Dataset Viewer
Auto-converted to Parquet Duplicate
text
stringlengths
10.2k
1.04M
source
stringclasses
6 values
token_length
int64
8.19k
730k
actual_len
int64
8.19k
8.19k
mean_distance
float64
1.55k
2.71k
local_ratio
float64
0.06
0.16
entropy
float64
0.72
0.89
classifier_score
null
label
stringclasses
3 values
Part 9 out of 21 "How do you do, Esther?" said she. "Do you recollect me?" I gave her my hand and told her yes and that she was very little "I wonder you remember those times, Esther," she returned with her old asperity. "They are changed now. Well! I am glad to see you, and glad you are not too proud to know me." But ...
fineweb
23,432
8,192
2,112.903291
0.110345
0.824389
null
ambiguous
// Copyright 2019 Kaleido // Licensed under the Apache License, Version 2.0 (the "License"); // you may not use this file except in compliance with the License. // You may obtain a copy of the License at // http://www.apache.org/licenses/LICENSE-2.0 // Unless required by applicable law or agreed to in writing, s...
stack_go
11,493
8,192
2,035.520243
0.122488
0.789693
null
ambiguous
"/*\n **** BEGIN LICENSE BLOCK *****\n * Version: EPL 1.0/GPL 2.0/LGPL 2.1\n *\n * The contents of t(...TRUNCATED)
stack_java
40,722
8,192
1,933.178472
0.128583
0.795993
null
low
"# -*- coding: utf-8 -*-\n# Generated by the protocol buffer compiler. DO NOT EDIT!\n# source: tens(...TRUNCATED)
stack_python
23,772
8,192
1,972.840603
0.093905
0.788319
null
ambiguous
"# Copyright 2009 by Peter Cock. All rights reserved.\n# This code is part of the Biopython distrib(...TRUNCATED)
stack_python
9,699
8,192
1,955.489943
0.108716
0.793965
null
ambiguous
"# -*- coding: utf-8 -*-\nr\"\"\"\nHyperplane Arrangements\n\nBefore talking about hyperplane arrang(...TRUNCATED)
stack_python
35,690
8,192
2,002.104228
0.11943
0.801868
null
ambiguous
"The President's Corner From the Editorís Desk\nLaurels and Darts Letters to the Editor\nOld New Or(...TRUNCATED)
fineweb
28,371
8,192
2,058.800885
0.121537
0.797659
null
ambiguous
"May 25, 2005\nAt Last, The Final\nThis is it. My last post here on Idol Tongues.\nI'd like to take (...TRUNCATED)
fineweb
8,368
8,192
2,179.011948
0.091953
0.838377
null
high
"# Copyright (c) 2011 X.commerce, a business unit of eBay Inc.\n# Copyright 2010 United States Gover(...TRUNCATED)
stack_python
18,793
8,192
1,986.135519
0.125956
0.798687
null
low
"// Copyright (c) 2011-2017 The Bitcoin Core developers\n// Distributed under the MIT software licen(...TRUNCATED)
stack_c++
8,249
8,192
1,984.963847
0.136907
0.78823
null
low
End of preview. Expand in Data Studio

YAML Metadata Warning:empty or missing yaml metadata in repo card

Check out the documentation for more information.

Qwen 8K Attention Gold Labels

4,998 text samples with attention pattern metrics extracted from Qwen2.5-7B-Instruct at 8K context length (layer 14).

Sources

Source Samples
fineweb 833
stack_python 833
stack_java 833
stack_go 833
stack_c++ 833
stack_javascript 833

Columns

  • text - Full document text
  • source - Data source (fineweb or stack_*)
  • token_length - Requested context length
  • actual_len - Actual tokens processed (8192)
  • mean_distance - Average attention distance in tokens
  • local_ratio - Fraction of attention within 128-token window
  • entropy - Normalized attention entropy (0-1)
  • label - high/low/ambiguous based on attention patterns

Usage

from datasets import load_dataset
ds = load_dataset("KevinDavidHayes/qwen-8k-attention-gold")
Downloads last month
8