that the domain space is inherently structured akin to a tree, reminiscent of cognitive science ontologies. Drawing from the essence of classical search algorithms and incorporating the power of LLMs, Explore-Instruct is conceived to actively traverse the domain space and generate instruction-tuning data, **not** necessitating a predefined tree structure. Specifically, Explore-Instruct employs two strategic operations: lookahead and backtracking exploration:
- **Lookahead** delves into a multitude of potential fine-grained sub-tasks, thereby mapping out a complex network of tasks
- **Backtracking** seeks alternative branches to widen the search boundary, hence extending the domain spectrum.
<p align="center">
<img src="https://github.com/fanqiwan/Explore-Instruct/blob/main/assets/fig2.png?raw=true" width="95%"> <br>
</p>
## Data Release
We release the Explore-Instruct data in brainstorming, rewriting, and math domains on 🤗 [Huggingface Datasets](https://huggingface.co/datasets?sort=trending&search=Explore_Instruct). Each domain includes two versions of datasets: the basic and extended version. The base version contains 10k instruction-tuning data and the extended version contains 16k, 32k, and 64k instruction-tuning data for each domain respectively. Each dataset is a structured data file in the JSON format. It consists of a list of dictionaries, with each dictionary containing the following fields:
- `instruction`: `str`, describes the task the model should perform.
- `input`: `str`, optional context or input for the task.
- `output`: `str`, ground-truth output text for the task and input text.
The results of data-centric analysis are shown as follows:
<p align="left">
<img src="https://github.com/fanqiwan/Explore-Instruct/blob/main/assets/fig1.png?raw=true" width="50%"> <br>
</p>
| Method | Brainstorming Unique<br/>V-N pairs | Rewriting Unique<br/>V-N pairs | Math Unique<br/>V-N pairs |
|:--------------------------------|:----------------------------------:|:------------------------------:|:-------------------------:|
| _Domain-Specific Human-Curated_ | 2 | 8 | 3 |
| _Domain-Aware Self-Instruct_ | 781 | 1715 | 451 |
| Explore-Instruct | **790** | **2015** | **917** |
## Model Release
We release the Explore-LM models in brainstorming, rewriting, and math domains on 🤗 [Huggingface Models](https://huggingface.co/models?sort=trending&search=Explore-LM). Each domain includes two versions of models: the basic and extended version trained with the corresponding version of dataset.
The results of automatic and human evaluation in three domains are shown as follows:
- Automatic evaluation:
| Automatic Comparison in the Brainstorming Domain | Win:Tie:Lose | Beat Rate |
|:-------------------------------------------------|:------------:|:---------:|
| Explore-LM vs Domain-Curated-LM | 194:1:13 | 93.72 |
| Explore-LM-Ext vs Domain-Curated-LM | 196:1:11 | 94.69 |
| Explore-LM vs Domain-Instruct-LM | 114:56:38 | 75.00 |
| Explore-LM-Ext vs Domain-Instruct-LM | 122:55:31 | 79.74 |
| Explore-LM vs ChatGPT | 52:71:85 | 37.96 |
| Explore-LM-Ext vs ChatGPT | 83:69:56 | 59.71 |
| Automatic Comparison in the Rewriting Domain | Win:Tie:Lose | Beat Rate |
|:---------------------------------------------|:------------:|:---------:|
| Explore-LM vs Domain-Curated-LM | 50:38:6 | 89.29 |
| Explore-LM-Ext vs Domain-Curated-LM | 53:37:4 | 92.98 |
| Explore-LM vs Domain-Instruct-LM | 34:49:11 | 75.56 |
| Explore-LM-Ext vs Domain-Instruct-LM | 35:53:6 | 85.37 |
| Explore-LM vs ChatGPT | 11:59:24 | 31.43 |
| Explore-LM-Ext vs ChatGPT | 12:56:26 | 31.58 |
| Automatic Comparison in the Math Domain | Accuracy Rate |
|:----------------------------------------|:-------------:|
| Domain-Curated-LM | 3.4 |
| Domain-Instruct-LM | 4.0 |
| Explore-LM | 6.8 |
| Explore-LM-Ext | 8.4 |
| ChatGPT | 34.8 |
- Human evaluation:
<p align="left">
<img src="https://github.com/fanqiwan/Explore-Instruct/blob/main/assets/fig5.png?raw=true" width="95%"> <br>
</p>
## Data Generation Process
To generate the domain-specific instruction-tuning data, please follow the following commands step by step:
### Domain Space Exploration
```
python3 generate_instruction.py \
--action extend \
--save_dir ./en_data/demo_domain \ # input dir include current domain tree for exploration
--out_dir ./en_data/demo_domain_exploration \ # output dir of the explored new domain tree
--lang <LANGUAGE> \ # currently support 'en'
--domain demo_domain \ # domain for exploration
--extend_nums <TASK_NUMBER_DEPTH_0>,...,<TASK_NUMBER_DEPTH_MAX_DEPTH-1> \ # exploration breadth at each depth
--max_depth <MAX_DEPTH> \ # exploration depth
--assistant_name <ASSISTANT_NAME> # currently support openai and claude
```
### Instruction-Tuning Data Generation
```
python3 generate_instruction.py \
--action enrich \
--save_dir ./en_data/demo_domain_exploration \ # input dir include current domain tree for data generation
--out_dir ./en_data/demo_domain_generation \ # output dir of the domain tree with generated data
--lang <LANGUAGE> \ # currently support 'en'
--domain demo_domain \ # domain for exploration
--enrich_nums <DATA_NUMBER_DEPTH_0>,...,<DATA_NUMBER_DEPTH_MAX_DEPTH> \ # data number for task at each depth
--enrich_batch_size <BATCH_SIZE> \ # batch size for data generation
--assistant_name <ASSISTANT_NAME> # currently support openai and claude
```
### Task Pruning
```
python3 generate_instruction.py \
--action prune \
--save_dir ./en_data/demo_domain_generation \ # input dir include current domain tree for task pruning
--out_dir ./en_data/demo_domain_pruning \ # output dir of the domain tree with 'pruned_subtasks_name.json' file
--lang <LANGUAGE> \ # currently support 'en'
--domain demo_domain \ # domain for exploration
--pruned_file ./en_data/demo_domain_pruning/pruned_subtasks_name.json \ # file of pruned tasks
--prune_threshold <PRUNE_THRESHOLD> \ # threshold of rouge-l overlap between task names
--assistant_name <ASSISTANT_NAME> # currently support openai and claude
```
### Data Filtering
```
python3 generate_instruction.py \
--action filter \
--save_dir ./en_data/demo_domain_pruning \ # input dir include current domain tree for data filtering
--out_dir ./en_data/demo_domain_filtering \ # output dir of the domain tree with fitered data
--lang <LANGUAGE> \ # currently support 'en'
--domain demo_domain \ # domain for exploration
--pruned_file ./en_data/demo_domain_pruning/pruned_subtasks_name.json \ # file of pruned tasks
--filter_threshold <FILTER_THRESHOLD> \ # threshold of rouge-l overlap between instructions
--assistant_name <ASSISTANT_NAME> # currently support openai and claude
```
### Data Sampling
```
python3 generate_instruction.py \
--action sample \
--save_dir ./en_data/demo_domain_filtering \ # input dir include current domain tree for data sampling
--out_dir ./en_data/demo_domain_sampling \ # output dir of the domain tree with sampled data
--lang <LANGUAGE> \ # currently support 'en'
--domain demo_domain \ # domain for exploration
--pruned_file ./en_data/demo_domain_filtering/pruned_subtasks_name.json \ # file of pruned tasks
--sample_example_num <SAMPLE_EXAMPLES_NUM> \ # number of sampled examples
--sample_max_depth <SAMPLE_MAX_DEPTH> \ # max depth for data sampling
--sample_use_pruned \ # do not sample from pruned tasks
--assistant_name <ASSISTANT_NAME> # currently support openai and claude
```
## Fine-tuning
We fine-tune LLaMA-7B with the following hyperparameters:
| Hyperparameter | Global Batch Size | Learning rate | Epochs | Max length | Weight decay |
|:----------------|-------------------:|---------------:|--------:|------------:|--------------:|
| LLaMA 7B | 128 | 2e-5 | 3 | 512| 0 |
To reproduce the training procedure, please use the following command:
```
deepspeed --num_gpus=8 ./train/train.py \
--deepspeed ./deepspeed_config/deepspeed_zero3_offload_config.json \
--model_name_or_path decapoda-research/llama-7b-hf \
--data_path ./en_data/demo_domain_sampling \
--fp16 True \
--output_dir ./training_results/explore-lm-7b-demo-domain \
--num_train_epochs 3 \
--per_device_train_batch_size 2 \
--per_device_eval_batch_size 2 \
--gradient_accumulation_steps 8 \
--evaluation_strategy "no" \
--model_max_length 512 \
--save_strategy "steps" \
--save_steps 2000 \
--save_total_limit 1 \
--learning_rate 2e-5 \
--weight_decay 0. \
--warmup_ratio 0.03 \
--lr_scheduler_type "cosine" \
--logging_steps 1 \
--prompt_type alpaca \
2>&1 | tee ./training_logs/explore-lm-7b-demo-domain.log
python3 ./train/zero_to_fp32.py \
--checkpoint_dir ./training_results/explore-lm-7b-demo-domain \
--output_file ./training_results/explore-lm-7b-demo-domain/pytorch_model.bin
```
## Evaluation
The evaluation datasets for different domains are as follows:
- Brainstorming and Rewriting: From the corresponding categories in the translated test set of BELLE. ([en_eval_set.jsonl](./eval/question/en_eval_set.jsonl))
- Math: From randomly selected 500 questions from the test set of MATH. ([MATH_eval_set_sample.jsonl](./eval/question/MATH_eval_set_sample.jsonl))
The evaluation metrics for different domains are as follows:
- Brainstorming and Rewriting: Both automatic and human evaluations following Vicuna.
- Math: Accuracy Rate metric in solving math problems.
The automatic evaluation commands for different domains are as follows:
```
# Brainstorming and Rewriting Domain
# 1. Inference
python3 ./eval/generate.py \
--model_id <MODEL_ID> \
--model_path <MODEL_PATH> \
--question_file ./eval/question/en_eval_set.jsonl \
--answer_file ./eval/answer/<MODEL_ID>.jsonl \
--num_gpus 8 \
--num_beams 1 \
--temperature 0.7 \
--max_new_tokens 512 \
--prompt_type alpaca \
--do_sample
# 2. Evaluation
python3 ./eval/chatgpt_score.py \
--baseline_file ./eval/answer/<MODEL_1>.jsonl \ # answer of baseline model to compare with
--answer_file ./eval/answer/<MODEL_2>.jsonl \ # answer of evaluation model
--review_file ./eval/review/<MODEL_1>_cp_<MODEL_2>_<DOMAIN>.jsonl \ # review from chatgpt
--prompt_file ./eval/prompt/en_review_prompt_compare.jsonl \ # evaluation prompt for chatgpt
--target_classes <DOMAIN> \ # evaluation domain
--batch_size <BATCH_SIZE> \
--review_model "gpt-3.5-turbo-0301"
```
```
# Math Domain
# 1. Inference
python3 ./eval/generate.py \
--model_id <MODEL_ID> \
--model_path <MODEL_PATH> \
--question_file ./eval/question/MATH_eval_set_sample.jsonl \
--answer_file ./eval/answer/<MODEL_ID>.jsonl \
--num_gpus 8 \
--num_beams 10 \
--temperature 1.0 \
--max_new_tokens 512 \
--prompt_type alpaca
# 2. Evaluation
python3 ./eval/auto_eval.py \
--question_file ./eval/question/MATH_eval_set_sample.jsonl \
--answer_file ./eval/answer/<MODEL_ID>.jsonl # answer of evaluation model
```
## Limitations
Explore-Instruct is still under development and needs a lot of improvements. We acknowledge that our work focuses on the enhancement of domain-specific instruction coverage and does not address other aspects of instruction-tuning, such as the generation of complex and challenging instructions or the mitigation of toxic and harmful instructions. Future work is needed to explore the potential of our approach in these areas.
## License
Explore-Instruct is intended and licensed for research use only. The dataset is CC BY NC 4.0 (allowing only non-commercial use) and models trained using the dataset should not be used outside of research purposes. The weights of Explore-LM models are also CC BY NC 4.0 (allowing only non-commercial use).
## Citation
If you find this work is relevant with your research or applications, please feel free to cite our work!
```
@misc{wan2023explore,
title={Explore-Instruct: Enhancing Domain-Specific Instruction Coverage through Active Exploration},
author={Fanqi, Wan and Xinting, Huang and Tao, Yang and Xiaojun, Quan and Wei, Bi and Shuming, Shi},
year={2023},
eprint={2310.09168},
archivePrefix={arXiv},
primaryClass={cs.CL}
}
```
## Acknowledgments
This repo benefits from [Stanford-Alpaca](https://github.com/tatsu-lab/stanford_alpaca) and [Vicuna](https://github.com/lm-sys/FastChat). Thanks for their wonderful works!
| 15,179 | [
[
-0.042694091796875,
-0.0736083984375,
0.025787353515625,
0.01262664794921875,
0.0161285400390625,
0.006938934326171875,
-0.0211639404296875,
-0.02081298828125,
0.00553131103515625,
0.0247955322265625,
-0.0706787109375,
-0.059661865234375,
-0.0380859375,
-0.0... |
Wanfq/Explore_Instruct_Math_64k | 2023-10-16T02:19:56.000Z | [
"language:en",
"license:cc-by-nc-4.0",
"arxiv:2310.09168",
"region:us"
] | Wanfq | null | null | 1 | 0 | 2023-10-12T14:29:49 | ---
license: cc-by-nc-4.0
language:
- en
---
<p align="center" width="100%">
</p>
<div id="top" align="center">
**Explore-Instruct: Enhancing Domain-Specific Instruction Coverage through Active Exploration**
<h4> |<a href="https://arxiv.org/abs/2310.09168"> 📑 Paper </a> |
<a href="https://huggingface.co/datasets?sort=trending&search=Explore_Instruct"> 🤗 Data </a> |
<a href="https://huggingface.co/models?sort=trending&search=Explore-LM"> 🤗 Model </a> |
<a href="https://github.com/fanqiwan/Explore-Instruct"> 🐱 Github Repo </a> |
</h4>
<!-- **Authors:** -->
_**Fanqi Wan<sup>†</sup>, Xinting Huang<sup>‡</sup>, Tao Yang<sup>†</sup>, Xiaojun Quan<sup>†</sup>, Wei Bi<sup>‡</sup>, Shuming Shi<sup>‡</sup>**_
<!-- **Affiliations:** -->
_<sup>†</sup> Sun Yat-sen University,
<sup>‡</sup> Tencent AI Lab_
</div>
## News
- **Oct 16, 2023:** 🔥 We're excited to announce that the Explore-Instruct datasets in brainstorming, rewriting, and math domains are now available on 🤗 [Huggingface Datasets](https://huggingface.co/datasets?sort=trending&search=Explore_Instruct)! Additionally, we've released Explore-LM models that have been initialized with LLaMA-7B and fine-tuned with the Explore-Instruct data in each domain. You can find these models on 🤗 [Huggingface Models](https://huggingface.co/models?sort=trending&search=Explore-LM). Happy exploring and instructing!
## Contents
- [Overview](#overview)
- [Data Release](#data-release)
- [Model Release](#model-release)
- [Data Generation Process](#data-generation-process)
- [Fine-tuning](#fine-tuning)
- [Evaluation](#evaluation)
- [Limitations](#limitations)
- [License](#license)
- [Citation](#citation)
- [Acknowledgements](#acknowledgments)
## Overview
We propose Explore-Instruct, a novel approach to enhancing domain-specific instruction coverage. We posit that the domain space is inherently structured akin to a tree, reminiscent of cognitive science ontologies. Drawing from the essence of classical search algorithms and incorporating the power of LLMs, Explore-Instruct is conceived to actively traverse the domain space and generate instruction-tuning data, **not** necessitating a predefined tree structure. Specifically, Explore-Instruct employs two strategic operations: lookahead and backtracking exploration:
- **Lookahead** delves into a multitude of potential fine-grained sub-tasks, thereby mapping out a complex network of tasks
- **Backtracking** seeks alternative branches to widen the search boundary, hence extending the domain spectrum.
<p align="center">
<img src="https://github.com/fanqiwan/Explore-Instruct/blob/main/assets/fig2.png?raw=true" width="95%"> <br>
</p>
## Data Release
We release the Explore-Instruct data in brainstorming, rewriting, and math domains on 🤗 [Huggingface Datasets](https://huggingface.co/datasets?sort=trending&search=Explore_Instruct). Each domain includes two versions of datasets: the basic and extended version. The base version contains 10k instruction-tuning data and the extended version contains 16k, 32k, and 64k instruction-tuning data for each domain respectively. Each dataset is a structured data file in the JSON format. It consists of a list of dictionaries, with each dictionary containing the following fields:
- `instruction`: `str`, describes the task the model should perform.
- `input`: `str`, optional context or input for the task.
- `output`: `str`, ground-truth output text for the task and input text.
