Hugging Face's logo

Datasets:
m-a-p
/
FineFineWeb-validation

Tasks:
Text Classification
Text Generation
Modalities:
Tabular
Text
Formats:
json
Languages:
English
Size:
10K - 100K
Libraries:
Datasets
Dask
License:
Dataset card Data Studio Files Community
FineFineWeb-validation / README.md
ZenMoore's picture
ZenMoore
Upload 5 files
6c3597c verified
preview code
|
raw
history blame
11.2 kB
metadata
license: apache-2.0
task_categories:
  - text-classification
  - text2text-generation
  - text-generation
language:
  - en
size_categories:
  - n>1T

FineFineWeb: A Comprehensive Study on Fine-Grained Domain Web Corpus

arXiv: Coming Soon

Project Page: Coming Soon

Blog: Coming Soon

Data Statistics

Domain (#tokens/#samples) Iteration 1 Iteration 2 Iteration 3 Total
medical 140.03B 813.46M 4.97B 145.81B
chemistry 28.69B 588.92M 131.46M 29.41B
systems_science 25.41B 11.64B 180.39M 37.22B
history 45.27B 1.56B 1.74B 48.57B
movie 13.09B 598.29M 155.10M 13.85B
design 96.58B 3.80B 450.00M 100.82B
fashion 18.72B 977.27M 264.01M 19.96B
celebrity 10.29B 745.94M 4.48B 15.52B
sociology 76.34B 3.59B 8.88B 88.82B
computer_science_and_technology 194.46B 3.95B 4.76B 203.16B
statistics 19.59B 1.15B 1.70B 22.44B
christianity 47.72B 403.68M 732.55M 48.86B
electronic_science 31.18B 8.01B 482.62M 39.67B
game 43.47B 2.36B 2.68B 48.51B
photo 6.77B 1.78B 41.44M 8.59B
biology 88.58B 371.29M 838.82M 89.79B
transportation_engineering 13.61B 6.61B 972.50M 21.19B
instrument_science 6.36B 2.17B 165.43M 8.70B
entertainment 152.92B 1.67B 5.06B 159.65B
travel 78.87B 584.78M 957.26M 80.41B
food 59.88B 136.32M 1.01B 61.03B
mathematics 5.70B 50.33M 261.65M 6.01B
mining_engineering 8.35B 206.05M 529.02M 9.08B
astronomy 5.54B 134.39M 92.27M 5.77B
artistic 178.25B 5.79B 3.75B 187.80B
automotive 37.88B 436.34M 911.65M 39.22B
music_and_dance 15.92B 745.94M 4.48B 15.52B
geography 110.18B 1.16B 192.67M 111.53B
ocean_science 2.36B 483.04M 229.43M 3.07B
textile_science 2.59B 3.00B 94.56M 5.69B
mechanical_engineering 86.13B 1.24B 129.96M 87.49B
topicality 34.87M 5.22M / 40.09M
psychology 51.53B 688.50M 2.56B 54.78B
philosophy 47.99B 121.26M 335.77M 48.44B
atmospheric_science 2.84B 102.04M 259.25M 3.21B
landscape_architecture 3.07B 2.93B 53.24M 15.12B
sports 122.47B 379.18M 1.83B 124.68B
politics 79.52B 253.26M 930.96M 80.70B
economics 205.01B 1.23B 2.63B 208.87B
environmental_science 56.98B 1.48B 920.77M 59.37B
agronomy 13.08B 1.02B 229.04M 14.33B
news 328.47B 12.37B 11.34B 352.18B
public_administration 100.13B 5.54B 716.81M 106.39B
pet 12.58B 154.14M 307.28M 13.05B
library 59.00B 5.01B 36.56M 64.05B
civil_engineering 9.46B 1.36B 402.91M 11.22B
literature 73.37B 7.22B 69.75B 150.34B
journalism_and_media_communication 440.98B 21.00B 1.55B 463.53B
nuclear_science 495.51M 79.89M 78.79M 654.19M
beauty 19.75B 709.15M 1.01B 21.47B
physics 22.29B 372.21M 191.17M 22.85B
painting 374.41M 429.63M 96.57M 900.61M
hobby 150.23B 42.78B 44.05B 237.06B
health 191.20B 427.93M 18.43B 210.06B
relationship 21.87B 3.69B 129.60M 25.69B
petroleum_and_natural_gas_engineering 950.08M 463.65M 121.56M 1.54B
optical_engineering 2.54B 253.06M 263.99M 3.06B
hydraulic_engineering 57.36M 75.40M 3.65M 136.41M
urban_planning 12.13B 2.93B 53.24M 15.12B
communication_engineering 9.21B 3.72B 327.66M 13.26B
aerospace 6.15B 261.63M 352.68M 6.77B
law 128.58B 455.19M 2.38B 131.42B
finance 151.07B 327.45M 1.13B 152.53B
drama_and_film 20.44B 11.22B 206.27M 31.86B
materials_science 18.95B 1.11B 303.66M 20.37B
weapons_science 80.62M 3.51B 140.89M 3.73B
gamble 30.12B 696.52M 158.48M 30.98B
Total 4007.48B 207.39B 207.99B 4422.86B

