update model card
Browse files
README.md
CHANGED
@@ -45,7 +45,7 @@ Model Pretrained using Masked Language Modelling on 2 million crystal structures
|
|
45 |
|
46 |
### Direct Use
|
47 |
|
48 |
-
The base model can be used for generating meaningful embeddings of bulk structures without further training.
|
49 |
This model is ideal if finetuned for narrowdown tasks.
|
50 |
|
51 |
### Downstream Use
|
@@ -55,84 +55,49 @@ This model can be used with fientuning for property prediction, classification o
|
|
55 |
|
56 |
## Bias, Risks, and Limitations
|
57 |
|
58 |
-
> Model was trained only on bulk structures (**n0w0f/MatText - pretrain2m** - dataset)
|
59 |
|
60 |
-
|
61 |
|
62 |
-
|
63 |
|
64 |
-
Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
|
65 |
|
66 |
## How to Get Started with the Model
|
67 |
|
68 |
-
|
69 |
-
|
70 |
-
|
|
|
71 |
|
72 |
## Training Details
|
73 |
|
74 |
### Training Data
|
75 |
|
76 |
-
|
|
|
77 |
|
78 |
-
|
79 |
|
80 |
-
|
81 |
-
|
82 |
-
<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
|
83 |
|
84 |
-
#### Preprocessing [optional]
|
85 |
|
86 |
-
|
87 |
|
88 |
|
89 |
#### Training Hyperparameters
|
90 |
|
91 |
-
- **Training regime:**
|
92 |
-
|
93 |
-
#### Speeds, Sizes, Times [optional]
|
94 |
|
95 |
-
<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
|
96 |
|
97 |
-
[More Information Needed]
|
98 |
-
|
99 |
-
## Evaluation
|
100 |
-
|
101 |
-
<!-- This section describes the evaluation protocols and provides the results. -->
|
102 |
|
103 |
### Testing Data, Factors & Metrics
|
104 |
|
105 |
#### Testing Data
|
106 |
|
107 |
-
|
108 |
-
|
109 |
-
[More Information Needed]
|
110 |
-
|
111 |
-
#### Factors
|
112 |
-
|
113 |
-
<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
|
114 |
|
115 |
-
[More Information Needed]
|
116 |
-
|
117 |
-
#### Metrics
|
118 |
-
|
119 |
-
<!-- These are the evaluation metrics being used, ideally with a description of why. -->
|
120 |
|
121 |
-
[More Information Needed]
|
122 |
-
|
123 |
-
### Results
|
124 |
|
125 |
-
[More Information Needed]
|
126 |
-
|
127 |
-
#### Summary
|
128 |
-
|
129 |
-
|
130 |
-
|
131 |
-
## Model Examination [optional]
|
132 |
-
|
133 |
-
<!-- Relevant interpretability work for the model goes here -->
|
134 |
-
|
135 |
-
[More Information Needed]
|
136 |
|
137 |
## Environmental Impact
|
138 |
|
@@ -140,31 +105,22 @@ Use the code below to get started with the model.
|
|
140 |
|
141 |
Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
|
142 |
|
143 |
-
- **Hardware Type:**
|
144 |
-
- **Hours used:**
|
145 |
-
- **Cloud Provider:**
|
146 |
-
- **Compute Region:**
|
147 |
-
- **Carbon Emitted:**
|
148 |
-
|
149 |
-
## Technical Specifications [optional]
|
150 |
-
|
151 |
-
### Model Architecture and Objective
|
152 |
-
|
153 |
-
[More Information Needed]
|
154 |
-
|
155 |
-
### Compute Infrastructure
|
156 |
|
157 |
-
[More Information Needed]
|
158 |
-
|
159 |
-
#### Hardware
|
160 |
|
161 |
-
|
162 |
|
163 |
#### Software
|
164 |
|
165 |
-
|
|
|
|
|
166 |
|
167 |
-
|
168 |
|
169 |
<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
|
170 |
|
@@ -176,20 +132,12 @@ Carbon emissions can be estimated using the [Machine Learning Impact calculator]
|
|
176 |
|
177 |
[More Information Needed]
|
178 |
|
179 |
-
## Glossary [optional]
|
180 |
-
|
181 |
-
<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
|
182 |
-
|
183 |
-
[More Information Needed]
|
184 |
-
|
185 |
-
## More Information [optional]
|
186 |
|
187 |
-
|
188 |
-
|
189 |
-
## Model Card Authors [optional]
|
190 |
|
191 |
-
[
|
192 |
|
193 |
## Model Card Contact
|
194 |
|
195 |
-
[
|
|
|
|
45 |
|
46 |
### Direct Use
|
47 |
|
48 |
+
The base model can be used for generating meaningful features/embeddings of bulk structures without further training.