The results of data-centric analysis are shown as follows:
<p align="left">
<img src="https://github.com/fanqiwan/Explore-Instruct/blob/main/assets/fig1.png?raw=true" width="50%"> <br>
</p>
| Method | Brainstorming Unique<br/>V-N pairs | Rewriting Unique<br/>V-N pairs | Math Unique<br/>V-N pairs |
|:--------------------------------|:----------------------------------:|:------------------------------:|:-------------------------:|
| _Domain-Specific Human-Curated_ | 2 | 8 | 3 |
| _Domain-Aware Self-Instruct_ | 781 | 1715 | 451 |
| Explore-Instruct | **790** | **2015** | **917** |
## Model Release
We release the Explore-LM models in brainstorming, rewriting, and math domains on 🤗 [Huggingface Models](https://huggingface.co/models?sort=trending&search=Explore-LM). Each domain includes two versions of models: the basic and extended version trained with the corresponding version of dataset.
The results of automatic and human evaluation in three domains are shown as follows:
- Automatic evaluation:
| Automatic Comparison in the Brainstorming Domain | Win:Tie:Lose | Beat Rate |
|:-------------------------------------------------|:------------:|:---------:|
| Explore-LM vs Domain-Curated-LM | 194:1:13 | 93.72 |
| Explore-LM-Ext vs Domain-Curated-LM | 196:1:11 | 94.69 |
| Explore-LM vs Domain-Instruct-LM | 114:56:38 | 75.00 |
| Explore-LM-Ext vs Domain-Instruct-LM | 122:55:31 | 79.74 |
| Explore-LM vs ChatGPT | 52:71:85 | 37.96 |
| Explore-LM-Ext vs ChatGPT | 83:69:56 | 59.71 |
| Automatic Comparison in the Rewriting Domain | Win:Tie:Lose | Beat Rate |
|:---------------------------------------------|:------------:|:---------:|
| Explore-LM vs Domain-Curated-LM | 50:38:6 | 89.29 |
| Explore-LM-Ext vs Domain-Curated-LM | 53:37:4 | 92.98 |
| Explore-LM vs Domain-Instruct-LM | 34:49:11 | 75.56 |
| Explore-LM-Ext vs Domain-Instruct-LM | 35:53:6 | 85.37 |
| Explore-LM vs ChatGPT | 11:59:24 | 31.43 |
| Explore-LM-Ext vs ChatGPT | 12:56:26 | 31.58 |
| Automatic Comparison in the Math Domain | Accuracy Rate |
|:----------------------------------------|:-------------:|
| Domain-Curated-LM | 3.4 |
| Domain-Instruct-LM | 4.0 |
| Explore-LM | 6.8 |
| Explore-LM-Ext | 8.4 |
| ChatGPT | 34.8 |
- Human evaluation:
<p align="left">
<img src="https://github.com/fanqiwan/Explore-Instruct/blob/main/assets/fig5.png?raw=true" width="95%"> <br>
</p>
## Data Generation Process
To generate the domain-specific instruction-tuning data, please follow the following commands step by step:
### Domain Space Exploration
```
python3 generate_instruction.py \
--action extend \
--save_dir ./en_data/demo_domain \ # input dir include current domain tree for exploration
--out_dir ./en_data/demo_domain_exploration \ # output dir of the explored new domain tree
--lang <LANGUAGE> \ # currently support 'en'
--domain demo_domain \ # domain for exploration
--extend_nums <TASK_NUMBER_DEPTH_0>,...,<TASK_NUMBER_DEPTH_MAX_DEPTH-1> \ # exploration breadth at each depth
--max_depth <MAX_DEPTH> \ # exploration depth
--assistant_name <ASSISTANT_NAME> # currently support openai and claude
```
### Instruction-Tuning Data Generation
```
python3 generate_instruction.py \
--action enrich \
--save_dir ./en_data/demo_domain_exploration \ # input dir include current domain tree for data generation
--out_dir ./en_data/demo_domain_generation \ # output dir of the domain tree with generated data
--lang <LANGUAGE> \ # currently support 'en'
--domain demo_domain \ # domain for exploration
--enrich_nums <DATA_NUMBER_DEPTH_0>,...,<DATA_NUMBER_DEPTH_MAX_DEPTH> \ # data number for task at each depth
--enrich_batch_size <BATCH_SIZE> \ # batch size for data generation
--assistant_name <ASSISTANT_NAME> # currently support openai and claude
```
### Task Pruning
```
python3 generate_instruction.py \
--action prune \
--save_dir ./en_data/demo_domain_generation \ # input dir include current domain tree for task pruning
--out_dir ./en_data/demo_domain_pruning \ # output dir of the domain tree with 'pruned_subtasks_name.json' file
--lang <LANGUAGE> \ # currently support 'en'
--domain demo_domain \ # domain for exploration
--pruned_file ./en_data/demo_domain_pruning/pruned_subtasks_name.json \ # file of pruned tasks
--prune_threshold <PRUNE_THRESHOLD> \ # threshold of rouge-l overlap between task names
--assistant_name <ASSISTANT_NAME> # currently support openai and claude
```
### Data Filtering
```
python3 generate_instruction.py \
--action filter \
--save_dir ./en_data/demo_domain_pruning \ # input dir include current domain tree for data filtering
--out_dir ./en_data/demo_domain_filtering \ # output dir of the domain tree with fitered data
--lang <LANGUAGE> \ # currently support 'en'
--domain demo_domain \ # domain for exploration
--pruned_file ./en_data/demo_domain_pruning/pruned_subtasks_name.json \ # file of pruned tasks
--filter_threshold <FILTER_THRESHOLD> \ # threshold of rouge-l overlap between instructions
--assistant_name <ASSISTANT_NAME> # currently support openai and claude
```
### Data Sampling
```
python3 generate_instruction.py \
--action sample \
--save_dir ./en_data/demo_domain_filtering \ # input dir include current domain tree for data sampling
--out_dir ./en_data/demo_domain_sampling \ # output dir of the domain tree with sampled data
--lang <LANGUAGE> \ # currently support 'en'
--domain demo_domain \ # domain for exploration
--pruned_file ./en_data/demo_domain_filtering/pruned_subtasks_name.json \ # file of pruned tasks
--sample_example_num <SAMPLE_EXAMPLES_NUM> \ # number of sampled examples
--sample_max_depth <SAMPLE_MAX_DEPTH> \ # max depth for data sampling
--sample_use_pruned \ # do not sample from pruned tasks
--assistant_name <ASSISTANT_NAME> # currently support openai and claude
```
## Fine-tuning
We fine-tune LLaMA-7B with the following hyperparameters:
| Hyperparameter | Global Batch Size | Learning rate | Epochs | Max length | Weight decay |
|:----------------|-------------------:|---------------:|--------:|------------:|--------------:|
| LLaMA 7B | 128 | 2e-5 | 3 | 512| 0 |
To reproduce the training procedure, please use the following command:
```
deepspeed --num_gpus=8 ./train/train.py \
--deepspeed ./deepspeed_config/deepspeed_zero3_offload_config.json \
--model_name_or_path decapoda-research/llama-7b-hf \
--data_path ./en_data/demo_domain_sampling \
--fp16 True \
--output_dir ./training_results/explore-lm-7b-demo-domain \
--num_train_epochs 3 \
--per_device_train_batch_size 2 \
--per_device_eval_batch_size 2 \
--gradient_accumulation_steps 8 \
--evaluation_strategy "no" \
--model_max_length 512 \
--save_strategy "steps" \
--save_steps 2000 \
--save_total_limit 1 \
--learning_rate 2e-5 \
--weight_decay 0. \
--warmup_ratio 0.03 \
--lr_scheduler_type "cosine" \
--logging_steps 1 \
--prompt_type alpaca \
2>&1 | tee ./training_logs/explore-lm-7b-demo-domain.log
python3 ./train/zero_to_fp32.py \
--checkpoint_dir ./training_results/explore-lm-7b-demo-domain \
--output_file ./training_results/explore-lm-7b-demo-domain/pytorch_model.bin
```
## Evaluation
The evaluation datasets for different domains are as follows:
- Brainstorming and Rewriting: From the corresponding categories in the translated test set of BELLE. ([en_eval_set.jsonl](./eval/question/en_eval_set.jsonl))
- Math: From randomly selected 500 questions from the test set of MATH. ([MATH_eval_set_sample.jsonl](./eval/question/MATH_eval_set_sample.jsonl))
The evaluation metrics for different domains are as follows:
- Brainstorming and Rewriting: Both automatic and human evaluations following Vicuna.
- Math: Accuracy Rate metric in solving math problems.
The automatic evaluation commands for different domains are as follows:
```
# Brainstorming and Rewriting Domain
# 1. Inference
python3 ./eval/generate.py \
--model_id <MODEL_ID> \
--model_path <MODEL_PATH> \
--question_file ./eval/question/en_eval_set.jsonl \
--answer_file ./eval/answer/<MODEL_ID>.jsonl \
--num_gpus 8 \
--num_beams 1 \
--temperature 0.7 \
--max_new_tokens 512 \
--prompt_type alpaca \
--do_sample
# 2. Evaluation
python3 ./eval/chatgpt_score.py \
--baseline_file ./eval/answer/<MODEL_1>.jsonl \ # answer of baseline model to compare with
--answer_file ./eval/answer/<MODEL_2>.jsonl \ # answer of evaluation model
--review_file ./eval/review/<MODEL_1>_cp_<MODEL_2>_<DOMAIN>.jsonl \ # review from chatgpt
--prompt_file ./eval/prompt/en_review_prompt_compare.jsonl \ # evaluation prompt for chatgpt
--target_classes <DOMAIN> \ # evaluation domain
--batch_size <BATCH_SIZE> \
--review_model "gpt-3.5-turbo-0301"
```
```
# Math Domain
# 1. Inference
python3 ./eval/generate.py \
--model_id <MODEL_ID> \
--model_path <MODEL_PATH> \
--question_file ./eval/question/MATH_eval_set_sample.jsonl \
--answer_file ./eval/answer/<MODEL_ID>.jsonl \
--num_gpus 8 \
--num_beams 10 \
--temperature 1.0 \
--max_new_tokens 512 \
--prompt_type alpaca
# 2. Evaluation
python3 ./eval/auto_eval.py \
--question_file ./eval/question/MATH_eval_set_sample.jsonl \
--answer_file ./eval/answer/<MODEL_ID>.jsonl # answer of evaluation model
```
## Limitations
Explore-Instruct is still under development and needs a lot of improvements. We acknowledge that our work focuses on the enhancement of domain-specific instruction coverage and does not address other aspects of instruction-tuning, such as the generation of complex and challenging instructions or the mitigation of toxic and harmful instructions. Future work is needed to explore the potential of our approach in these areas.
## License
Explore-Instruct is intended and licensed for research use only. The dataset is CC BY NC 4.0 (allowing only non-commercial use) and models trained using the dataset should not be used outside of research purposes. The weights of Explore-LM models are also CC BY NC 4.0 (allowing only non-commercial use).
## Citation
If you find this work is relevant with your research or applications, please feel free to cite our work!
```
@misc{wan2023explore,
title={Explore-Instruct: Enhancing Domain-Specific Instruction Coverage through Active Exploration},
author={Fanqi, Wan and Xinting, Huang and Tao, Yang and Xiaojun, Quan and Wei, Bi and Shuming, Shi},
year={2023},
eprint={2310.09168},
archivePrefix={arXiv},
primaryClass={cs.CL}
}
```
## Acknowledgments
This repo benefits from [Stanford-Alpaca](https://github.com/tatsu-lab/stanford_alpaca) and [Vicuna](https://github.com/lm-sys/FastChat). Thanks for their wonderful works!
| 15,179 | [
[
-0.042694091796875,
-0.0736083984375,
0.025787353515625,
0.01262664794921875,
0.0161285400390625,
0.006938934326171875,
-0.0211639404296875,
-0.02081298828125,
0.00553131103515625,
0.0247955322265625,
-0.0706787109375,
-0.059661865234375,
-0.0380859375,
-0.0... |
RyuNumchon/artery-ultrasound-siit | 2023-10-12T14:44:48.000Z | [
"region:us"
] | RyuNumchon | null | null | 0 | 0 | 2023-10-12T14:44:36 | ---
dataset_info:
features:
- name: pixel_values
dtype: image
- name: label
dtype: image
splits:
- name: train
num_bytes: 230791779.0
num_examples: 100
download_size: 0
dataset_size: 230791779.0
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
---
# Dataset Card for "artery-ultrasound-siit"
[More Information needed](https://github.com/huggingface/datasets/blob/main/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards) | 491 | [
[
-0.040008544921875,
-0.0071258544921875,
0.03070068359375,
0.00791168212890625,
-0.031463623046875,
-0.0053863525390625,
0.034271240234375,
-0.028472900390625,
0.0701904296875,
0.0211639404296875,
-0.039825439453125,
-0.044342041015625,
-0.061920166015625,
-... |
S3Eval/Easy | 2023-10-12T14:51:43.000Z | [
"region:us"
] | S3Eval | null | null | 0 | 0 | 2023-10-12T14:51:09 | Entry not found | 15 | [
[
-0.0213775634765625,
-0.01497650146484375,
0.05718994140625,
0.02880859375,
-0.0350341796875,
0.046478271484375,
0.052490234375,
0.00507354736328125,
0.051361083984375,
0.0170135498046875,
-0.052093505859375,
-0.01497650146484375,
-0.0604248046875,
0.0379028... |
1aurent/STORK | 2023-10-12T15:49:47.000Z | [
"task_categories:image-classification",
"size_categories:n<1K",
"license:mit",
"biology",
"IVF",
"embryo",
"region:us"
] | 1aurent | null | null | 0 | 0 | 2023-10-12T15:41:07 | ---
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
- split: test
path: data/test-*
dataset_info:
features:
- name: image
dtype: image
- name: label
dtype:
class_label:
names:
'0': good
'1': poor
splits:
- name: train
num_bytes: 4513394
num_examples: 84
- name: test
num_bytes: 729815
num_examples: 14
download_size: 5243240
dataset_size: 5243209
license: mit
task_categories:
- image-classification
tags:
- biology
- IVF
- embryo
size_categories:
- n<1K
---
# Stork
**Homepage**: https://github.com/ih-lab/STORK/ \
**Publication Date**: 2019-01-18 \
**License**: [MIT](https://github.com/ih-lab/STORK/blob/master/LICENSE)
 | 815 | [
[
-0.0124969482421875,
-0.039398193359375,
0.026092529296875,
0.0302276611328125,
-0.037811279296875,
0.0194549560546875,
0.01702880859375,
-0.0333251953125,
0.043365478515625,
0.032318115234375,
-0.0288238525390625,
-0.06378173828125,
-0.02606201171875,
-0.00... |
YiYiXu/testing-images | 2023-10-22T08:41:37.000Z | [
"region:us"
] | YiYiXu | null | null | 0 | 0 | 2023-10-12T15:42:06 | Entry not found | 15 | [
[
-0.02142333984375,
-0.01495361328125,
0.05718994140625,
0.0288238525390625,
-0.035064697265625,
0.046539306640625,
0.052520751953125,
0.005062103271484375,
0.0513916015625,
0.016998291015625,
-0.052093505859375,
-0.014984130859375,
-0.060394287109375,
0.0379... |
Greenvs/FR1-test | 2023-10-12T15:50:11.000Z | [
"region:us"
] | Greenvs | null | null | 0 | 0 | 2023-10-12T15:46:41 | Entry not found | 15 | [
[
-0.02142333984375,
-0.01495361328125,
0.05718994140625,
0.0288238525390625,
-0.035064697265625,
0.046539306640625,
0.052520751953125,
0.005062103271484375,
0.0513916015625,
0.016998291015625,
-0.052093505859375,
-0.014984130859375,
-0.060394287109375,
0.0379... |
xu3kev/BIRD-SQL-data-train | 2023-10-12T16:00:17.000Z | [
"region:us"
] | xu3kev | null | null | 0 | 0 | 2023-10-12T15:56:51 | ---
dataset_info:
features:
- name: db_id
dtype: string
- name: question
dtype: string
- name: evidence
dtype: string
- name: SQL
dtype: string
- name: schema
dtype: string
splits:
- name: train
num_bytes: 49782288
num_examples: 9428
download_size: 2331031
dataset_size: 49782288
---
# Dataset Card for "BIRD-SQL-data-train"
Data from [BIRD-SQL](https://bird-bench.github.io/) benchmark training set.
