Metallic Glass · foundry-ml
dataset_metallicglass_rc_llm
LLM-extracted critical cooling rates for 297 metallic glasses across research literature.
What dataset_metallicglass_rc_llm is
dataset_metallicglass_rc_llm is a tabular materials science dataset developed by Maciej P. Polak and Dane Morgan, published through the Materials Data Facility and Foundry-ML. It contains experimental critical cooling rates (Rc) for 297 metallic glass compositions. The critical cooling rate indicates how rapidly a liquid alloy must be cooled to bypass crystallisation and form a solid amorphous metallic glass.
The dataset was created by extracting property values from scientific research papers using conversational large language models combined with prompt engineering. It acts as a text-mined companion to structured critical cooling rate datasets, preserving the source text passages and references for each entry.
Materials scientists and computational researchers can use this dataset to train machine learning models to predict glass-forming ability, benchmark property regression algorithms, and examine language model data extraction performance on technical literature.
How it was made
The data was gathered from scientific papers using conversational language models guided by prompt engineering techniques to identify relevant excerpts and extract experimental values. Each record links the extracted numerical data back to its context.
The tabular structure includes input features containing the extracted literature passage, the publication DOI, the alloy chemical composition, the raw critical cooling rate, and its reported units. The designated regression target is log(Rc), representing the critical cooling rate on a logarithmic scale in kelvins per second (K/s).
Key facts
- Publisher
- Materials Data Facility / Foundry-ML
- Publication Year
- 2024
- Dataset Size
- 297 metallic glasses
- Data Modality
- Tabular, text
- File Format
- Parquet
- Target Property
- log(Rc) in K/s
- DOI
- 10.18126/ndyp-yv32
What you can use it for
- Critical Cooling Rate Regression
- Train machine learning regressors to predict the logarithmic critical cooling rate log(Rc) directly from alloy chemical compositions.
- Glass-Forming Classification
- Group compositions by cooling rate thresholds to classify alloys according to their glass-forming ability.
- Information Extraction Validation
- Evaluate the accuracy of language-model-based scientific information extraction pipelines against original passage text and source DOIs.
Explore the data
The schema, real rows, full-text search and a SQL filter — read in your browser, nothing downloaded.
Explore
Read the real rows without downloading anything
Reading rows…
LiveRead from foundry-ml/dataset_metallicglass_rc_llm at the moment you asked. Nothing is cached or stored — every row above came from that request.
How the columns are distributed
Over 297 rows of default/train
DOI
string_text- min
- 25
- median
- 29
- mean
- 29.15
- max
- 34
Material
string_text- min
- 6
- median
- 15
- mean
- 17.56
- max
- 40
Passage
string_text- min
- 167
- median
- 434
- mean
- 445.17
- max
- 964
Rc
float- min
- 2.0e-5
- median
- 10
- mean
- 5.4B
- max
- 1000.0B
Unit
string_label- K/s100%
- K0%
log(Rc)
float- min
- -4.68
- median
- 1
- mean
- 1.31
- max
- 12
Strengths and limits
Good at
- Provides direct traceability for every record through the original text passage and publication DOI.
- Supplies both raw critical cooling rate values and standard logarithmic targets log(Rc) in kelvins per second.
- Offers targeted experimental property data for 297 metallic glass compositions.
Watch out for
- Contains a relatively small sample set of 297 entries.
- Provides only a single training split with no pre-partitioned validation or test sets.
Questions
Can I look inside dataset_metallicglass_rc_llm without downloading it?
Yes. This page reads its schema, real rows and per-column statistics in the browser, with full-text search and a SQL filter. Nothing is downloaded to your machine.
What licence is dataset_metallicglass_rc_llm under?
Its authors declare other. Read the terms on the original page before reusing the data.
What target variable does the dataset define?
The target variable is log(Rc), which is the critical cooling rate expressed on a logarithmic scale in units of kelvins per second (K/s).
How was the data extracted from scientific publications?
The data was retrieved from published papers using conversational language models guided by prompt engineering strategies.
What columns are included in each record?
Each row contains the source text passage, original DOI, alloy material composition, critical cooling rate (Rc), units, and log(Rc).
How to cite it
Polak, Maciej P. and Morgan, Dane. Extracting Accurate Materials Data from Research Papers with Conversational Language Models and Prompt Engineering. Materials Data Facility, 2024. doi:10.18126/ndyp-yv32.
Written from dataset_metallicglass_rc_llm's own dataset card. Statements citing a figure that does not appear there are removed before publication. Licence and running details come from our registry.