Data Construction Workflow

finefineweb-data-workflow

The data construction workflow can be summarized as follows:

  1. Deduplicate: The FineWeb dataset is deduplicated using exact deduplication and MinHash techniques to remove redundant data.

  2. URL Labeling: Root URLs from FineWeb are counted, and the top 1 million URLs are labeled using GPT-4. This step generates DoI (Domain-of-Interest) Coarse-Grained URLs and DoNI (Domain-of-Non-Interest) Coarse-Grained URLs as seed data sources.

  3. Coarse Recall:

    a. Based on the labeled root URLs, data is sampled for each domain.

    b. The sampled data is labeled using Qwen2-7B-Instruct, producing 500K DoI Positive Data and 500K DoI Negative Data (note that for N>1 iterations, each 500K samples are composed of 250K sampled original seed data and 250K refined data after Fine Recall).

    c. A binary FastText model is trained per domain using the labeled data.

    d. The FastText model performs coarse recall on FineWeb, generating Coarse DoI Data.

  4. Fine Recall:

    a. The Coarse DoI Data is labeled using Qwen2-72B-Instruct to produce 100K DoI Positive Data and 50K DoI Negative Data, with the latter further augmented with 50K negative samples from earlier FastText training.

    b. A BERT model is trained using this labeled data.

    c. The BERT model performs fine recall on the Coarse DoI Data, producing a refined dataset, which is the DoI subset of FineFineWeb.

  5. Coarse-Fine Recall Iteration: The workflow of coarse and fine recall iterates for 3 rounds with the following adjustments:

    a. FastText is re-trained using updated seed data, which combines BERT-recalled samples, BERT-dropped samples, and previously labeled seed data.

    b. The BERT model keeps frozen during subsequent iterations.

    c. Steps for training FastText, coarse recall, and fine recall are repeated without re-labeling data with Qwen2-Instruct models.

Domain-Domain Similarity Analysis

  1. Perform proportional weighted sampling of the domain subsets based on the sample size of each domain, with a total of 1 billion tokens sampled from the domain subsets.
  2. Use the BGE-M3 model to compute the embeddings of the samples in each domain subset, referred to as domain embeddings.
  3. Use the BGE-M3 model to compute the embeddings of the samples in each benchmark, referred to as benchmark embeddings (bench embeddings).
  4. Calculate the MMD distance and the Wasserstein distance between the domain embeddings and the benchmark embeddings.

domain-benchmark similarity

The results above reveal the following observations:

  1. The two code-related benchmarks, MBPP and HumanEval, exhibit relatively large distances from nearly all domains, indicating that the proportion of code data in the training set is relatively small. Notably, their distance to the mathematics domain is comparatively smaller, suggesting a certain degree of overlap between mathematics data and code data.
  2. Benchmarks such as Hellaswag, ARC, MMLU, and BoolQ have distances that are close to almost all domains, except for the gamble domain. This indicates that the samples in these benchmarks involve synergetic effects across multiple domains of knowledge, with a wide distribution.
  3. GSM8K and TriviaQA show significant discrepancies with a small number of domains, suggesting that the distribution differences between domains are more pronounced for samples involving grade-school mathematics and fact-based question answering. Some domains contain a substantial amount of this type of data, while others do not.
  4. The gamble domain exhibits substantial differences from other domains and has large distances from all benchmarks, indicating that pretraining data related to gambling provides limited benefits for these benchmarks.

Domain-Domain Duplication

Let $D_1, D_2, \dots, D_N$ represent $N$ distinct domains, where we select top-20 URLs for each domain $D_i$, denoted as ${U_{i1}, U_{i2}, \dots, U_{i20}}$,. The total set of URLs across all domains is represented as $\mathcal{U}$, and the total number of URLs is $M = |\mathcal{U}|$.

For each URL $U_k \in \mathcal{U}$, the term frequency (TF) is defined as the proportion of $U_k$ in the total set of URLs:

$\text{TF}(U_k) = \frac{\text{count}(U_k)}{M}$

where $\text{count}(U_k)$ is the number of times $U_k$ appears in $\mathcal{U}$. Additionally, the document frequency $K_k$ of $U_k$ is the number of domains in which $U_k$ appears. Based on this, the inverse document frequency (IDF) is calculated as:

$\text{IDF}(U_k) = \log(\frac{N}{K_k})$

The TF-IDF value for each URL $U_{ij}$ in a specific domain $D_i$ is then computed as:

$\text{TF-IDF}(U_{ij}) = \text{TF}(U_{ij}) \times \text{IDF}(U_{ij})$

domain-domain URL duplication

Using the TF-IDF values of all URLs within a domain, the domain-domain duplicate rate can be analyzed by comparing the distribution of TF-IDF values across domains. If a domain has many URLs with high TF-IDF values, it indicates that the domain’s URLs are relatively unique and significant within the entire set of URLs. Conversely, if a domain has many URLs with low TF-IDF values, it suggests that the domain's URLs are more common across other domains. Analyzing these values helps assess how similar or redundant a domain's content is in relation to others based on its URL composition.

As shown in the figure, most domains have low duplication rates, except for topicality, pet, and atmospheric science.

Domain-Benchmark BPC-Acc Correlation

Experimental method: Using 28 models (see the paper), we first calculate BPC for all domains to obtain a model ranking $R_D$. Similarly, we compute scores across all benchmarks to obtain a model ranking $R_M$. We then calculate the Spearman correlation between $R_D$ and $R_M$.

domain-benchmark BPC-Acc correlation

  • For benchmarks like ARC, MMLU, GSM8K, HumanEval, and MBPP, STEM-related domains show higher correlation rankings, particularly mathematics, physics, and systems science.
  • For TriviaQA, which emphasizes factual knowledge over reasoning, domains rich in world knowledge such as literature, history, and library science demonstrate higher correlation rankings.

Bibtex 

@misc{
title={FineFineWeb: A Comprehensive Study on Fine-grained Domain Web Corpus},
url={[https://huggingface.co/datasets/m-a-p/FineFineWeb](https://huggingface.co/datasets/m-a-p/FineFineWeb)},
author = {M-A-P, Ge Zhang*, Xinrun Du*, Zhimiao Yu*, Zili Wang*, Zekun Wang, Shuyue Guo, Tianyu Zheng, Kang Zhu, Jerry Liu, Shawn Yue, Binbin Liu, Zhongyuan Peng, Yifan Yao, Jack Yang, Ziming Li, Bingni Zhang, Wenhu Chen, Minghao Liu, Tianyu Liu, Xiaohuan Zhou, Qian Liu, Taifeng Wang+, Wenhao Huang+},
publisher={huggingface},
verision={v0.1.0},
month={December},
year={2024}
}