| 449 | [
[
-0.0128173828125,
-0.0145721435546875,
-0.0200042724609375,
0.009552001953125,
-0.02435302734375,
-0.00042366981506347656,
0.01441192626953125,
-0.0110015869140625,
0.01486968994140625,
0.026458740234375,
-0.050506591796875,
-0.04620361328125,
-0.013442993164062... |
Daniel-Prieto/Dataset-pruebas | 2023-10-12T16:10:05.000Z | [
"region:us"
] | Daniel-Prieto | null | null | 0 | 0 | 2023-10-12T16:10:00 | ---
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
- split: test
path: data/test-*
dataset_info:
features:
- name: image
dtype: image
- name: words
sequence: string
- name: bbox
sequence:
sequence: int64
- name: ner_tags
sequence: string
splits:
- name: train
num_bytes: 13531808.0
num_examples: 150
- name: test
num_bytes: 5449811.0
num_examples: 51
download_size: 18496767
dataset_size: 18981619.0
---
# Dataset Card for "Dataset-pruebas"
[More Information needed](https://github.com/huggingface/datasets/blob/main/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards) | 676 | [
[
-0.0455322265625,
-0.0192718505859375,
0.007350921630859375,
0.0273284912109375,
-0.0207977294921875,
0.00007051229476928711,
0.00814056396484375,
-0.009765625,
0.07666015625,
0.04571533203125,
-0.054290771484375,
-0.05810546875,
-0.0418701171875,
-0.0137405... |
Imofy/pzizo | 2023-10-12T16:12:53.000Z | [
"region:us"
] | Imofy | null | null | 0 | 0 | 2023-10-12T16:12:53 | Entry not found | 15 | [
[
-0.02142333984375,
-0.01495361328125,
0.057220458984375,
0.0288238525390625,
-0.035064697265625,
0.046539306640625,
0.052520751953125,
0.00507354736328125,
0.0513916015625,
0.0169830322265625,
-0.052032470703125,
-0.014984130859375,
-0.060455322265625,
0.037... |
Jaynm31245/answers | 2023-10-12T16:40:55.000Z | [
"region:us"
] | Jaynm31245 | null | null | 0 | 0 | 2023-10-12T16:36:00 | Entry not found | 15 | [
[
-0.02142333984375,
-0.01495361328125,
0.05718994140625,
0.0288238525390625,
-0.035064697265625,
0.046539306640625,
0.052520751953125,
0.005062103271484375,
0.0513916015625,
0.016998291015625,
-0.052093505859375,
-0.014984130859375,
-0.060394287109375,
0.0379... |
Adun/pantip_qa_1000 | 2023-10-12T16:49:40.000Z | [
"region:us"
] | Adun | null | null | 0 | 0 | 2023-10-12T16:47:07 | Entry not found | 15 | [
[
-0.02142333984375,
-0.01495361328125,
0.05718994140625,
0.0288238525390625,
-0.035064697265625,
0.046539306640625,
0.052520751953125,
0.005062103271484375,
0.0513916015625,
0.016998291015625,
-0.052093505859375,
-0.014984130859375,
-0.060394287109375,
0.0379... |
qq67878980/Niggermaxxx_benchmark | 2023-10-12T17:03:43.000Z | [
"license:cc",
"region:us"
] | qq67878980 | null | null | 0 | 0 | 2023-10-12T16:47:09 | ---
license: cc
---
A benchmark for LLMs, real tests for real people, and real usecases. THE benchmark for the ages. | 116 | [
[
-0.0335693359375,
-0.03961181640625,
0.036956787109375,
0.0261993408203125,
-0.0102081298828125,
-0.003726959228515625,
0.022735595703125,
-0.03973388671875,
-0.0127716064453125,
0.021484375,
-0.050445556640625,
-0.0164947509765625,
0.0013227462768554688,
-0... |
mardiek/ML2 | 2023-10-12T17:11:41.000Z | [
"region:us"
] | mardiek | null | null | 0 | 0 | 2023-10-12T17:10:10 | Entry not found | 15 | [
[
-0.0213775634765625,
-0.014984130859375,
0.05718994140625,
0.0288543701171875,
-0.0350341796875,
0.046478271484375,
0.052520751953125,
0.005062103271484375,
0.051361083984375,
0.016998291015625,
-0.0521240234375,
-0.01496124267578125,
-0.0604248046875,
0.037... |
jonasknobloch/cs_squad | 2023-10-12T17:30:21.000Z | [
"task_categories:question-answering",
"task_ids:extractive-qa",
"annotations_creators:crowdsourced",
"language_creators:crowdsourced",
"multilinguality:monolingual",
"size_categories:1K<n<10K",
"source_datasets:original",
"language:cs",
"license:lgpl-3.0",
"czech QA",
"wikipedia QA",
"region:u... | jonasknobloch | null | null | 0 | 0 | 2023-10-12T17:16:09 | ---
annotations_creators:
- crowdsourced
language:
- cs
language_creators:
- crowdsourced
license:
- lgpl-3.0
multilinguality:
- monolingual
pretty_name: Czech Simple Question Answering Dataset
size_categories:
- 1K<n<10K
source_datasets:
- original
tags:
- czech QA
- wikipedia QA
task_categories:
- question-answering
task_ids:
- extractive-qa
---
DO NOT USE
Forked from https://huggingface.co/datasets/fewshot-goes-multilingual/cs_squad-3.0
# Dataset Card for Czech Simple Question Answering Dataset 2.0
This a processed and filtered adaptation of an existing dataset. For raw and larger dataset, see `Dataset Source` section.
## Dataset Description
The data contains questions and answers based on Czech wikipeadia articles.
Each question has an answer (or more) and a selected part of the context as the evidence.
A majority of the answers are extractive - i.e. they are present in the context in the exact form. The remaining cases are
- yes/no questions
- answer is almost in the exact form present in the text, but the form of words was changed to suit the question (declension, ...)
- answered in own words (should be rare, but is not)
All questions in the dataset are answerable from the context. Small minority of questions have multiple answers.
Sometimes it means that any of them is correct (e.g. either "Pacifik" or "Tichý oceán" are correct terms for Pacific Ocean)
and sometimes it means that all of them together are a correct answer (e.g., Who was Leonardo da Vinci? ["painter", "engineer"])
Total number of examples is around:
- 6,250 in train
- 570 in validation
- 850 in test.
## Dataset Features
Each example contains:
- `item_id`: string id of the
- `context`: "reasonably" big chunk (string) of wikipedia article that contains the answer
- `question`: string
- `answers`: list of all answers (string). mostly list of length 1
- `evidence_text`: substring of context (typically one sentence) that is sufficient to answer the question
- `evidence_start`: index in context, such that `context[evidence_start:evidence_end] == evidence_text`
- `evidence_end`: index in context
- `occurences`:
list of (dictionaries) occurences of the answer(s) in the evidence.
Each answer was searched with word boundaries ("\b" in regex) and case-sensitive in the evidence.
If nothing found, try again but case-insensitive.
If nothing found, try again but case-sensitive without word boundaries.
If nothing found, try again but case-insensitive without word boundaries.
This process should supress "false positive" occurences of the answer in the evidence.
- `start`: index in context
- `end`: index in context
- `text`: the answer looked for
- `url`: link to the wikipedia article
- `original_article`: original parsed wikipedia article from which the context is taken
- `question_type`: type of the question, one of: ['ABBREVIATION', 'DATETIME', 'DENOTATION', 'ENTITY', 'LOCATION', 'NUMERIC', 'ORGANIZATION', 'OTHER', 'PERSON', 'YES_NO']
- `answer_type`: type of the answer, one of: ['ABBREVIATION', 'ADJ_PHRASE', 'CLAUSE', 'DATETIME', 'ENTITY', 'LOCATION', 'NUMERIC', 'OTHER', 'PERSON', 'VERB_PHRASE']
## Dataset Source
The dataset is a preprocessed adaptation of existing SQAD 3.0 dataset [link to data](https://lindat.cz/repository/xmlui/handle/11234/1-3069).
This adaptation contains (almost) same data, but converted to a convenient format.
The data was also filtered to remove a statistical bias where the answer was contained
in the first sentence in the article (around 50% of all data in the original dataset, likely
caused by the data collection process).
## Citation
Cite authors of the [original dataset](https://lindat.cz/repository/xmlui/handle/11234/1-3069):
```bibtex
@misc{11234/1-3069,
title = {sqad 3.0},
author = {Medve{\v d}, Marek and Hor{\'a}k, Ale{\v s}},
url = {http://hdl.handle.net/11234/1-3069},
note = {{LINDAT}/{CLARIAH}-{CZ} digital library at the Institute of Formal and Applied Linguistics ({{\'U}FAL}), Faculty of Mathematics and Physics, Charles University},
copyright = {{GNU} Library or "Lesser" General Public License 3.0 ({LGPL}-3.0)},
year = {2019}
}
```
| 4,146 | [
[
-0.03973388671875,
-0.08856201171875,
0.038909912109375,
0.00821685791015625,
-0.02252197265625,
-0.023284912109375,
-0.0181732177734375,
-0.02923583984375,
0.029510498046875,
0.0269317626953125,
-0.05841064453125,
-0.038421630859375,
-0.0297698974609375,
0.... |
dim/camel_ai_physics | 2023-10-12T17:17:57.000Z | [
"region:us"
] | dim | null | null | 0 | 0 | 2023-10-12T17:17:30 | ---
dataset_info:
features:
- name: role_1
dtype: string
- name: topic;
dtype: string
- name: sub_topic
dtype: string
- name: message_1
dtype: string
- name: message_2
dtype: string
splits:
- name: train
num_bytes: 51650490
num_examples: 20000
download_size: 18889012
dataset_size: 51650490
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
---
# Dataset Card for "camel_ai_physics"
[More Information needed](https://github.com/huggingface/datasets/blob/main/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards) | 599 | [
[
-0.0343017578125,
-0.0173492431640625,
-0.01036834716796875,
0.0163726806640625,
-0.0196075439453125,
-0.00455474853515625,
0.022979736328125,
-0.0238189697265625,
0.045257568359375,
0.01374053955078125,
-0.05596923828125,
-0.046600341796875,
-0.0325927734375,
... |
nlplabtdtu/sickr-sts-vi | 2023-10-13T00:52:08.000Z | [
"region:us"
] | nlplabtdtu | null | null | 0 | 0 | 2023-10-12T17:20:37 | Entry not found | 15 | [
[
-0.0213775634765625,
-0.014984130859375,
0.05718994140625,
0.0288543701171875,
-0.0350341796875,
0.046478271484375,
0.052520751953125,
0.005062103271484375,
0.051361083984375,
0.016998291015625,
-0.0521240234375,
-0.01496124267578125,
-0.0604248046875,
0.037... |
dim/camel_ai_biology | 2023-10-12T17:27:15.000Z | [
"region:us"
] | dim | null | null | 0 | 0 | 2023-10-12T17:26:35 | ---
dataset_info:
features:
- name: role_1
dtype: string
- name: topic;
dtype: string
- name: sub_topic
dtype: string
- name: message_1
dtype: string
- name: message_2
dtype: string
splits:
- name: train
num_bytes: 61275986
num_examples: 20000
download_size: 22376128
dataset_size: 61275986
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
---
# Dataset Card for "camel_ai_biology"
[More Information needed](https://github.com/huggingface/datasets/blob/main/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards) | 599 | [
[
-0.0295562744140625,
-0.01995849609375,
-0.0166015625,
0.01123046875,
-0.0209503173828125,
-0.004375457763671875,
0.0263214111328125,
-0.031341552734375,
0.055755615234375,
0.02362060546875,
-0.0521240234375,
-0.05718994140625,
-0.040252685546875,
-0.0080032... |
open-llm-leaderboard/details_Sao10K__Euryale-1.3-L2-70B | 2023-10-26T00:12:02.000Z | [
"region:us"
] | open-llm-leaderboard | null | null | 0 | 0 | 2023-10-12T17:36:47 | ---
pretty_name: Evaluation run of Sao10K/Euryale-1.3-L2-70B
dataset_summary: "Dataset automatically created during the evaluation run of model\
\ [Sao10K/Euryale-1.3-L2-70B](https://huggingface.co/Sao10K/Euryale-1.3-L2-70B)\
\ on the [Open LLM Leaderboard](https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard).\n\
\nThe dataset is composed of 64 configuration, each one coresponding to one of the\
\ evaluated task.\n\nThe 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.\n\
\nAn additional configuration \"results\" store all the aggregated results of the\
\ run (and is used to compute and display the agregated metrics on the [Open LLM\
\ Leaderboard](https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard)).\n\
\nTo load the details from a run, you can for instance do the following:\n```python\n\
from datasets import load_dataset\ndata = load_dataset(\"open-llm-leaderboard/details_Sao10K__Euryale-1.3-L2-70B\"\
,\n\t\"harness_winogrande_5\",\n\tsplit=\"train\")\n```\n\n## Latest results\n\n\
These are the [latest results from run 2023-10-26T00:11:50.324232](https://huggingface.co/datasets/open-llm-leaderboard/details_Sao10K__Euryale-1.3-L2-70B/blob/main/results_2023-10-26T00-11-50.324232.json)(note\
\ that their might be results for other tasks in the repos if successive evals didn't\
\ cover the same tasks. You find each in the results and the \"latest\" split for\
\ each eval):\n\n```python\n{\n \"all\": {\n \"em\": 0.5388003355704698,\n\
\ \"em_stderr\": 0.005105027329360947,\n \"f1\": 0.6009920302013437,\n\
\ \"f1_stderr\": 0.004740248039821831,\n \"acc\": 0.5849328585370874,\n\
\ \"acc_stderr\": 0.011836910620214903\n },\n \"harness|drop|3\": {\n\
\ \"em\": 0.5388003355704698,\n \"em_stderr\": 0.005105027329360947,\n\
\ \"f1\": 0.6009920302013437,\n \"f1_stderr\": 0.004740248039821831\n\
\ },\n \"harness|gsm8k|5\": {\n \"acc\": 0.3419257012888552,\n \
\ \"acc_stderr\": 0.013066089625182799\n },\n \"harness|winogrande|5\"\
: {\n \"acc\": 0.8279400157853196,\n \"acc_stderr\": 0.010607731615247007\n\
\ }\n}\n```"
repo_url: https://huggingface.co/Sao10K/Euryale-1.3-L2-70B
leaderboard_url: https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard
point_of_contact: clementine@hf.co
configs:
- config_name: harness_arc_challenge_25
data_files:
- split: 2023_10_12T17_36_24.431746
path:
- '**/details_harness|arc:challenge|25_2023-10-12T17-36-24.431746.parquet'
- split: latest
path:
- '**/details_harness|arc:challenge|25_2023-10-12T17-36-24.431746.parquet'
- config_name: harness_drop_3
data_files:
- split: 2023_10_26T00_11_50.324232
path:
- '**/details_harness|drop|3_2023-10-26T00-11-50.324232.parquet'
- split: latest
path:
- '**/details_harness|drop|3_2023-10-26T00-11-50.324232.parquet'
- config_name: harness_gsm8k_5
data_files:
- split: 2023_10_26T00_11_50.324232
path:
- '**/details_harness|gsm8k|5_2023-10-26T00-11-50.324232.parquet'
- split: latest
path:
- '**/details_harness|gsm8k|5_2023-10-26T00-11-50.324232.parquet'
- config_name: harness_hellaswag_10
data_files:
- split: 2023_10_12T17_36_24.431746
path:
- '**/details_harness|hellaswag|10_2023-10-12T17-36-24.431746.parquet'
- split: latest
path:
- '**/details_harness|hellaswag|10_2023-10-12T17-36-24.431746.parquet'
- config_name: harness_hendrycksTest_5
data_files:
- split: 2023_10_12T17_36_24.431746
path:
- '**/details_harness|hendrycksTest-abstract_algebra|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-anatomy|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-astronomy|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-business_ethics|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-clinical_knowledge|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-college_biology|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-college_chemistry|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-college_computer_science|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-college_mathematics|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-college_medicine|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-college_physics|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-computer_security|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-conceptual_physics|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-econometrics|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-electrical_engineering|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-elementary_mathematics|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-formal_logic|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-global_facts|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-high_school_biology|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-high_school_chemistry|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-high_school_computer_science|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-high_school_european_history|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-high_school_geography|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-high_school_government_and_politics|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-high_school_macroeconomics|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-high_school_mathematics|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-high_school_microeconomics|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-high_school_physics|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-high_school_psychology|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-high_school_statistics|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-high_school_us_history|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-high_school_world_history|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-human_aging|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-human_sexuality|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-international_law|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-jurisprudence|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-logical_fallacies|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-machine_learning|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-management|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-marketing|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-medical_genetics|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-miscellaneous|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-moral_disputes|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-moral_scenarios|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-nutrition|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-philosophy|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-prehistory|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-professional_accounting|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-professional_law|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-professional_medicine|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-professional_psychology|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-public_relations|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-security_studies|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-sociology|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-us_foreign_policy|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-virology|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-world_religions|5_2023-10-12T17-36-24.431746.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-abstract_algebra|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-anatomy|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-astronomy|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-business_ethics|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-clinical_knowledge|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-college_biology|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-college_chemistry|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-college_computer_science|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-college_mathematics|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-college_medicine|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-college_physics|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-computer_security|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-conceptual_physics|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-econometrics|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-electrical_engineering|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-elementary_mathematics|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-formal_logic|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-global_facts|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-high_school_biology|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-high_school_chemistry|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-high_school_computer_science|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-high_school_european_history|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-high_school_geography|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-high_school_government_and_politics|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-high_school_macroeconomics|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-high_school_mathematics|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-high_school_microeconomics|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-high_school_physics|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-high_school_psychology|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-high_school_statistics|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-high_school_us_history|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-high_school_world_history|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-human_aging|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-human_sexuality|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-international_law|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-jurisprudence|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-logical_fallacies|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-machine_learning|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-management|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-marketing|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-medical_genetics|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-miscellaneous|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-moral_disputes|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-moral_scenarios|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-nutrition|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-philosophy|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-prehistory|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-professional_accounting|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-professional_law|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-professional_medicine|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-professional_psychology|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-public_relations|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-security_studies|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-sociology|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-us_foreign_policy|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-virology|5_2023-10-12T17-36-24.431746.parquet'