|
49 |
This model is ideal if finetuned for narrowdown tasks.
|
50 |
|
51 |
### Downstream Use
|
|
|
55 |
|
56 |
## Bias, Risks, and Limitations
|
57 |
|
58 |
+
> Model was trained only on bulk structures (**n0w0f/MatText - pretrain2m** - dataset).
|
59 |
|
60 |
+
The pertaining dataset is a subset of the materials deposited in the NOMAD archive. We queried only 3D-connected structures (i.e., excluding 2D materials, which often require special treatment) and, for consistency, limited our query to materials for which the bandgap has been computed using the PBE functional and the VASP code.
|
61 |
|
62 |
+
### Recommendations
|
63 |
|
|
|
64 |
|
65 |
## How to Get Started with the Model
|
66 |
|
67 |
+
```python
|
68 |
+
from transformers import AutoModel
|
69 |
+
model = AutoModel.from_pretrained("n0w0f/MatText-cifp1-2m")
|
70 |
+
```
|
71 |
|
72 |
## Training Details
|
73 |
|
74 |
### Training Data
|
75 |
|
76 |
+
**n0w0f/MatText - pretrain2m**
|
77 |
+
The dataset contains crystal structures in various text representations and labels for some subsets.
|
78 |
|
79 |
+
https://huggingface.co/datasets/n0w0f/MatText
|
80 |
|
81 |
+
<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
|
|
|
|
|
82 |
|
|
|
83 |
|
84 |
+
### Training Procedure
|
85 |
|
86 |
|
87 |
#### Training Hyperparameters
|
88 |
|
89 |
+
- **Training regime:** fp32 <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
|
|
|
|
|
90 |
|
|
|
91 |
|
|
|
|
|
|
|
|
|
|
|
92 |
|
93 |
### Testing Data, Factors & Metrics
|
94 |
|
95 |
#### Testing Data
|
96 |
|
97 |
+
https://huggingface.co/datasets/n0w0f/MatText/viewer/pretrain2m/test
|
|
|
|
|
|
|
|
|
|
|
|
|
98 |
|
|
|
|
|
|
|
|
|
|
|
99 |
|
|
|
|
|
|
|
100 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
101 |
|
102 |
## Environmental Impact
|
103 |
|
|
|
105 |
|
106 |
Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
|
107 |
|
108 |
+
- **Hardware Type:** 8 A100 GPUs with 40GB
|
109 |
+
- **Hours used:** 72h
|
110 |
+
- **Cloud Provider:** Private Infrastructure
|
111 |
+
- **Compute Region:** US/EU
|
112 |
+
- **Carbon Emitted:** 250W x 72h = 18 kWh x 0.432 kg eq. CO2/kWh = 7.78 kg eq. CO2
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
113 |
|
|
|
|
|
|
|
114 |
|
115 |
+
## Technical Specifications
|
116 |
|
117 |
#### Software
|
118 |
|
119 |
+
Pretrained using https://github.com/lamalab-org/MatText
|
120 |
+
|
121 |
+
## Citation
|
122 |
|
123 |
+
To be published
|
124 |
|
125 |
<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
|
126 |
|
|
|
132 |
|
133 |
[More Information Needed]
|
134 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
135 |
|
136 |
+
## Model Card Authors
|
|
|
|
|
137 |
|
138 |
+
The model was trained by Nawaf Alampara ([n0w0f](https://github.com/n0w0f)), Santiago Miret ([LinkedIn]()), and Kevin Maik Jablonka ([kjappelbaum](https://github.com/kjappelbaum)).
|
139 |
|
140 |
## Model Card Contact
|
141 |
|
142 |
+
[Nawaf](https://github.com/n0w0f),
|
143 |
+
[Kevin](https://github.com/kjappelbaum)
|