- '**/details_harness|hendrycksTest-world_religions|5_2023-10-12T17-36-24.431746.parquet'
- config_name: harness_hendrycksTest_abstract_algebra_5
data_files:
- split: 2023_10_12T17_36_24.431746
path:
- '**/details_harness|hendrycksTest-abstract_algebra|5_2023-10-12T17-36-24.431746.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-abstract_algebra|5_2023-10-12T17-36-24.431746.parquet'
- config_name: harness_hendrycksTest_anatomy_5
data_files:
- split: 2023_10_12T17_36_24.431746
path:
- '**/details_harness|hendrycksTest-anatomy|5_2023-10-12T17-36-24.431746.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-anatomy|5_2023-10-12T17-36-24.431746.parquet'
- config_name: harness_hendrycksTest_astronomy_5
data_files:
- split: 2023_10_12T17_36_24.431746
path:
- '**/details_harness|hendrycksTest-astronomy|5_2023-10-12T17-36-24.431746.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-astronomy|5_2023-10-12T17-36-24.431746.parquet'
- config_name: harness_hendrycksTest_business_ethics_5
data_files:
- split: 2023_10_12T17_36_24.431746
path:
- '**/details_harness|hendrycksTest-business_ethics|5_2023-10-12T17-36-24.431746.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-business_ethics|5_2023-10-12T17-36-24.431746.parquet'
- config_name: harness_hendrycksTest_clinical_knowledge_5
data_files:
- split: 2023_10_12T17_36_24.431746
path:
- '**/details_harness|hendrycksTest-clinical_knowledge|5_2023-10-12T17-36-24.431746.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-clinical_knowledge|5_2023-10-12T17-36-24.431746.parquet'
- config_name: harness_hendrycksTest_college_biology_5
data_files:
- split: 2023_10_12T17_36_24.431746
path:
- '**/details_harness|hendrycksTest-college_biology|5_2023-10-12T17-36-24.431746.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-college_biology|5_2023-10-12T17-36-24.431746.parquet'
- config_name: harness_hendrycksTest_college_chemistry_5
data_files:
- split: 2023_10_12T17_36_24.431746
path:
- '**/details_harness|hendrycksTest-college_chemistry|5_2023-10-12T17-36-24.431746.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-college_chemistry|5_2023-10-12T17-36-24.431746.parquet'
- config_name: harness_hendrycksTest_college_computer_science_5
data_files:
- split: 2023_10_12T17_36_24.431746
path:
- '**/details_harness|hendrycksTest-college_computer_science|5_2023-10-12T17-36-24.431746.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-college_computer_science|5_2023-10-12T17-36-24.431746.parquet'
- config_name: harness_hendrycksTest_college_mathematics_5
data_files:
- split: 2023_10_12T17_36_24.431746
path:
- '**/details_harness|hendrycksTest-college_mathematics|5_2023-10-12T17-36-24.431746.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-college_mathematics|5_2023-10-12T17-36-24.431746.parquet'
- config_name: harness_hendrycksTest_college_medicine_5
data_files:
- split: 2023_10_12T17_36_24.431746
path:
- '**/details_harness|hendrycksTest-college_medicine|5_2023-10-12T17-36-24.431746.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-college_medicine|5_2023-10-12T17-36-24.431746.parquet'
- config_name: harness_hendrycksTest_college_physics_5
data_files:
- split: 2023_10_12T17_36_24.431746
path:
- '**/details_harness|hendrycksTest-college_physics|5_2023-10-12T17-36-24.431746.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-college_physics|5_2023-10-12T17-36-24.431746.parquet'
- config_name: harness_hendrycksTest_computer_security_5
data_files:
- split: 2023_10_12T17_36_24.431746
path:
- '**/details_harness|hendrycksTest-computer_security|5_2023-10-12T17-36-24.431746.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-computer_security|5_2023-10-12T17-36-24.431746.parquet'
- config_name: harness_hendrycksTest_conceptual_physics_5
data_files:
- split: 2023_10_12T17_36_24.431746
path:
- '**/details_harness|hendrycksTest-conceptual_physics|5_2023-10-12T17-36-24.431746.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-conceptual_physics|5_2023-10-12T17-36-24.431746.parquet'
- config_name: harness_hendrycksTest_econometrics_5
data_files:
- split: 2023_10_12T17_36_24.431746
path:
- '**/details_harness|hendrycksTest-econometrics|5_2023-10-12T17-36-24.431746.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-econometrics|5_2023-10-12T17-36-24.431746.parquet'
- config_name: harness_hendrycksTest_electrical_engineering_5
data_files:
- split: 2023_10_12T17_36_24.431746
path:
- '**/details_harness|hendrycksTest-electrical_engineering|5_2023-10-12T17-36-24.431746.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-electrical_engineering|5_2023-10-12T17-36-24.431746.parquet'
- config_name: harness_hendrycksTest_elementary_mathematics_5
data_files:
- split: 2023_10_12T17_36_24.431746
path:
- '**/details_harness|hendrycksTest-elementary_mathematics|5_2023-10-12T17-36-24.431746.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-elementary_mathematics|5_2023-10-12T17-36-24.431746.parquet'
- config_name: harness_hendrycksTest_formal_logic_5
data_files:
- split: 2023_10_12T17_36_24.431746
path:
- '**/details_harness|hendrycksTest-formal_logic|5_2023-10-12T17-36-24.431746.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-formal_logic|5_2023-10-12T17-36-24.431746.parquet'
- config_name: harness_hendrycksTest_global_facts_5
data_files:
- split: 2023_10_12T17_36_24.431746
path:
- '**/details_harness|hendrycksTest-global_facts|5_2023-10-12T17-36-24.431746.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-global_facts|5_2023-10-12T17-36-24.431746.parquet'
- config_name: harness_hendrycksTest_high_school_biology_5
data_files:
- split: 2023_10_12T17_36_24.431746
path:
- '**/details_harness|hendrycksTest-high_school_biology|5_2023-10-12T17-36-24.431746.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-high_school_biology|5_2023-10-12T17-36-24.431746.parquet'
- config_name: harness_hendrycksTest_high_school_chemistry_5
data_files:
- split: 2023_10_12T17_36_24.431746
path:
- '**/details_harness|hendrycksTest-high_school_chemistry|5_2023-10-12T17-36-24.431746.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-high_school_chemistry|5_2023-10-12T17-36-24.431746.parquet'
- config_name: harness_hendrycksTest_high_school_computer_science_5
data_files:
- split: 2023_10_12T17_36_24.431746
path:
- '**/details_harness|hendrycksTest-high_school_computer_science|5_2023-10-12T17-36-24.431746.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-high_school_computer_science|5_2023-10-12T17-36-24.431746.parquet'
- config_name: harness_hendrycksTest_high_school_european_history_5
data_files:
- split: 2023_10_12T17_36_24.431746
path:
- '**/details_harness|hendrycksTest-high_school_european_history|5_2023-10-12T17-36-24.431746.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-high_school_european_history|5_2023-10-12T17-36-24.431746.parquet'
- config_name: harness_hendrycksTest_high_school_geography_5
data_files:
- split: 2023_10_12T17_36_24.431746
path:
- '**/details_harness|hendrycksTest-high_school_geography|5_2023-10-12T17-36-24.431746.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-high_school_geography|5_2023-10-12T17-36-24.431746.parquet'
- config_name: harness_hendrycksTest_high_school_government_and_politics_5
data_files:
- split: 2023_10_12T17_36_24.431746
path:
- '**/details_harness|hendrycksTest-high_school_government_and_politics|5_2023-10-12T17-36-24.431746.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-high_school_government_and_politics|5_2023-10-12T17-36-24.431746.parquet'
- config_name: harness_hendrycksTest_high_school_macroeconomics_5
data_files:
- split: 2023_10_12T17_36_24.431746
path:
- '**/details_harness|hendrycksTest-high_school_macroeconomics|5_2023-10-12T17-36-24.431746.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-high_school_macroeconomics|5_2023-10-12T17-36-24.431746.parquet'
- config_name: harness_hendrycksTest_high_school_mathematics_5
data_files:
- split: 2023_10_12T17_36_24.431746
path:
- '**/details_harness|hendrycksTest-high_school_mathematics|5_2023-10-12T17-36-24.431746.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-high_school_mathematics|5_2023-10-12T17-36-24.431746.parquet'
- config_name: harness_hendrycksTest_high_school_microeconomics_5
data_files:
- split: 2023_10_12T17_36_24.431746
path:
- '**/details_harness|hendrycksTest-high_school_microeconomics|5_2023-10-12T17-36-24.431746.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-high_school_microeconomics|5_2023-10-12T17-36-24.431746.parquet'
- config_name: harness_hendrycksTest_high_school_physics_5
data_files:
- split: 2023_10_12T17_36_24.431746
path:
- '**/details_harness|hendrycksTest-high_school_physics|5_2023-10-12T17-36-24.431746.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-high_school_physics|5_2023-10-12T17-36-24.431746.parquet'
- config_name: harness_hendrycksTest_high_school_psychology_5
data_files:
- split: 2023_10_12T17_36_24.431746
path:
- '**/details_harness|hendrycksTest-high_school_psychology|5_2023-10-12T17-36-24.431746.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-high_school_psychology|5_2023-10-12T17-36-24.431746.parquet'
- config_name: harness_hendrycksTest_high_school_statistics_5
data_files:
- split: 2023_10_12T17_36_24.431746
path:
- '**/details_harness|hendrycksTest-high_school_statistics|5_2023-10-12T17-36-24.431746.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-high_school_statistics|5_2023-10-12T17-36-24.431746.parquet'
- config_name: harness_hendrycksTest_high_school_us_history_5
data_files:
- split: 2023_10_12T17_36_24.431746
path:
- '**/details_harness|hendrycksTest-high_school_us_history|5_2023-10-12T17-36-24.431746.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-high_school_us_history|5_2023-10-12T17-36-24.431746.parquet'
- config_name: harness_hendrycksTest_high_school_world_history_5
data_files:
- split: 2023_10_12T17_36_24.431746
path:
- '**/details_harness|hendrycksTest-high_school_world_history|5_2023-10-12T17-36-24.431746.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-high_school_world_history|5_2023-10-12T17-36-24.431746.parquet'
- config_name: harness_hendrycksTest_human_aging_5
data_files:
- split: 2023_10_12T17_36_24.431746
path:
- '**/details_harness|hendrycksTest-human_aging|5_2023-10-12T17-36-24.431746.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-human_aging|5_2023-10-12T17-36-24.431746.parquet'
- config_name: harness_hendrycksTest_human_sexuality_5
data_files:
- split: 2023_10_12T17_36_24.431746
path:
- '**/details_harness|hendrycksTest-human_sexuality|5_2023-10-12T17-36-24.431746.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-human_sexuality|5_2023-10-12T17-36-24.431746.parquet'
- config_name: harness_hendrycksTest_international_law_5
data_files:
- split: 2023_10_12T17_36_24.431746
path:
- '**/details_harness|hendrycksTest-international_law|5_2023-10-12T17-36-24.431746.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-international_law|5_2023-10-12T17-36-24.431746.parquet'
- config_name: harness_hendrycksTest_jurisprudence_5
data_files:
- split: 2023_10_12T17_36_24.431746
path:
- '**/details_harness|hendrycksTest-jurisprudence|5_2023-10-12T17-36-24.431746.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-jurisprudence|5_2023-10-12T17-36-24.431746.parquet'
- config_name: harness_hendrycksTest_logical_fallacies_5
data_files:
- split: 2023_10_12T17_36_24.431746
path:
- '**/details_harness|hendrycksTest-logical_fallacies|5_2023-10-12T17-36-24.431746.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-logical_fallacies|5_2023-10-12T17-36-24.431746.parquet'
- config_name: harness_hendrycksTest_machine_learning_5
data_files:
- split: 2023_10_12T17_36_24.431746
path:
- '**/details_harness|hendrycksTest-machine_learning|5_2023-10-12T17-36-24.431746.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-machine_learning|5_2023-10-12T17-36-24.431746.parquet'
- config_name: harness_hendrycksTest_management_5
data_files:
- split: 2023_10_12T17_36_24.431746
path:
- '**/details_harness|hendrycksTest-management|5_2023-10-12T17-36-24.431746.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-management|5_2023-10-12T17-36-24.431746.parquet'
- config_name: harness_hendrycksTest_marketing_5
data_files:
- split: 2023_10_12T17_36_24.431746
path:
- '**/details_harness|hendrycksTest-marketing|5_2023-10-12T17-36-24.431746.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-marketing|5_2023-10-12T17-36-24.431746.parquet'
- config_name: harness_hendrycksTest_medical_genetics_5
data_files:
- split: 2023_10_12T17_36_24.431746
path:
- '**/details_harness|hendrycksTest-medical_genetics|5_2023-10-12T17-36-24.431746.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-medical_genetics|5_2023-10-12T17-36-24.431746.parquet'
- config_name: harness_hendrycksTest_miscellaneous_5
data_files:
- split: 2023_10_12T17_36_24.431746
path:
- '**/details_harness|hendrycksTest-miscellaneous|5_2023-10-12T17-36-24.431746.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-miscellaneous|5_2023-10-12T17-36-24.431746.parquet'
- config_name: harness_hendrycksTest_moral_disputes_5
data_files:
- split: 2023_10_12T17_36_24.431746
path:
- '**/details_harness|hendrycksTest-moral_disputes|5_2023-10-12T17-36-24.431746.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-moral_disputes|5_2023-10-12T17-36-24.431746.parquet'
- config_name: harness_hendrycksTest_moral_scenarios_5
data_files:
- split: 2023_10_12T17_36_24.431746
path:
- '**/details_harness|hendrycksTest-moral_scenarios|5_2023-10-12T17-36-24.431746.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-moral_scenarios|5_2023-10-12T17-36-24.431746.parquet'
- config_name: harness_hendrycksTest_nutrition_5
data_files:
- split: 2023_10_12T17_36_24.431746
path:
- '**/details_harness|hendrycksTest-nutrition|5_2023-10-12T17-36-24.431746.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-nutrition|5_2023-10-12T17-36-24.431746.parquet'
- config_name: harness_hendrycksTest_philosophy_5
data_files:
- split: 2023_10_12T17_36_24.431746
path:
- '**/details_harness|hendrycksTest-philosophy|5_2023-10-12T17-36-24.431746.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-philosophy|5_2023-10-12T17-36-24.431746.parquet'
- config_name: harness_hendrycksTest_prehistory_5
data_files:
- split: 2023_10_12T17_36_24.431746
path:
- '**/details_harness|hendrycksTest-prehistory|5_2023-10-12T17-36-24.431746.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-prehistory|5_2023-10-12T17-36-24.431746.parquet'
- config_name: harness_hendrycksTest_professional_accounting_5
data_files:
- split: 2023_10_12T17_36_24.431746
path:
- '**/details_harness|hendrycksTest-professional_accounting|5_2023-10-12T17-36-24.431746.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-professional_accounting|5_2023-10-12T17-36-24.431746.parquet'
- config_name: harness_hendrycksTest_professional_law_5
data_files:
- split: 2023_10_12T17_36_24.431746
path:
- '**/details_harness|hendrycksTest-professional_law|5_2023-10-12T17-36-24.431746.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-professional_law|5_2023-10-12T17-36-24.431746.parquet'
- config_name: harness_hendrycksTest_professional_medicine_5
data_files:
- split: 2023_10_12T17_36_24.431746
path:
- '**/details_harness|hendrycksTest-professional_medicine|5_2023-10-12T17-36-24.431746.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-professional_medicine|5_2023-10-12T17-36-24.431746.parquet'
- config_name: harness_hendrycksTest_professional_psychology_5
data_files:
- split: 2023_10_12T17_36_24.431746
path:
- '**/details_harness|hendrycksTest-professional_psychology|5_2023-10-12T17-36-24.431746.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-professional_psychology|5_2023-10-12T17-36-24.431746.parquet'
- config_name: harness_hendrycksTest_public_relations_5
data_files:
- split: 2023_10_12T17_36_24.431746
path:
- '**/details_harness|hendrycksTest-public_relations|5_2023-10-12T17-36-24.431746.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-public_relations|5_2023-10-12T17-36-24.431746.parquet'
- config_name: harness_hendrycksTest_security_studies_5
data_files:
- split: 2023_10_12T17_36_24.431746
path:
- '**/details_harness|hendrycksTest-security_studies|5_2023-10-12T17-36-24.431746.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-security_studies|5_2023-10-12T17-36-24.431746.parquet'
- config_name: harness_hendrycksTest_sociology_5
data_files:
- split: 2023_10_12T17_36_24.431746
path:
- '**/details_harness|hendrycksTest-sociology|5_2023-10-12T17-36-24.431746.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-sociology|5_2023-10-12T17-36-24.431746.parquet'
- config_name: harness_hendrycksTest_us_foreign_policy_5
data_files:
- split: 2023_10_12T17_36_24.431746
path:
- '**/details_harness|hendrycksTest-us_foreign_policy|5_2023-10-12T17-36-24.431746.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-us_foreign_policy|5_2023-10-12T17-36-24.431746.parquet'
- config_name: harness_hendrycksTest_virology_5
data_files:
- split: 2023_10_12T17_36_24.431746
path:
- '**/details_harness|hendrycksTest-virology|5_2023-10-12T17-36-24.431746.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-virology|5_2023-10-12T17-36-24.431746.parquet'
- config_name: harness_hendrycksTest_world_religions_5
data_files:
- split: 2023_10_12T17_36_24.431746
path:
- '**/details_harness|hendrycksTest-world_religions|5_2023-10-12T17-36-24.431746.parquet'
- split: latest
path:
- '**/details_harness|hendrycksTest-world_religions|5_2023-10-12T17-36-24.431746.parquet'
- config_name: harness_truthfulqa_mc_0
data_files:
- split: 2023_10_12T17_36_24.431746
path:
- '**/details_harness|truthfulqa:mc|0_2023-10-12T17-36-24.431746.parquet'
- split: latest
path:
- '**/details_harness|truthfulqa:mc|0_2023-10-12T17-36-24.431746.parquet'
- config_name: harness_winogrande_5
data_files:
- split: 2023_10_26T00_11_50.324232
path:
- '**/details_harness|winogrande|5_2023-10-26T00-11-50.324232.parquet'
- split: latest
path:
- '**/details_harness|winogrande|5_2023-10-26T00-11-50.324232.parquet'
- config_name: results
data_files:
- split: 2023_10_12T17_36_24.431746
path:
- results_2023-10-12T17-36-24.431746.parquet
- split: 2023_10_26T00_11_50.324232
path:
- results_2023-10-26T00-11-50.324232.parquet
- split: latest
path:
- results_2023-10-26T00-11-50.324232.parquet
---
# Dataset Card for Evaluation run of Sao10K/Euryale-1.3-L2-70B
## Dataset Description
- **Homepage:**
- **Repository:** https://huggingface.co/Sao10K/Euryale-1.3-L2-70B
- **Paper:**
- **Leaderboard:** https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard
- **Point of Contact:** clementine@hf.co
### Dataset Summary
Dataset automatically created during the evaluation run of model [Sao10K/Euryale-1.3-L2-70B](https://huggingface.co/Sao10K/Euryale-1.3-L2-70B) on the [Open LLM Leaderboard](https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard).
The dataset is composed of 64 configuration, each one coresponding 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 additional configuration "results" store all the aggregated results of the run (and is used to compute and display the agregated metrics on the [Open LLM Leaderboard](https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard)).
To load the details from a run, you can for instance do the following:
```python
from datasets import load_dataset
data = load_dataset("open-llm-leaderboard/details_Sao10K__Euryale-1.3-L2-70B",
"harness_winogrande_5",
split="train")
```
## Latest results
These are the [latest results from run 2023-10-26T00:11:50.324232](https://huggingface.co/datasets/open-llm-leaderboard/details_Sao10K__Euryale-1.3-L2-70B/blob/main/results_2023-10-26T00-11-50.324232.json)(note that their might be results for other tasks in the repos if successive evals didn't cover the same tasks. You find each in the results and the "latest" split for each eval):
```python
{
"all": {
"em": 0.5388003355704698,
"em_stderr": 0.005105027329360947,
"f1": 0.6009920302013437,
"f1_stderr": 0.004740248039821831,
"acc": 0.5849328585370874,
"acc_stderr": 0.011836910620214903
},
"harness|drop|3": {
"em": 0.5388003355704698,
"em_stderr": 0.005105027329360947,
"f1": 0.6009920302013437,
"f1_stderr": 0.004740248039821831
},
"harness|gsm8k|5": {
"acc": 0.3419257012888552,
"acc_stderr": 0.013066089625182799
},
"harness|winogrande|5": {
"acc": 0.8279400157853196,
"acc_stderr": 0.010607731615247007
}
}
```
### Supported Tasks and Leaderboards
[More Information Needed]
### Languages
[More Information Needed]
## Dataset Structure
### Data Instances
[More Information Needed]
### Data Fields
[More Information Needed]
### Data Splits
[More Information Needed]
## Dataset Creation
### Curation Rationale
[More Information Needed]
### Source Data
#### Initial Data Collection and Normalization
[More Information Needed]
#### Who are the source language producers?
[More Information Needed]
### Annotations
#### Annotation process
[More Information Needed]
#### Who are the annotators?
[More Information Needed]
### Personal and Sensitive Information
[More Information Needed]
## Considerations for Using the Data
### Social Impact of Dataset
[More Information Needed]
### Discussion of Biases
[More Information Needed]
### Other Known Limitations
[More Information Needed]
## Additional Information
### Dataset Curators
[More Information Needed]
### Licensing Information
[More Information Needed]
### Citation Information
[More Information Needed]
### Contributions
[More Information Needed] | 38,602 | [
[
-0.0295257568359375,
-0.046966552734375,
0.01049041748046875,
0.021087646484375,
-0.00846099853515625,
0.00902557373046875,
-0.023223876953125,
-0.0179290771484375,
0.03570556640625,
0.0400390625,
-0.041015625,
-0.06341552734375,
-0.0496826171875,
0.01959228... |
GPTZERO/PERSUADE | 2023-10-12T18:06:06.000Z | [
"region:us"
] | GPTZERO | null | null | 0 | 0 | 2023-10-12T18:05:42 | Entry not found | 15 | [
[
-0.02142333984375,
-0.01495361328125,
0.05718994140625,
0.0288238525390625,
-0.035064697265625,
0.046539306640625,
0.052520751953125,
0.005062103271484375,
0.0513916015625,
0.016998291015625,
-0.052093505859375,
-0.014984130859375,
-0.060394287109375,
0.0379... |
V-Z/V-BOT | 2023-10-12T18:46:34.000Z | [
"region:us"
] | V-Z | null | null | 0 | 0 | 2023-10-12T18:46:34 | Entry not found | 15 | [
[
-0.0213775634765625,
-0.014984130859375,
0.05718994140625,
0.0288543701171875,
-0.0350341796875,
0.046478271484375,
0.052520751953125,
0.005062103271484375,
0.051361083984375,
0.016998291015625,
-0.0521240234375,
-0.01496124267578125,
-0.0604248046875,
0.037... |
open-llm-leaderboard/details_pe-nlp__llama-2-13b-vicuna-wizard | 2023-10-12T18:46:52.000Z | [
"region:us"
] | open-llm-leaderboard | null | null | 0 | 0 | 2023-10-12T18:46:43 | ---
pretty_name: Evaluation run of pe-nlp/llama-2-13b-vicuna-wizard
dataset_summary: "Dataset automatically created during the evaluation run of model\
\ [pe-nlp/llama-2-13b-vicuna-wizard](https://huggingface.co/pe-nlp/llama-2-13b-vicuna-wizard)\
\ on the [Open LLM Leaderboard](https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard).\n\
\nThe dataset is composed of 3 configuration, each one coresponding to one of the\
\ evaluated task.\n\nThe 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.\n\
\nAn additional configuration \"results\" store all the aggregated results of the\
\ run (and is used to compute and display the agregated metrics on the [Open LLM\
\ Leaderboard](https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard)).\n\
\nTo load the details from a run, you can for instance do the following:\n```python\n\
from datasets import load_dataset\ndata = load_dataset(\"open-llm-leaderboard/details_pe-nlp__llama-2-13b-vicuna-wizard\"\
,\n\t\"harness_winogrande_5\",\n\tsplit=\"train\")\n```\n\n## Latest results\n\n\
These are the [latest results from run 2023-10-12T18:46:39.910816](https://huggingface.co/datasets/open-llm-leaderboard/details_pe-nlp__llama-2-13b-vicuna-wizard/blob/main/results_2023-10-12T18-46-39.910816.json)(note\
\ that their might be results for other tasks in the repos if successive evals didn't\
\ cover the same tasks. You find each in the results and the \"latest\" split for\
\ each eval):\n\n```python\n{\n \"all\": {\n \"em\": 0.3976510067114094,\n\
\ \"em_stderr\": 0.0050120430065395205,\n \"f1\": 0.43937709731543745,\n\
\ \"f1_stderr\": 0.004888666829236633,\n \"acc\": 0.3794502424345056,\n\
\ \"acc_stderr\": 0.007394168076612409\n },\n \"harness|drop|3\": {\n\
\ \"em\": 0.3976510067114094,\n \"em_stderr\": 0.0050120430065395205,\n\
\ \"f1\": 0.43937709731543745,\n \"f1_stderr\": 0.004888666829236633\n\
\ },\n \"harness|gsm8k|5\": {\n \"acc\": 0.009097801364670205,\n \
\ \"acc_stderr\": 0.0026153265107756725\n },\n \"harness|winogrande|5\"\
: {\n \"acc\": 0.749802683504341,\n \"acc_stderr\": 0.012173009642449144\n\
\ }\n}\n```"
repo_url: https://huggingface.co/pe-nlp/llama-2-13b-vicuna-wizard
leaderboard_url: https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard
point_of_contact: clementine@hf.co
configs:
- config_name: harness_drop_3
data_files:
- split: 2023_10_12T18_46_39.910816
path:
- '**/details_harness|drop|3_2023-10-12T18-46-39.910816.parquet'
- split: latest
path:
- '**/details_harness|drop|3_2023-10-12T18-46-39.910816.parquet'
- config_name: harness_gsm8k_5
data_files:
- split: 2023_10_12T18_46_39.910816
path:
- '**/details_harness|gsm8k|5_2023-10-12T18-46-39.910816.parquet'
- split: latest
path:
- '**/details_harness|gsm8k|5_2023-10-12T18-46-39.910816.parquet'
- config_name: harness_winogrande_5
data_files:
- split: 2023_10_12T18_46_39.910816
path:
- '**/details_harness|winogrande|5_2023-10-12T18-46-39.910816.parquet'
- split: latest
path:
- '**/details_harness|winogrande|5_2023-10-12T18-46-39.910816.parquet'
- config_name: results
data_files:
- split: 2023_10_12T18_46_39.910816
path:
- results_2023-10-12T18-46-39.910816.parquet
- split: latest
path:
- results_2023-10-12T18-46-39.910816.parquet
---
# Dataset Card for Evaluation run of pe-nlp/llama-2-13b-vicuna-wizard
## Dataset Description
- **Homepage:**
- **Repository:** https://huggingface.co/pe-nlp/llama-2-13b-vicuna-wizard
- **Paper:**
- **Leaderboard:** https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard
- **Point of Contact:** clementine@hf.co
### Dataset Summary
Dataset automatically created during the evaluation run of model [pe-nlp/llama-2-13b-vicuna-wizard](https://huggingface.co/pe-nlp/llama-2-13b-vicuna-wizard) on the [Open LLM Leaderboard](https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard).
The dataset is composed of 3 configuration, each one coresponding 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 additional configuration "results" store all the aggregated results of the run (and is used to compute and display the agregated metrics on the [Open LLM Leaderboard](https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard)).
To load the details from a run, you can for instance do the following:
```python
from datasets import load_dataset
data = load_dataset("open-llm-leaderboard/details_pe-nlp__llama-2-13b-vicuna-wizard",
"harness_winogrande_5",
split="train")
```
## Latest results
These are the [latest results from run 2023-10-12T18:46:39.910816](https://huggingface.co/datasets/open-llm-leaderboard/details_pe-nlp__llama-2-13b-vicuna-wizard/blob/main/results_2023-10-12T18-46-39.910816.json)(note that their might be results for other tasks in the repos if successive evals didn't cover the same tasks. You find each in the results and the "latest" split for each eval):
```python
{
"all": {
"em": 0.3976510067114094,
"em_stderr": 0.0050120430065395205,
"f1": 0.43937709731543745,
"f1_stderr": 0.004888666829236633,
"acc": 0.3794502424345056,
"acc_stderr": 0.007394168076612409
},
"harness|drop|3": {
"em": 0.3976510067114094,
"em_stderr": 0.0050120430065395205,
"f1": 0.43937709731543745,
"f1_stderr": 0.004888666829236633
},
"harness|gsm8k|5": {
"acc": 0.009097801364670205,
"acc_stderr": 0.0026153265107756725
},
"harness|winogrande|5": {
"acc": 0.749802683504341,
"acc_stderr": 0.012173009642449144
}
}
```
### Supported Tasks and Leaderboards
[More Information Needed]
### Languages
[More Information Needed]
## Dataset Structure
### Data Instances
[More Information Needed]
### Data Fields
[More Information Needed]
### Data Splits
[More Information Needed]
## Dataset Creation
### Curation Rationale
[More Information Needed]
### Source Data
#### Initial Data Collection and Normalization
[More Information Needed]
#### Who are the source language producers?
[More Information Needed]
### Annotations
#### Annotation process
[More Information Needed]
#### Who are the annotators?
[More Information Needed]
### Personal and Sensitive Information
[More Information Needed]
## Considerations for Using the Data
### Social Impact of Dataset
[More Information Needed]
### Discussion of Biases
[More Information Needed]
### Other Known Limitations
[More Information Needed]
## Additional Information
### Dataset Curators
[More Information Needed]
### Licensing Information
[More Information Needed]
### Citation Information
[More Information Needed]
### Contributions
[More Information Needed] | 7,289 | [
[
-0.0261993408203125,
-0.05242919921875,
0.01617431640625,
0.025848388671875,
-0.00927734375,
0.0068206787109375,
-0.029083251953125,
-0.0166473388671875,
0.031951904296875,
0.043212890625,
-0.048095703125,
-0.06939697265625,
-0.049560546875,
0.02310180664062... |
pccl-org/formal-logic-simple-order-simple-objects-clavorier-500 | 2023-10-12T19:22:37.000Z | [
"region:us"
] | pccl-org | null | null | 0 | 0 | 2023-10-12T18:56:06 | ---
dataset_info:
features:
- name: greater_than
dtype: string
- name: less_than
dtype: string
- name: correct_example
sequence: string
- name: incorrect_example
sequence: string
- name: distance
dtype: int64
- name: index
dtype: int64
splits:
- name: train
num_bytes: 19386150
num_examples: 124750
download_size: 0
dataset_size: 19386150
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
---
# Dataset Card for "formal-logic-simple-order-simple-objects-clavorier-500"
[More Information needed](https://github.com/huggingface/datasets/blob/main/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards) | 691 | [
[
-0.034698486328125,
-0.0197906494140625,
0.023040771484375,
0.01139068603515625,
-0.0196380615234375,
-0.033050537109375,
0.0159454345703125,
-0.0203399658203125,
0.033721923828125,
0.039154052734375,
-0.0506591796875,
-0.055877685546875,
-0.044891357421875,
... |
ostapeno/platy_icl5_maxD50_maxC1000000_prmt00_1 | 2023-10-12T20:16:58.000Z | [
"region:us"
] | ostapeno | null | null | 0 | 0 | 2023-10-12T20:16:46 | ## model_setting_name: platy
## max_context_length: 512
## icl_examples: 5
## icl_dataset_name: lukaemon/mmlu
## max_documents_per_subject: 50
## max_contexts_per_subject: 1000000
## icl_use_out_options: True
## seed_dataset: sordonia/my-wiki-latex_mmlu_from_valid_all
## subjects: SUB_10
## response_template: 0
## inverse_template: 0
| 336 | [
[
-0.037078857421875,
-0.0241241455078125,
0.0255279541015625,
0.03460693359375,
-0.028594970703125,
-0.019561767578125,
-0.003643035888671875,
0.0164337158203125,
-0.00916290283203125,
0.031707763671875,
-0.0650634765625,
-0.03997802734375,
-0.026458740234375,
... |
ostapeno/platy_icl5_maxD50_maxC1000000_prmt01_1 | 2023-10-12T20:57:37.000Z | [
"region:us"
] | ostapeno | null | null | 0 | 0 | 2023-10-12T20:57:24 | ## model_setting_name: platy
## max_context_length: 512
## icl_examples: 5
## icl_dataset_name: lukaemon/mmlu
## max_documents_per_subject: 50
## max_contexts_per_subject: 1000000
## icl_use_out_options: True
## seed_dataset: sordonia/my-wiki-latex_mmlu_from_valid_all
## subjects: SUB_10
## response_template: 0
## inverse_template: 1
| 336 | [
[
-0.037689208984375,
-0.0238800048828125,
0.02545166015625,
0.03460693359375,
-0.02874755859375,
-0.0206756591796875,
-0.00374603271484375,
0.0159454345703125,
-0.0091705322265625,
0.0311279296875,
-0.065185546875,
-0.03948974609375,
-0.026153564453125,
0.019... |
ostapeno/platy_icl5_subset1.0_maxD1000000_3 | 2023-10-12T21:07:18.000Z | [
"region:us"
] | ostapeno | null | null | 0 | 0 | 2023-10-12T21:07:03 | ## model_setting_name: platy
## max_context_length: 512
## subset: 1.0
## icl_examples: 5
## icl_dataset_name: lukaemon/mmlu
## max_documents_per_subject: 1000000
## icl_use_out_options: True
## seed_dataset: sordonia/my-wiki-latex_mmlu_from_valid_all
## subjects: SUB_10
| 272 | [
[
-0.036376953125,
-0.02606201171875,
0.0289154052734375,
0.03533935546875,
-0.03289794921875,
-0.0164642333984375,
-0.002685546875,
0.01611328125,
-0.007537841796875,
0.0357666015625,
-0.06488037109375,
-0.041656494140625,
-0.030120849609375,
0.02348327636718... |
Maruti-IO/Generate_SQL | 2023-10-13T02:26:15.000Z | [
"license:apache-2.0",
"region:us"
] | Maruti-IO | null | null | 0 | 0 | 2023-10-12T21:20:03 | ---
license: apache-2.0
---
## Created By Maruti.io
This dataset prompts to create SQL Queries, and completions to those prompts. | 130 | [
[
-0.023406982421875,
-0.018890380859375,
0.0259246826171875,
0.033905029296875,
-0.0124053955078125,
0.00008243322372436523,
0.0217132568359375,
0.0174560546875,
0.046630859375,
0.04791259765625,
-0.09600830078125,
-0.02777099609375,
0.0057373046875,
-0.00032... |
ostapeno/platy_icl5_maxD50_maxC1000000_prmt10_1 | 2023-10-12T21:21:55.000Z | [
"region:us"
] | ostapeno | null | null | 0 | 0 | 2023-10-12T21:21:42 | ## model_setting_name: platy
## max_context_length: 512
## icl_examples: 5
## icl_dataset_name: lukaemon/mmlu
## max_documents_per_subject: 50
## max_contexts_per_subject: 1000000
## icl_use_out_options: True
## seed_dataset: sordonia/my-wiki-latex_mmlu_from_valid_all
## subjects: SUB_10
## response_template: 1
## inverse_template: 0
| 336 | [
[
-0.037628173828125,
-0.02398681640625,
0.0255126953125,
0.0347900390625,
-0.0284576416015625,
-0.020751953125,
-0.0034656524658203125,
0.01617431640625,
-0.0097503662109375,
0.0311279296875,
-0.064697265625,
-0.039459228515625,
-0.02642822265625,
0.019104003... |
Hani89/Medical_ASR_45HRs | 2023-10-13T01:38:01.000Z | [
"task_categories:automatic-speech-recognition",
"size_categories:10K<n<100K",
"language:en",
"license:apache-2.0",
"medical",
"region:us"
] | Hani89 | null | null | 0 | 0 | 2023-10-12T21:22:48 | ---
license: apache-2.0
task_categories:
- automatic-speech-recognition
language:
- en
tags:
- medical
size_categories:
- 10K<n<100K
---
# Medical Dataset for ASR
The dataset is a part taken from [The MedDialog dataset](https://huggingface.co/datasets/medical_dialog). We used only icliniq_dialogue.txt and done some preprocessing:
- Remove all chars except for [a-z|A-Z|0-9|,|.].
- Break each conversation into rows of 32 to 35 words.
- Remove Duplication.
- Fix typos using GPT-3 instructons' model.
- Used Suno/Bark to create ~15K audio clips with different voices [*In Progress*]
#### Note:
- We are expecting about ~45 hours of medical audio clips.
- The dataset will be released soon, for any inqueries please contact me on(hmthubaiti@uqu.edu.sa) | 755 | [
[
-0.0110321044921875,
-0.0294342041015625,
0.0154266357421875,
0.0155181884765625,
-0.03387451171875,
0.0119781494140625,
-0.004364013671875,
-0.029052734375,
0.0140533447265625,
0.060028076171875,
-0.0577392578125,
-0.03289794921875,
-0.0288238525390625,
0.0... |
ostapeno/platy_icl5_maxD50_maxC1000000_prmt11_1 | 2023-10-12T21:39:55.000Z | [
"region:us"
] | ostapeno | null | null | 0 | 0 | 2023-10-12T21:39:42 | ## model_setting_name: platy
## max_context_length: 512
## icl_examples: 5
## icl_dataset_name: lukaemon/mmlu
## max_documents_per_subject: 50
## max_contexts_per_subject: 1000000
## icl_use_out_options: True
## seed_dataset: sordonia/my-wiki-latex_mmlu_from_valid_all
## subjects: SUB_10
## response_template: 1
## inverse_template: 1
| 336 | [
[
-0.03802490234375,
-0.02410888671875,
0.0254974365234375,
0.03515625,
-0.02899169921875,
-0.0215911865234375,
-0.0037384033203125,
0.0162200927734375,
-0.0105133056640625,
0.031494140625,
-0.064697265625,
-0.038848876953125,
-0.0262908935546875,
0.0194244384... |
ostapeno/platy_icl5_maxD50_maxC1000000_prmt20_1 | 2023-10-12T22:06:49.000Z | [
"region:us"
] | ostapeno | null | null | 0 | 0 | 2023-10-12T21:52:21 | ## model_setting_name: platy
## max_context_length: 512
## icl_examples: 5
## icl_dataset_name: lukaemon/mmlu
## max_documents_per_subject: 50
## max_contexts_per_subject: 1000000
## icl_use_out_options: True
## seed_dataset: sordonia/my-wiki-latex_mmlu_from_valid_all
## subjects: SUB_10
## response_template: 2
## inverse_template: 0
| 336 | [
[
-0.037322998046875,
-0.0240936279296875,
0.025543212890625,
0.035064697265625,
-0.0287322998046875,
-0.0206451416015625,
-0.0037822723388671875,
0.01548004150390625,
-0.01015472412109375,
0.030975341796875,
-0.06475830078125,
-0.03900146484375,
-0.026123046875,
... |
bleugreen/scraped | 2023-10-12T21:59:23.000Z | [
"region:us"
] | bleugreen | null | null | 0 | 0 | 2023-10-12T21:59:23 | Entry not found | 15 | [
[
-0.02142333984375,
-0.01495361328125,
0.05718994140625,
0.0288238525390625,
-0.035064697265625,
0.046539306640625,
0.052520751953125,
0.005062103271484375,
0.0513916015625,
0.016998291015625,
-0.052093505859375,
-0.014984130859375,
-0.060394287109375,
0.0379... |
ostapeno/platy_icl5_maxD50_maxC1000000_prmt21_1 | 2023-10-12T22:08:01.000Z | [
"region:us"
] | ostapeno | null | null | 0 | 0 | 2023-10-12T22:07:48 | ## model_setting_name: platy
## max_context_length: 512
## icl_examples: 5
## icl_dataset_name: lukaemon/mmlu
## max_documents_per_subject: 50
## max_contexts_per_subject: 1000000
## icl_use_out_options: True
## seed_dataset: sordonia/my-wiki-latex_mmlu_from_valid_all
## subjects: SUB_10
## response_template: 2
## inverse_template: 1
| 336 | [
[
-0.038116455078125,
-0.024444580078125,
0.0254669189453125,
0.03533935546875,
-0.0290985107421875,
-0.021942138671875,
-0.003997802734375,
0.01558685302734375,
-0.0109100341796875,
0.0313720703125,
-0.06488037109375,
-0.038360595703125,
-0.0258636474609375,
... |
open-llm-leaderboard/details_WizardLM__WizardMath-13B-V1.0 | 2023-10-12T22:46:05.000Z | [
"region:us"
] | open-llm-leaderboard | null | null | 0 | 0 | 2023-10-12T22:45:57 | ---
pretty_name: Evaluation run of WizardLM/WizardMath-13B-V1.0
dataset_summary: "Dataset automatically created during the evaluation run of model\
\ [WizardLM/WizardMath-13B-V1.0](https://huggingface.co/WizardLM/WizardMath-13B-V1.0)\
\ on the [Open LLM Leaderboard](https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard).\n\
\nThe dataset is composed of 3 configuration, each one coresponding to one of the\
\ evaluated task.\n\nThe 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.\n\
\nAn additional configuration \"results\" store all the aggregated results of the\
\ run (and is used to compute and display the agregated metrics on the [Open LLM\
\ Leaderboard](https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard)).\n\
\nTo load the details from a run, you can for instance do the following:\n```python\n\
from datasets import load_dataset\ndata = load_dataset(\"open-llm-leaderboard/details_WizardLM__WizardMath-13B-V1.0\"\
,\n\t\"harness_winogrande_5\",\n\tsplit=\"train\")\n```\n\n## Latest results\n\n\
These are the [latest results from run 2023-10-12T22:45:52.861079](https://huggingface.co/datasets/open-llm-leaderboard/details_WizardLM__WizardMath-13B-V1.0/blob/main/results_2023-10-12T22-45-52.861079.json)(note\
\ that their might be results for other tasks in the repos if successive evals didn't\
\ cover the same tasks. You find each in the results and the \"latest\" split for\
\ each eval):\n\n```python\n{\n \"all\": {\n \"em\": 0.0024119127516778523,\n\
\ \"em_stderr\": 0.0005023380498893313,\n \"f1\": 0.07075817953020154,\n\
\ \"f1_stderr\": 0.0015254513833319102,\n \"acc\": 0.4212998893591507,\n\
\ \"acc_stderr\": 0.010848795701326375\n },\n \"harness|drop|3\": {\n\
\ \"em\": 0.0024119127516778523,\n \"em_stderr\": 0.0005023380498893313,\n\
\ \"f1\": 0.07075817953020154,\n \"f1_stderr\": 0.0015254513833319102\n\
\ },\n \"harness|gsm8k|5\": {\n \"acc\": 0.12357846853677028,\n \
\ \"acc_stderr\": 0.009065050306776925\n },\n \"harness|winogrande|5\"\
: {\n \"acc\": 0.7190213101815311,\n \"acc_stderr\": 0.012632541095875825\n\
\ }\n}\n```"
repo_url: https://huggingface.co/WizardLM/WizardMath-13B-V1.0
leaderboard_url: https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard
point_of_contact: clementine@hf.co
configs:
- config_name: harness_drop_3
data_files:
- split: 2023_10_12T22_45_52.861079
path:
- '**/details_harness|drop|3_2023-10-12T22-45-52.861079.parquet'
- split: latest
path:
- '**/details_harness|drop|3_2023-10-12T22-45-52.861079.parquet'
- config_name: harness_gsm8k_5
data_files:
- split: 2023_10_12T22_45_52.861079
path:
- '**/details_harness|gsm8k|5_2023-10-12T22-45-52.861079.parquet'
- split: latest
path:
- '**/details_harness|gsm8k|5_2023-10-12T22-45-52.861079.parquet'
- config_name: harness_winogrande_5
data_files:
- split: 2023_10_12T22_45_52.861079
path:
- '**/details_harness|winogrande|5_2023-10-12T22-45-52.861079.parquet'
- split: latest
path:
- '**/details_harness|winogrande|5_2023-10-12T22-45-52.861079.parquet'
- config_name: results
data_files:
- split: 2023_10_12T22_45_52.861079
path:
- results_2023-10-12T22-45-52.861079.parquet
- split: latest
path:
- results_2023-10-12T22-45-52.861079.parquet
---
# Dataset Card for Evaluation run of WizardLM/WizardMath-13B-V1.0
## Dataset Description
- **Homepage:**
- **Repository:** https://huggingface.co/WizardLM/WizardMath-13B-V1.0
- **Paper:**
- **Leaderboard:** https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard
- **Point of Contact:** clementine@hf.co
### Dataset Summary
Dataset automatically created during the evaluation run of model [WizardLM/WizardMath-13B-V1.0](https://huggingface.co/WizardLM/WizardMath-13B-V1.0) on the [Open LLM Leaderboard](https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard).
The dataset is composed of 3 configuration, each one coresponding 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 additional configuration "results" store all the aggregated results of the run (and is used to compute and display the agregated metrics on the [Open LLM Leaderboard](https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard)).
To load the details from a run, you can for instance do the following:
```python
from datasets import load_dataset
data = load_dataset("open-llm-leaderboard/details_WizardLM__WizardMath-13B-V1.0",
"harness_winogrande_5",
split="train")
```
## Latest results
These are the [latest results from run 2023-10-12T22:45:52.861079](https://huggingface.co/datasets/open-llm-leaderboard/details_WizardLM__WizardMath-13B-V1.0/blob/main/results_2023-10-12T22-45-52.861079.json)(note that their might be results for other tasks in the repos if successive evals didn't cover the same tasks. You find each in the results and the "latest" split for each eval):
```python
{
"all": {
"em": 0.0024119127516778523,
"em_stderr": 0.0005023380498893313,
"f1": 0.07075817953020154,
"f1_stderr": 0.0015254513833319102,
"acc": 0.4212998893591507,
"acc_stderr": 0.010848795701326375
},
"harness|drop|3": {
"em": 0.0024119127516778523,
"em_stderr": 0.0005023380498893313,
"f1": 0.07075817953020154,
"f1_stderr": 0.0015254513833319102
},
"harness|gsm8k|5": {
"acc": 0.12357846853677028,
"acc_stderr": 0.009065050306776925
},
"harness|winogrande|5": {
"acc": 0.7190213101815311,
"acc_stderr": 0.012632541095875825
}
}
```
### Supported Tasks and Leaderboards
[More Information Needed]
### Languages
[More Information Needed]
## Dataset Structure
### Data Instances
[More Information Needed]
### Data Fields
[More Information Needed]
### Data Splits
[More Information Needed]
## Dataset Creation
### Curation Rationale
[More Information Needed]
### Source Data
#### Initial Data Collection and Normalization
[More Information Needed]
#### Who are the source language producers?
[More Information Needed]
### Annotations
#### Annotation process
[More Information Needed]
#### Who are the annotators?
[More Information Needed]
### Personal and Sensitive Information
[More Information Needed]
## Considerations for Using the Data
### Social Impact of Dataset
[More Information Needed]
### Discussion of Biases
[More Information Needed]
### Other Known Limitations
[More Information Needed]
## Additional Information
### Dataset Curators
[More Information Needed]
### Licensing Information
[More Information Needed]
### Citation Information
[More Information Needed]
### Contributions
[More Information Needed] | 7,255 | [
[
-0.03399658203125,
-0.049560546875,
0.0068817138671875,
0.020904541015625,
-0.0011806488037109375,
0.00691986083984375,
-0.018951416015625,
-0.009552001953125,
0.0243682861328125,
0.042938232421875,
-0.058929443359375,
-0.06536865234375,
-0.047271728515625,
... |
pharaouk/biology_dataset_standardized_unified | 2023-10-13T21:20:02.000Z | [
"region:us"
] | pharaouk | null | null | 0 | 0 | 2023-10-12T23:48:34 | ---
dataset_info:
features:
- name: text
dtype: string
- name: conversation_id
dtype: int64
splits:
- name: train
num_bytes: 59401701
num_examples: 19999
download_size: 0
dataset_size: 59401701
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
---
# Dataset Card for "biology_dataset_standardized_unified"
[More Information needed](https://github.com/huggingface/datasets/blob/main/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards) | 504 | [
[
-0.021148681640625,
-0.006404876708984375,
0.012847900390625,
0.006549835205078125,
-0.0292510986328125,
-0.0008254051208496094,
0.017059326171875,
-0.0157318115234375,
0.06494140625,
0.029754638671875,
-0.0379638671875,
-0.06280517578125,
-0.034393310546875,
... |
pharaouk/biology_dataset_standardized_embedded | 2023-10-13T21:20:50.000Z | [
"region:us"
] | pharaouk | null | null | 0 | 0 | 2023-10-12T23:59:50 | ---
dataset_info:
features:
- name: text
dtype: string
- name: conversation_id
dtype: int64
- name: embedding
sequence: float32
splits:
- name: train
num_bytes: 141397601
num_examples: 19999
download_size: 0
dataset_size: 141397601
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
---
# Dataset Card for "biology_dataset_standardized_embedded"
[More Information needed](https://github.com/huggingface/datasets/blob/main/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards) | 549 | [
[
-0.0244140625,
-0.0268096923828125,
0.019744873046875,
0.008880615234375,
-0.033172607421875,
0.00713348388671875,
0.016632080078125,
-0.02032470703125,
0.06964111328125,
0.016387939453125,
-0.0360107421875,
-0.0675048828125,
-0.041046142578125,
-0.002647399... |
ZelaAI/lj_speech_encodec_packed | 2023-10-13T00:21:44.000Z | [
"region:us"
] | ZelaAI | null | null | 0 | 0 | 2023-10-13T00:19:21 | Entry not found | 15 | [
[
-0.0213775634765625,
-0.01497650146484375,
0.05718994140625,
0.02880859375,
-0.0350341796875,
0.046478271484375,
0.052490234375,
0.00507354736328125,
0.051361083984375,
0.0170135498046875,
-0.052093505859375,
-0.01497650146484375,
-0.0604248046875,
0.0379028... |
open-llm-leaderboard/details_bigscience__bloomz-7b1-mt | 2023-10-13T00:23:26.000Z | [
"region:us"
] | open-llm-leaderboard | null | null | 0 | 0 | 2023-10-13T00:23:18 | ---
pretty_name: Evaluation run of bigscience/bloomz-7b1-mt
dataset_summary: "Dataset automatically created during the evaluation run of model\
\ [bigscience/bloomz-7b1-mt](https://huggingface.co/bigscience/bloomz-7b1-mt) on\
\ the [Open LLM Leaderboard](https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard).\n\
\nThe dataset is composed of 3 configuration, each one coresponding to one of the\
\ evaluated task.\n\nThe 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.\n\
\nAn additional configuration \"results\" store all the aggregated results of the\
\ run (and is used to compute and display the agregated metrics on the [Open LLM\
\ Leaderboard](https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard)).\n\
\nTo load the details from a run, you can for instance do the following:\n```python\n\
from datasets import load_dataset\ndata = load_dataset(\"open-llm-leaderboard/details_bigscience__bloomz-7b1-mt\"\
,\n\t\"harness_winogrande_5\",\n\tsplit=\"train\")\n```\n\n## Latest results\n\n\
These are the [latest results from run 2023-10-13T00:23:14.934221](https://huggingface.co/datasets/open-llm-leaderboard/details_bigscience__bloomz-7b1-mt/blob/main/results_2023-10-13T00-23-14.934221.json)(note\
\ that their might be results for other tasks in the repos if successive evals didn't\
\ cover the same tasks. You find each in the results and the \"latest\" split for\
\ each eval):\n\n```python\n{\n \"all\": {\n \"em\": 0.23668204697986578,\n\
\ \"em_stderr\": 0.004352863496663657,\n \"f1\": 0.26898280201342256,\n\
\ \"f1_stderr\": 0.004366632346809137,\n \"acc\": 0.31531176006314127,\n\
\ \"acc_stderr\": 0.006782235298026759\n },\n \"harness|drop|3\": {\n\
\ \"em\": 0.23668204697986578,\n \"em_stderr\": 0.004352863496663657,\n\
\ \"f1\": 0.26898280201342256,\n \"f1_stderr\": 0.004366632346809137\n\
\ },\n \"harness|gsm8k|5\": {\n \"acc\": 0.0,\n \"acc_stderr\"\
: 0.0\n },\n \"harness|winogrande|5\": {\n \"acc\": 0.6306235201262825,\n\
\ \"acc_stderr\": 0.013564470596053518\n }\n}\n```"
repo_url: https://huggingface.co/bigscience/bloomz-7b1-mt
leaderboard_url: https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard
point_of_contact: clementine@hf.co
configs:
- config_name: harness_drop_3
data_files:
- split: 2023_10_13T00_23_14.934221
path:
- '**/details_harness|drop|3_2023-10-13T00-23-14.934221.parquet'
- split: latest
path:
- '**/details_harness|drop|3_2023-10-13T00-23-14.934221.parquet'
- config_name: harness_gsm8k_5
data_files:
- split: 2023_10_13T00_23_14.934221
path:
- '**/details_harness|gsm8k|5_2023-10-13T00-23-14.934221.parquet'
- split: latest
path:
- '**/details_harness|gsm8k|5_2023-10-13T00-23-14.934221.parquet'
- config_name: harness_winogrande_5
data_files:
- split: 2023_10_13T00_23_14.934221
path:
- '**/details_harness|winogrande|5_2023-10-13T00-23-14.934221.parquet'
- split: latest
path:
- '**/details_harness|winogrande|5_2023-10-13T00-23-14.934221.parquet'
- config_name: results
data_files:
- split: 2023_10_13T00_23_14.934221
path:
- results_2023-10-13T00-23-14.934221.parquet
- split: latest
path:
- results_2023-10-13T00-23-14.934221.parquet
---
# Dataset Card for Evaluation run of bigscience/bloomz-7b1-mt
## Dataset Description
- **Homepage:**
- **Repository:** https://huggingface.co/bigscience/bloomz-7b1-mt
- **Paper:**
- **Leaderboard:** https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard
- **Point of Contact:** clementine@hf.co
### Dataset Summary
Dataset automatically created during the evaluation run of model [bigscience/bloomz-7b1-mt](https://huggingface.co/bigscience/bloomz-7b1-mt) on the [Open LLM Leaderboard](https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard).
The dataset is composed of 3 configuration, each one coresponding 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 additional configuration "results" store all the aggregated results of the run (and is used to compute and display the agregated metrics on the [Open LLM Leaderboard](https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard)).
To load the details from a run, you can for instance do the following:
```python
from datasets import load_dataset
data = load_dataset("open-llm-leaderboard/details_bigscience__bloomz-7b1-mt",
"harness_winogrande_5",
split="train")
```
## Latest results
These are the [latest results from run 2023-10-13T00:23:14.934221](https://huggingface.co/datasets/open-llm-leaderboard/details_bigscience__bloomz-7b1-mt/blob/main/results_2023-10-13T00-23-14.934221.json)(note that their might be results for other tasks in the repos if successive evals didn't cover the same tasks. You find each in the results and the "latest" split for each eval):
```python
{
"all": {
"em": 0.23668204697986578,
"em_stderr": 0.004352863496663657,
"f1": 0.26898280201342256,
"f1_stderr": 0.004366632346809137,
"acc": 0.31531176006314127,
"acc_stderr": 0.006782235298026759
},
"harness|drop|3": {
"em": 0.23668204697986578,
"em_stderr": 0.004352863496663657,
"f1": 0.26898280201342256,
"f1_stderr": 0.004366632346809137
},
"harness|gsm8k|5": {
"acc": 0.0,
"acc_stderr": 0.0
},
"harness|winogrande|5": {
"acc": 0.6306235201262825,
"acc_stderr": 0.013564470596053518
}
}
```
### Supported Tasks and Leaderboards
[More Information Needed]
### Languages
[More Information Needed]
## Dataset Structure
### Data Instances
[More Information Needed]
### Data Fields
[More Information Needed]
### Data Splits
[More Information Needed]
## Dataset Creation
### Curation Rationale
[More Information Needed]
### Source Data
#### Initial Data Collection and Normalization
[More Information Needed]
#### Who are the source language producers?
[More Information Needed]
### Annotations
#### Annotation process
[More Information Needed]
#### Who are the annotators?
[More Information Needed]
### Personal and Sensitive Information
[More Information Needed]
## Considerations for Using the Data
### Social Impact of Dataset
[More Information Needed]
### Discussion of Biases
[More Information Needed]
### Other Known Limitations
[More Information Needed]
## Additional Information
### Dataset Curators
[More Information Needed]
### Licensing Information
[More Information Needed]
### Citation Information
[More Information Needed]
### Contributions
[More Information Needed] | 7,122 | [
[
-0.0299072265625,
-0.046966552734375,
0.024169921875,
0.0228729248046875,
-0.0072021484375,
0.007282257080078125,
-0.0340576171875,
-0.0108489990234375,
0.030059814453125,
0.0360107421875,
-0.05584716796875,
-0.0750732421875,
-0.0482177734375,
0.013153076171... |
nekofura/Comis | 2023-10-13T06:36:42.000Z | [
"region:us"
] | nekofura | null | null | 0 | 0 | 2023-10-13T00:47:15 | Entry not found | 15 | [
[
-0.02142333984375,
-0.01495361328125,
0.05718994140625,
0.0288238525390625,
-0.035064697265625,
0.046539306640625,
0.052520751953125,
0.005062103271484375,
0.0513916015625,
0.016998291015625,
-0.052093505859375,
-0.014984130859375,
-0.060394287109375,
0.0379... |
nlplabtdtu/biosses-sts-vi | 2023-10-13T00:55:01.000Z | [
"region:us"
] | nlplabtdtu | null | null | 0 | 0 | 2023-10-13T00:54:49 | Entry not found | 15 | [
[
-0.02142333984375,
-0.01495361328125,
0.05718994140625,
0.0288238525390625,
-0.035064697265625,
0.046539306640625,
0.052520751953125,
0.005062103271484375,
0.0513916015625,
0.016998291015625,
-0.052093505859375,
-0.014984130859375,
-0.060394287109375,
0.0379... |
nlplabtdtu/sts12-vi | 2023-10-13T00:57:42.000Z | [
"region:us"
] | nlplabtdtu | null | null | 0 | 0 | 2023-10-13T00:57:28 | Entry not found | 15 | [
[
-0.02142333984375,
-0.01495361328125,
0.05718994140625,
0.0288238525390625,
-0.035064697265625,
0.046539306640625,
0.052520751953125,
0.005062103271484375,
0.0513916015625,
0.016998291015625,
-0.052093505859375,
-0.014984130859375,
-0.060394287109375,
0.0379... |
20GaugeDeer/AlvaAnyaAI | 2023-10-13T00:57:29.000Z | [
"region:us"
] | 20GaugeDeer | null | null | 0 | 0 | 2023-10-13T00:57:29 | Entry not found | 15 | [
[
-0.02142333984375,
-0.01495361328125,
0.05718994140625,
0.0288238525390625,
-0.035064697265625,
0.046539306640625,
0.052520751953125,
0.005062103271484375,
0.0513916015625,
0.016998291015625,
-0.052093505859375,
-0.014984130859375,
-0.060394287109375,
0.0379... |
nlplabtdtu/sts13-vi | 2023-10-13T00:58:35.000Z | [
"region:us"
] | nlplabtdtu | null | null | 0 | 0 | 2023-10-13T00:57:57 | Entry not found | 15 | [
[
-0.02142333984375,
-0.01495361328125,
0.05718994140625,
0.0288238525390625,
-0.035064697265625,
0.046539306640625,
0.052520751953125,
0.005062103271484375,
0.0513916015625,
0.016998291015625,
-0.052093505859375,
-0.014984130859375,
-0.060394287109375,
0.0379... |
SKAInet98/infer | 2023-10-13T01:41:11.000Z | [
"region:us"
] | SKAInet98 | null | null | 0 | 0 | 2023-10-13T01:38:11 | Entry not found | 15 | [
[
-0.02142333984375,
-0.01495361328125,
0.05718994140625,
0.0288238525390625,
-0.035064697265625,
0.046539306640625,
0.052520751953125,
0.005062103271484375,
0.0513916015625,
0.016998291015625,
-0.052093505859375,
-0.014984130859375,
-0.060394287109375,
0.0379... |
RitikkumarS/Dreambooth1 | 2023-10-13T02:18:43.000Z | [
"region:us"
] | RitikkumarS | null | null | 0 | 0 | 2023-10-13T01:56:04 | Entry not found | 15 | [
[
-0.02142333984375,
-0.01495361328125,
0.05718994140625,
0.0288238525390625,
-0.035064697265625,
0.046539306640625,
0.052520751953125,
0.005062103271484375,
0.0513916015625,
0.016998291015625,
-0.052093505859375,
-0.014984130859375,
-0.060394287109375,
0.0379... |
realCyberWEI/anime_dataset_100k | 2023-10-13T02:39:03.000Z | [
"region:us"
] | realCyberWEI | null | null | 0 | 0 | 2023-10-13T02:11:10 | Entry not found | 15 | [
[
-0.02142333984375,
-0.01495361328125,
0.05718994140625,
0.0288238525390625,
-0.035064697265625,
0.046539306640625,
0.052520751953125,
0.005062103271484375,
0.0513916015625,
0.016998291015625,
-0.052093505859375,
-0.014984130859375,
-0.060394287109375,
0.0379... |
pharaouk/biology_dataset_standardized_cluster_0 | 2023-10-13T21:21:10.000Z | [
"region:us"
] | pharaouk | null | null | 0 | 0 | 2023-10-13T02:13:44 | ---
dataset_info:
features: []
splits:
- name: train
num_bytes: 0
num_examples: 0
download_size: 0
dataset_size: 0
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
---
# Dataset Card for "biology_dataset_standardized_cluster_0"
[More Information needed](https://github.com/huggingface/datasets/blob/main/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards) | 415 | [
[
-0.0234832763671875,
-0.0196380615234375,
0.0301971435546875,
0.01151275634765625,
-0.0231475830078125,
0.0003428459167480469,
0.0190887451171875,
-0.011688232421875,
0.0799560546875,
0.0184326171875,
-0.04241943359375,
-0.0745849609375,
-0.04022216796875,
-... |
pharaouk/biology_dataset_standardized_cluster_1 | 2023-10-13T21:24:28.000Z | [
"region:us"
] | pharaouk | null | null | 0 | 0 | 2023-10-13T02:13:53 | ---
dataset_info:
features: []
splits:
- name: train
num_bytes: 0
num_examples: 0
download_size: 0
dataset_size: 0
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
---
# Dataset Card for "biology_dataset_standardized_cluster_1"
[More Information needed](https://github.com/huggingface/datasets/blob/main/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards) | 415 | [
[
-0.0272216796875,
-0.021392822265625,
0.0213470458984375,
0.0192718505859375,
-0.0267791748046875,
-0.0025577545166015625,
0.0204315185546875,
-0.0058746337890625,
0.07611083984375,
0.0189666748046875,
-0.04974365234375,
-0.0743408203125,
-0.04473876953125,
... |
pharaouk/biology_dataset_standardized_cluster_2 | 2023-10-13T02:14:04.000Z | [
"region:us"
] | pharaouk | null | null | 0 | 0 | 2023-10-13T02:14:02 | ---
dataset_info:
features: []
splits:
- name: train
num_bytes: 0
num_examples: 0
download_size: 324
dataset_size: 0
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
---
# Dataset Card for "biology_dataset_standardized_cluster_2"
[More Information needed](https://github.com/huggingface/datasets/blob/main/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards) | 417 | [
[
-0.0151214599609375,
-0.01800537109375,
0.0242767333984375,
0.017974853515625,
-0.0264129638671875,
-0.00067901611328125,
0.0186920166015625,
-0.019317626953125,
0.06048583984375,
0.0159912109375,
-0.038177490234375,
-0.05938720703125,
-0.0479736328125,
-0.0... |
pharaouk/biology_dataset_standardized_cluster_3 | 2023-10-13T02:14:14.000Z | [
"region:us"
] | pharaouk | null | null | 0 | 0 | 2023-10-13T02:14:12 | ---
dataset_info:
features: []
splits:
- name: train
num_bytes: 0
num_examples: 0
download_size: 324
dataset_size: 0
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
---
# Dataset Card for "biology_dataset_standardized_cluster_3"
[More Information needed](https://github.com/huggingface/datasets/blob/main/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards) | 417 | [
[
-0.0237579345703125,
-0.0160064697265625,
0.03765869140625,
0.021026611328125,
-0.0199127197265625,
-0.0036640167236328125,
0.0236663818359375,
-0.0188751220703125,
0.06256103515625,
0.0219573974609375,
-0.03472900390625,
-0.0701904296875,
-0.038604736328125,
... |
pharaouk/biology_dataset_standardized_cluster_4 | 2023-10-13T02:14:23.000Z | [
"region:us"
] | pharaouk | null | null | 0 | 0 | 2023-10-13T02:14:21 | ---
dataset_info:
features: []
splits:
- name: train
num_bytes: 0
num_examples: 0
download_size: 324
dataset_size: 0
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
---
# Dataset Card for "biology_dataset_standardized_cluster_4"
[More Information needed](https://github.com/huggingface/datasets/blob/main/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards) | 417 | [
[
-0.024566650390625,
-0.0124969482421875,
0.036041259765625,
0.01776123046875,
-0.0205841064453125,
0.005069732666015625,
0.021942138671875,
-0.01458740234375,
0.06585693359375,
0.0179443359375,
-0.039398193359375,
-0.07232666015625,
-0.034820556640625,
0.017... |
pharaouk/biology_dataset_standardized_cluster_5 | 2023-10-13T02:14:32.000Z | [
"region:us"
] | pharaouk | null | null | 0 | 0 | 2023-10-13T02:14:30 | ---
dataset_info:
features: []
splits:
- name: train
num_bytes: 0
num_examples: 0
download_size: 324
dataset_size: 0
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
---
# Dataset Card for "biology_dataset_standardized_cluster_5"
[More Information needed](https://github.com/huggingface/datasets/blob/main/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards) | 417 | [
[
-0.030029296875,
-0.0075836181640625,
0.0306854248046875,
0.0181732177734375,
-0.02459716796875,
0.0015535354614257812,
0.0194244384765625,
-0.0186920166015625,
0.059844970703125,
0.015777587890625,
-0.04315185546875,
-0.07940673828125,
-0.041015625,
0.01725... |
pharaouk/biology_dataset_standardized_cluster_6 | 2023-10-13T02:14:42.000Z | [
"region:us"
] | pharaouk | null | null | 0 | 0 | 2023-10-13T02:14:40 | ---
dataset_info:
features: []
splits:
- name: train
num_bytes: 0
num_examples: 0
download_size: 324
dataset_size: 0
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
---
# Dataset Card for "biology_dataset_standardized_cluster_6"
[More Information needed](https://github.com/huggingface/datasets/blob/main/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards) | 417 | [
[
-0.022308349609375,
-0.00970458984375,
0.0289306640625,
0.0114288330078125,
-0.0222625732421875,
-0.0046234130859375,
0.0189208984375,
-0.0190582275390625,
0.06396484375,
0.01580810546875,
-0.04083251953125,
-0.0670166015625,
-0.0401611328125,
0.011512756347... |
pharaouk/biology_dataset_standardized_cluster_7 | 2023-10-13T02:14:51.000Z | [
"region:us"
] | pharaouk | null | null | 0 | 0 | 2023-10-13T02:14:49 | ---
dataset_info:
features: []
splits:
- name: train
num_bytes: 0
num_examples: 0
download_size: 324
dataset_size: 0
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
---
# Dataset Card for "biology_dataset_standardized_cluster_7"
[More Information needed](https://github.com/huggingface/datasets/blob/main/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards) | 417 | [
[
-0.02728271484375,
-0.00972747802734375,
0.030914306640625,
0.01552581787109375,
-0.03192138671875,
0.0014286041259765625,
0.0177001953125,
-0.015716552734375,
0.06317138671875,
0.0242156982421875,
-0.03131103515625,
-0.0731201171875,
-0.047210693359375,
0.0... |
pharaouk/biology_dataset_standardized_cluster_8 | 2023-10-13T02:15:00.000Z | [
"region:us"
] | pharaouk | null | null | 0 | 0 | 2023-10-13T02:14:58 | ---
dataset_info:
features: []
splits:
- name: train
num_bytes: 0
num_examples: 0
download_size: 324
dataset_size: 0
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
---
# Dataset Card for "biology_dataset_standardized_cluster_8"
[More Information needed](https://github.com/huggingface/datasets/blob/main/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards) | 417 | [
[
-0.024139404296875,
-0.00890350341796875,
0.031768798828125,
0.01380157470703125,
-0.0214691162109375,
0.00223541259765625,
0.01861572265625,
-0.016876220703125,
0.06597900390625,
0.0224609375,
-0.030670166015625,
-0.06524658203125,
-0.0474853515625,
0.01255... |
pharaouk/biology_dataset_standardized_cluster_9 | 2023-10-13T02:15:10.000Z | [
"region:us"
] | pharaouk | null | null | 0 | 0 | 2023-10-13T02:15:07 | ---
dataset_info:
features: []
splits:
- name: train
num_bytes: 0
num_examples: 0
download_size: 324
dataset_size: 0
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
---
# Dataset Card for "biology_dataset_standardized_cluster_9"
[More Information needed](https://github.com/huggingface/datasets/blob/main/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards) | 417 | [
[
-0.0207366943359375,
-0.01506805419921875,
0.033782958984375,
0.01552581787109375,
-0.0244140625,
0.01468658447265625,
0.0110015869140625,
-0.0121002197265625,
0.0718994140625,
0.0205230712890625,
-0.03741455078125,
-0.06414794921875,
-0.045379638671875,
0.0... |
pharaouk/biology_dataset_standardized_cluster_10 | 2023-10-13T02:15:19.000Z | [
"region:us"
] | pharaouk | null | null | 0 | 0 | 2023-10-13T02:15:17 | ---
dataset_info:
features: []
splits:
- name: train
num_bytes: 0
num_examples: 0
download_size: 324
dataset_size: 0
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
---
# Dataset Card for "biology_dataset_standardized_cluster_10"
[More Information needed](https://github.com/huggingface/datasets/blob/main/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards) | 418 | [
[
-0.0302581787109375,
-0.016143798828125,
0.03143310546875,
0.0217132568359375,
-0.0182952880859375,
0.0013017654418945312,
0.00983428955078125,
-0.01384735107421875,
0.07244873046875,
0.015350341796875,
-0.038970947265625,
-0.06597900390625,
-0.04638671875,
... |
pharaouk/biology_dataset_standardized_cluster_11 | 2023-10-13T02:15:28.000Z | [
"region:us"
] | pharaouk | null | null | 0 | 0 | 2023-10-13T02:15:26 | ---
dataset_info:
features: []
splits:
- name: train
num_bytes: 0
num_examples: 0
download_size: 324
dataset_size: 0
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
---
# Dataset Card for "biology_dataset_standardized_cluster_11"
[More Information needed](https://github.com/huggingface/datasets/blob/main/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards) | 418 | [
[
-0.029388427734375,
-0.01506805419921875,
0.031768798828125,
0.0156402587890625,
-0.0240020751953125,
0.007068634033203125,
0.014007568359375,
-0.01922607421875,
0.0731201171875,
0.0135040283203125,
-0.037017822265625,
-0.06903076171875,
-0.05035400390625,
0... |
pharaouk/biology_dataset_standardized_cluster_12 | 2023-10-13T02:15:37.000Z | [
"region:us"
] | pharaouk | null | null | 0 | 0 | 2023-10-13T02:15:35 | ---
dataset_info:
features: []
splits:
- name: train
num_bytes: 0
num_examples: 0
download_size: 324
dataset_size: 0
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
---
# Dataset Card for "biology_dataset_standardized_cluster_12"
[More Information needed](https://github.com/huggingface/datasets/blob/main/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards) | 418 | [
[
-0.0263671875,
-0.0153656005859375,
0.02972412109375,
0.02203369140625,
-0.0223846435546875,
0.00652313232421875,
0.0128936767578125,
-0.01788330078125,
0.07000732421875,
0.01263427734375,
-0.049530029296875,
-0.0711669921875,
-0.048004150390625,
0.005138397... |
pharaouk/biology_dataset_standardized_cluster_13 | 2023-10-13T02:15:47.000Z | [
"region:us"
] | pharaouk | null | null | 0 | 0 | 2023-10-13T02:15:45 | ---
dataset_info:
features: []
splits:
- name: train
num_bytes: 0
num_examples: 0
download_size: 324
dataset_size: 0
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
---
# Dataset Card for "biology_dataset_standardized_cluster_13"
[More Information needed](https://github.com/huggingface/datasets/blob/main/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards) | 418 | [
[
-0.0281829833984375,
-0.019805908203125,
0.028656005859375,
0.0241851806640625,
-0.025634765625,
0.0026721954345703125,
0.01273345947265625,
-0.01314544677734375,
0.06805419921875,
0.0129241943359375,
-0.050018310546875,
-0.0697021484375,
-0.048736572265625,
... |
pharaouk/biology_dataset_standardized_cluster_14 | 2023-10-13T02:15:57.000Z | [
"region:us"
] | pharaouk | null | null | 0 | 0 | 2023-10-13T02:15:54 | ---
dataset_info:
features: []
splits:
- name: train
num_bytes: 0
num_examples: 0
download_size: 324
dataset_size: 0
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
---
# Dataset Card for "biology_dataset_standardized_cluster_14"
[More Information needed](https://github.com/huggingface/datasets/blob/main/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards) | 418 | [
[
-0.03143310546875,
-0.0145721435546875,
0.0287017822265625,
0.024444580078125,
-0.025360107421875,
-0.002147674560546875,
0.0147552490234375,
-0.01439666748046875,
0.0626220703125,
0.0227813720703125,
-0.0484619140625,
-0.07171630859375,
-0.047454833984375,
... |
pharaouk/biology_dataset_standardized_cluster_15 | 2023-10-13T02:16:06.000Z | [
"region:us"
] | pharaouk | null | null | 0 | 0 | 2023-10-13T02:16:04 | ---
dataset_info:
features: []
splits:
- name: train
num_bytes: 0
num_examples: 0
download_size: 324
dataset_size: 0
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
---
# Dataset Card for "biology_dataset_standardized_cluster_15"
[More Information needed](https://github.com/huggingface/datasets/blob/main/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards) | 418 | [
[
-0.031890869140625,
-0.0212554931640625,
0.0279541015625,
0.022979736328125,
-0.0213623046875,
0.005313873291015625,
0.0131378173828125,
-0.0183563232421875,
0.06365966796875,
0.013397216796875,
-0.048126220703125,
-0.07281494140625,
-0.03912353515625,
0.011... |
pharaouk/biology_dataset_standardized_cluster_16 | 2023-10-13T02:16:15.000Z | [
"region:us"
] | pharaouk | null | null | 0 | 0 | 2023-10-13T02:16:13 | ---
dataset_info:
features: []
splits:
- name: train
num_bytes: 0
num_examples: 0
download_size: 324
dataset_size: 0
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
---
# Dataset Card for "biology_dataset_standardized_cluster_16"
[More Information needed](https://github.com/huggingface/datasets/blob/main/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards) | 418 | [
[
-0.0261383056640625,
-0.0178680419921875,
0.0289306640625,
0.012725830078125,
-0.0260162353515625,
0.005062103271484375,
0.01192474365234375,
-0.01824951171875,
0.06561279296875,
0.0136260986328125,
-0.045074462890625,
-0.07061767578125,
-0.04083251953125,
0... |
pharaouk/biology_dataset_standardized_cluster_17 | 2023-10-13T02:16:25.000Z | [
"region:us"
] | pharaouk | null | null | 0 | 0 | 2023-10-13T02:16:23 | ---
dataset_info:
features: []
splits:
- name: train
num_bytes: 0
num_examples: 0
download_size: 324
dataset_size: 0
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
---
# Dataset Card for "biology_dataset_standardized_cluster_17"
[More Information needed](https://github.com/huggingface/datasets/blob/main/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards) | 418 | [
[
-0.0252685546875,
-0.019622802734375,
0.027923583984375,
0.01377105712890625,
-0.02447509765625,
0.006378173828125,
0.010101318359375,
-0.0160980224609375,
0.060699462890625,
0.013031005859375,
-0.044586181640625,
-0.07220458984375,
-0.0423583984375,
0.00304... |
pharaouk/biology_dataset_standardized_cluster_18 | 2023-10-13T02:16:34.000Z | [
"region:us"
] | pharaouk | null | null | 0 | 0 | 2023-10-13T02:16:32 | ---
dataset_info:
features: []
splits:
- name: train
num_bytes: 0
num_examples: 0
download_size: 324
dataset_size: 0
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
---
# Dataset Card for "biology_dataset_standardized_cluster_18"
[More Information needed](https://github.com/huggingface/datasets/blob/main/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards) | 418 | [
[
-0.0269775390625,
-0.0238189697265625,
0.0275421142578125,
0.018310546875,
-0.02105712890625,
0.004909515380859375,
0.01251220703125,
-0.021087646484375,
0.06378173828125,
0.0185699462890625,
-0.04840087890625,
-0.066650390625,
-0.04058837890625,
0.007232666... |
open-llm-leaderboard/details_ziqingyang__chinese-llama-2-13b | 2023-10-13T02:16:41.000Z | [
"region:us"
] | open-llm-leaderboard | null | null | 0 | 0 | 2023-10-13T02:16:32 | ---
pretty_name: Evaluation run of ziqingyang/chinese-llama-2-13b
dataset_summary: "Dataset automatically created during the evaluation run of model\
\ [ziqingyang/chinese-llama-2-13b](https://huggingface.co/ziqingyang/chinese-llama-2-13b)\
\ on the [Open LLM Leaderboard](https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard).\n\
\nThe dataset is composed of 3 configuration, each one coresponding to one of the\
\ evaluated task.\n\nThe 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.\n\
\nAn additional configuration \"results\" store all the aggregated results of the\
\ run (and is used to compute and display the agregated metrics on the [Open LLM\
\ Leaderboard](https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard)).\n\
\nTo load the details from a run, you can for instance do the following:\n```python\n\
from datasets import load_dataset\ndata = load_dataset(\"open-llm-leaderboard/details_ziqingyang__chinese-llama-2-13b\"\
,\n\t\"harness_winogrande_5\",\n\tsplit=\"train\")\n```\n\n## Latest results\n\n\
These are the [latest results from run 2023-10-13T02:16:28.624292](https://huggingface.co/datasets/open-llm-leaderboard/details_ziqingyang__chinese-llama-2-13b/blob/main/results_2023-10-13T02-16-28.624292.json)(note\
\ that their might be results for other tasks in the repos if successive evals didn't\
\ cover the same tasks. You find each in the results and the \"latest\" split for\
\ each eval):\n\n```python\n{\n \"all\": {\n \"em\": 0.37741191275167785,\n\
\ \"em_stderr\": 0.004964183842623747,\n \"f1\": 0.42850880872483355,\n\
\ \"f1_stderr\": 0.004835429715953239,\n \"acc\": 0.39816494163081856,\n\
\ \"acc_stderr\": 0.008707972830386747\n },\n \"harness|drop|3\": {\n\
\ \"em\": 0.37741191275167785,\n \"em_stderr\": 0.004964183842623747,\n\
\ \"f1\": 0.42850880872483355,\n \"f1_stderr\": 0.004835429715953239\n\
\ },\n \"harness|gsm8k|5\": {\n \"acc\": 0.039423805913570885,\n \
\ \"acc_stderr\": 0.005360280030342443\n },\n \"harness|winogrande|5\"\
: {\n \"acc\": 0.7569060773480663,\n \"acc_stderr\": 0.01205566563043105\n\
\ }\n}\n```"
repo_url: https://huggingface.co/ziqingyang/chinese-llama-2-13b
leaderboard_url: https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard
point_of_contact: clementine@hf.co
configs:
- config_name: harness_drop_3
data_files:
- split: 2023_10_13T02_16_28.624292
path:
- '**/details_harness|drop|3_2023-10-13T02-16-28.624292.parquet'
- split: latest
path:
- '**/details_harness|drop|3_2023-10-13T02-16-28.624292.parquet'
- config_name: harness_gsm8k_5
data_files:
- split: 2023_10_13T02_16_28.624292
path:
- '**/details_harness|gsm8k|5_2023-10-13T02-16-28.624292.parquet'
- split: latest
path:
- '**/details_harness|gsm8k|5_2023-10-13T02-16-28.624292.parquet'
- config_name: harness_winogrande_5
data_files:
- split: 2023_10_13T02_16_28.624292
path:
- '**/details_harness|winogrande|5_2023-10-13T02-16-28.624292.parquet'
- split: latest
path:
- '**/details_harness|winogrande|5_2023-10-13T02-16-28.624292.parquet'
- config_name: results
data_files:
- split: 2023_10_13T02_16_28.624292
path:
- results_2023-10-13T02-16-28.624292.parquet
- split: latest
path:
- results_2023-10-13T02-16-28.624292.parquet
---
# Dataset Card for Evaluation run of ziqingyang/chinese-llama-2-13b
## Dataset Description
- **Homepage:**
- **Repository:** https://huggingface.co/ziqingyang/chinese-llama-2-13b
- **Paper:**
- **Leaderboard:** https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard
- **Point of Contact:** clementine@hf.co
### Dataset Summary
Dataset automatically created during the evaluation run of model [ziqingyang/chinese-llama-2-13b](https://huggingface.co/ziqingyang/chinese-llama-2-13b) on the [Open LLM Leaderboard](https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard).
The dataset is composed of 3 configuration, each one coresponding 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 additional configuration "results" store all the aggregated results of the run (and is used to compute and display the agregated metrics on the [Open LLM Leaderboard](https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard)).
To load the details from a run, you can for instance do the following:
```python
from datasets import load_dataset
data = load_dataset("open-llm-leaderboard/details_ziqingyang__chinese-llama-2-13b",
"harness_winogrande_5",
split="train")
```
## Latest results
These are the [latest results from run 2023-10-13T02:16:28.624292](https://huggingface.co/datasets/open-llm-leaderboard/details_ziqingyang__chinese-llama-2-13b/blob/main/results_2023-10-13T02-16-28.624292.json)(note that their might be results for other tasks in the repos if successive evals didn't cover the same tasks. You find each in the results and the "latest" split for each eval):
```python
{
"all": {
"em": 0.37741191275167785,
"em_stderr": 0.004964183842623747,
"f1": 0.42850880872483355,
"f1_stderr": 0.004835429715953239,
"acc": 0.39816494163081856,
"acc_stderr": 0.008707972830386747
},
"harness|drop|3": {
"em": 0.37741191275167785,
"em_stderr": 0.004964183842623747,
"f1": 0.42850880872483355,
"f1_stderr": 0.004835429715953239
},
"harness|gsm8k|5": {
"acc": 0.039423805913570885,
"acc_stderr": 0.005360280030342443
},
"harness|winogrande|5": {
"acc": 0.7569060773480663,
"acc_stderr": 0.01205566563043105
}
}
```
### Supported Tasks and Leaderboards
[More Information Needed]
### Languages
[More Information Needed]
## Dataset Structure
### Data Instances
[More Information Needed]
### Data Fields
[More Information Needed]
### Data Splits
[More Information Needed]
## Dataset Creation
### Curation Rationale
[More Information Needed]
### Source Data
#### Initial Data Collection and Normalization
[More Information Needed]
#### Who are the source language producers?
[More Information Needed]
### Annotations
#### Annotation process
[More Information Needed]
#### Who are the annotators?
[More Information Needed]
### Personal and Sensitive Information
[More Information Needed]
## Considerations for Using the Data
### Social Impact of Dataset
[More Information Needed]
### Discussion of Biases
[More Information Needed]
### Other Known Limitations
[More Information Needed]
## Additional Information
### Dataset Curators
[More Information Needed]
### Licensing Information
[More Information Needed]
### Citation Information
[More Information Needed]
### Contributions
[More Information Needed] | 7,265 | [
[
-0.0246734619140625,
-0.04241943359375,
0.0161285400390625,
0.0247344970703125,
-0.0192413330078125,
0.0126953125,
-0.02935791015625,
-0.02093505859375,
0.0283660888671875,
0.033416748046875,
-0.054107666015625,
-0.06744384765625,
-0.04888916015625,
0.014968... |
pharaouk/biology_dataset_standardized_cluster_19 | 2023-10-13T02:16:43.000Z | [
"region:us"
] | pharaouk | null | null | 0 | 0 | 2023-10-13T02:16:41 | ---
dataset_info:
features: []
splits:
- name: train
num_bytes: 0
num_examples: 0
download_size: 324
dataset_size: 0
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
---
# Dataset Card for "biology_dataset_standardized_cluster_19"
[More Information needed](https://github.com/huggingface/datasets/blob/main/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards) | 418 | [
[
-0.0268707275390625,
-0.0255584716796875,
0.027008056640625,
0.0235137939453125,
-0.0249786376953125,
0.004062652587890625,
0.0123291015625,
-0.0152435302734375,
0.06842041015625,
0.01690673828125,
-0.04327392578125,
-0.06317138671875,
-0.043701171875,
0.005... |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.